Janssens, F., Zhang, L., De Moor, B., & Glänzel, W. (2009). Hybrid clustering for validation and improvement of subject-classification schemes. Information Processing & Management, 45(6), 683-702.

Janssens, F., Zhang, L., De Moor, B., & Glänzel, W. (2009). Hybrid clustering for validation and improvement of subject-classification schemes. Information Processing & Management, 45(6), 683-702.



科學認知映射(cognitive mapping of science)能將科學的結構(the structure of science)加以視覺化，起初應用於資訊服務(information services)、後來發現也可將其應用在科學政策(science policy)與研究評鑑(research evaluation)，現在則有越來越多將其應用在發現新興與正在融合中的領域以及主題劃分(subject delineation)的改善。

科學認知映射主要可分為依據引用資訊、依據文本以及混合上述兩種資訊等三種方法。本研究利用文本與引用資訊混合的方法，將2002-2006年Web of Science資料庫內的期刊進行集群，以集群結果產生的認知映射，檢驗目前的期刊主題分類架構，如果可行的話，也將提出改善方式。

本研究首先評估ESI (Essential Science Indicators) 的22個領域主題分類架構，並且將其視覺化。圖1的左右分別是以交互引用與文本方式測量22個ESI領域的Silhouette值，從圖上發現生物學及生物化學(#2)、臨床醫學(#4)、工程學(#7)、植物及動物科學(#19)以及社會科學(#21)等領域上的期刊並沒有足夠好的一致性(coherent)。



從詞語的TF-IDF可以找出每個領域的描述詞語，而這裡也可發現不少領域的描述語互有重疊，例如工程學(#7)與電腦科學(#5)、化學(#3)與材料科學(#11)、植物及動物科學(#19)與環境/生態學(#8)，以及生物學及生物化學(#2)、分子生物學及遺傳學(#14)與臨床醫學(#4)等等，此外，從描述社會科學(#21)的詞語也可以了解這個領域高度的異質性(heterogeneity)。


圖2則是以Pajek畫出22個ESI領域的結構圖，圖上也可發現生物學及生物化學(#2)與臨床醫學(#4)、化學(#3)與材料科學(#11)、電腦科學(#5)與工程學(#7)、環境/生態學(#8)與植物及動物科學(#19)等領域之間有很強的關連。

然後將約8300種期刊利用餘弦相似法及Wade的凝聚式階層集群演算法(Wade's  agglomerative hierarchical cluster algorithm)進行集群，再比較集群結果與分類架構。值得說明的是文字部分的資訊可以提供集群結果的標示(labelling)，而引用部分則可產生交互引用圖(cross-citation graph)提供視覺化，並且輸入PageRank演算法以決定代表性期刊。

決定集群的數目可以根據集群結果的品質，而集群品質有內在或外在驗證測量(internal or external validation measures)等兩種評估方式。內在驗證只考慮資料與集群的統計特性，例如dendrogram、Silhouette值與模組性(modularity)等；外在驗證需要將集群結果與一個已知的劃分標準進行比較，例如計算兩者間的Jaccard相似性 (Jaccard similarity)。本研究以dendrogram的視覺化方式將期刊首先分為三大群，再分為七群，最後分為22群，三大群約等於自然與應用科學(生物學、農學及環境科學；物理、化學及工程學；數學及電腦科學)、醫學與社會科學以及人文學。以TF-IDF描述語來看，七個群組中有三個屬於自然與應用科學，兩個是生命科學(生物科學及生物醫學與臨床、實驗醫學及神經科學)與兩個是社會科學以及人文學(經濟學、商學及政治學與心理學、社會學及教育學)

圖6是22個群組的結構圖，圖上可看到屬於社會科學以及人文學的群組(#1、#6、#14及#22與#9、#11及#21)、地理學、環境科學、生物學及農學(#2、#15及#19)、物理、化學及工程學(#4、#20及 #5)、數學及電腦科學(#8及#18)、生物科學及生物醫學(#3、#13及#16)與臨床、實驗醫學及神經科學(#7、#10、#12及#17)。



表3比較22個ESI領域與本研究利用引用、文本以及混合等方法產生的22個群組的集群品質，可以發現混合引用及文本資料在各種指標上幾乎都有最好的表現。


圖8是利用Jaccard指標比較集群結果與ESI架構的一致性(concordance)。



最後，本研究並分析期刊轉移(migration)的情形，也就是期刊不屬於原本ESI架構的領域所對應的集群，而被分配到另一個不同集群的現象，好的轉移(Good migration)能使分類的一致性增加，也就是Silhouette值或是模組性增加。本研究希望利用這個現象，從集群與領域的一致性的基礎上提出改善目前期刊主題分類架構的方法。

The main bibliometric techniques are characterised by three major approaches, particularly the analysis of citation links (cross-citations, bibliographic coupling, co-citations), the lexical approach (text mining), and their combination.

A hybrid text/citation-based method is used to cluster journals covered by the Web of Science database in the period 2002–2006. The objective is to use this clustering to validate and, if possible, to improve existing journal-based subject-classification schemes.

In a first step, the 22-field subject-classification scheme of the Essential Science Indicators (ESI) is evaluated and visualised. In a second step, the hybrid clustering method is applied to classify the about 8300 journals meeting the selection criteria concerning continuity, size and impact.

The hybrid method proves superior to its two components when applied separately. The choice of 22 clusters also allows a direct field-to-cluster comparison, and we substantiate that the science areas resulting from cluster analysis form a more coherent structure than the ‘‘intellectual” reference scheme, the ESI subject scheme.

Moreover, the textual component of the hybrid method allows labelling the clusters using cognitive characteristics, while the citation component allows visualising the cross-citation graph and determining representative journals suggested by the PageRank algorithm.

Finally, the analysis of journal ‘migration’ allows the improvement of existing classification schemes on the basis of the concordance between fields and clusters.

The history of cognitive mapping of science is as long as the history of computerised scientometrics itself. While the first visualisations of the structure of science were considered part of information services, i.e., an extension of scientific review literature (Garfield, 1975, 1988), bibliometricians soon recognised the potential value of structural science studies for science policy and research evaluation as well. At present, the identification of emerging and converging fields and the improvement of subject delineation are in the foreground.

The main bibliometric techniques are characterised by three major approaches, particularly the analysis of citation links (cross-citations, bibliographic coupling, co-citations), the lexical approach (text mining), and their combination.

For instance, clustering based on co-citation and bibliographic coupling has to cope with several severe methodological problems. This has been reported, among others by Hicks (1987) in the context of cocitation analysis and by Janssens, Glänzel, and De Moor (2008) with regard to bibliographic coupling. One promising solution is to combine these techniques with other methods such as text mining (e.g., combined co-citation and word analysis: Braam, Moed, & Van Raan, 1991a; combination of coupling and co-word analysis: Small (1998); hybrid coupling-lexical approach: Janssens, Glänzel, & De Moor, 2007; Janssens et al., 2008).

Jarneving (2005) proposed a combination of bibliometric structure–analytical techniques with statistical methods to generate and visualise subject coherent and meaningful clusters. His conclusions drawn from the comparison with ‘intellectual’ classification were rather sceptical.

Despite several limitations, which will be discussed further in the course of the present study, cognitive maps proved useful tools in visualising the structure of science and can be used to adjust existing subject-classification schemes even on the large scale as we will demonstrate in the following.

The main objective of this study is to compare (hybrid) cluster techniques for cognitive mapping with traditional ‘intellectual’ subject-classifications schemes.

In a first study by authors related to the current work, the pilot study of Glenisson, Glänzel, & Persson (2005), further extended and confirmed by Glenisson, Glänzel, Janssens et al. (2005), full-text analysis and traditional bibliometric methods were serially combined to improve the efficiency of the individual methods. It was clear that clusters found through application of text mining provided additional information that could be used to extend and explain structures found by bibliometric methods, and vice versa. However, the integration was still limited to serial combination.

All textual content was indexed with the Jakarta Lucene platform (Hatcher & Gospodnetic, 2004) and encoded in the Vector Space Model using the TF-IDF weighting scheme reviewed by Baeza-Yates & Ribeiro-Neto (1999). Stop words were neglected during indexing and the Porter stemmer was applied to all remaining terms from titles, abstracts, and keyword fields. The resulting term-by-document matrix contained nine and a half million term dimensions (9,473,061), but by ignoring all tokens that occurred in one sole document, only 669,860 term dimensions were retained. Those ignored terms with document frequency equal to one are useless for clustering purposes.

The dimensionality was further reduced from 669,860 term dimensions to 200 factors by Latent Semantic Indexing (LSI) (Berry, Dumais, & O’brien, 1995; Deerwester, Dumais, Furnas, Landauer, & Harshman, 1990), which is based on the Singular Value Decomposition (SVD).

Text-based similarities were calculated as the cosine of the angle between the vector representations of two papers (Salton & Mcgill, 1986).

For simplicity and efficiency, the method used to summarise the subject of a field or cluster is based on selecting the terms with the highest mean TF-IDF weights over all journal papers in the field or cluster, where the IDF factor is calculated on the complete term-by-paper matrix (more than six million papers).

For example, Treeratpituk and Callan (2006) automatically select and assign a few concise labels to hierarchical clusters by combining statistical features from the cluster, parent cluster, and a corpus of general English into a descriptive score.

Geraci, Maggini, Pellegrini, and Sebastiani (2008) label clusters by combining intra-cluster and inter-cluster term extraction based on a variant of the information gain measure, and by looking within the titles of Web pages for the substring that best matches the selected top-scoring words.

The similarities Sij used for clustering were found by calculating the cosine of the angle between the pair of vectors containing all symmetric journal cross-citation values between the two respective journals (i and j) and all other journals (i.e., row or column of the matrix C):

The journal cross-citation graph is also analysed to identify important high-impact journals. We use the PageRank algorithm (Brin & Page, 1998) to determine representative journals in each cluster. Besides, the graph can also be used to evaluate the quality of a clustering outcome.

In order to subdivide the journal set into clusters we used the agglomerative hierarchical cluster algorithm with Ward’s method (Jain & Dubes, 1988).

In general, the number of clusters is determined by comparing the quality of different clustering solutions based on various numbers of clusters. Cluster quality can be assessed by internal or external validation measures. Internal validation solely considers the statistical properties of the data and clusters, whereas external validation compares the clustering result to a known gold standard partition

This compound strategy encompasses observation of a dendrogram, text- and citation-based mean Silhouette curves, and modularity curves. Besides, the Jaccard similarity coefficient is used to compare the obtained results with an intellectual classification scheme.

Up to a multiplicative constant, modularity measures the number of intra-cluster citations minus the expected number in an equivalent network with the same clusters but with citations given at random. Intuitively, in a good clustering there are more citations within (and fewer citations between) clusters than could be expected from random citing.

In Fig. 8, we use the Jaccard index to compare each cluster with every field from the intellectual ESI classification, in order to detect the best-matching fields for each cluster.

Nowadays two ISI systems are widely used, in particular, the ISI Subject Categories, which are available in the JCR and through journal assignment in the Web of Science as well, and the Essential Science Indicators (ESI).

While the first system assigns multiple categories to each journal and is too fine grained (254 categories) for comparison with cluster analysis, the ESI scheme is forming a partition (with practically unique journal assignment) and the 22 fields are large enough. ... This subject-classification scheme is in principle based on unique assignment; only about 0.6% of all journals were assigned to more than one field over a 5-year period.

Fig. 1 presents the evaluation of the 22 ESI fields based on the cross-citation- (left) and text-based (right) Silhouette values (see Section 3.3.3). Since the ESI fields form a partition, this approach allows to evaluate their consistency as if the fields were results of a clustering procedure. Multi-, interand cross-disciplinarity of journals can certainly affect the results.

Several fields seem not to be coherent enough from both perspectives (i.e., the cross-citation and textual approach). Above all, the Silhouette values of field #2 (Biology and Biochemistry), #4 (Clinical Medicine), #7 (Engineering), #19 (Plant and Animal Science) and #21 (Social Sciences) substantiate that at least five of the 22 fields are not sufficiently coherent.

Simultaneously to the above validation, the textual approach also provides the best TF-IDF terms – out of a vocabulary of 669,860 terms – describing the individual fields. These terms are presented in Table 2. Although these terms already provide an acceptable characterisation of the topics covered by the 22 fields, considerable overlaps are apparent between pairs of fields, respectively: Engineering (#7) and Computer Science (#5), Chemistry (#3) and Materials Science (#11), Plant and Animal Science (#19) and Environment/Ecology (#8), as well as Biology and Biochemistry (#2), Molecular Biology and Genetics (#14) and Clinical Medicine (#4). In addition, the terms characterising the social sciences (#21) reflect a pronounced heterogeneity of the field.

The structural map of the 22 ESI fields based on cross-citation links is presented in Fig. 2. For the visualisation we used Pajek (Batagelj & Mrvar, 2003). The network map confirms the strong links we have found based on the best terms between fields #2 and #14, #3 and #11, #5 and #7, and #8 and #19, respectively.

In Table 3 we compare the quality of the partition of 22 ESI fields with the quality of the 22 clusters resulting from citation-based, text-based and hybrid clustering.

The cluster dendrogram shows the structure in a hierarchical order (see Fig. 4). We visually find a first clear cut-off point at three clusters, a second one around seven, and 22 clusters also seemed to be an acceptable/appropriate number.

The number of three clusters results in an almost trivial classification. Intuitively, these three high-level clusters should comprise natural and applied sciences, medical sciences, and social sciences and humanities.

The solution comprising of seven clusters results in a non-trivial classification. The best TF-IDF terms (see Table 5) show that three of these clusters represent the natural/applied sciences, whereas two classes each stand for the life sciences and the social sciences and humanities. This situation is also reflected by the cluster dendrogram in Fig. 4. A closer look at the best TF-IDF terms reveals that the social-sciences cluster (#1 of the 3-cluster solution) is split into the cluster #1 (economics, business and political science) and #6 (psychology, sociology, education), the life-science cluster (#3 in the 3-cluster scheme) is split into clusters #3 (biosciences and biomedical research) and #7 (clinical, experimental medicine and neurosciences) and, finally, the sciences cluster #2 of the 3-cluster scheme is distributed over three clusters in the 7-cluster solution, particularly, the cluster comprising biology, agriculture and environmental sciences (#2), physics, chemistry and engineering (#4) as well as mathematics and computer science (#5).

The social-sciences and humanities clusters form two groups that are each strongly interlinked; one consists of clusters #1, #6, #14 and #22 with focus on humanities, economics, business, political and library science, the other one comprises #9, #11 and #21 with sociology, education and psychology. This is in line with the hierarchical structure shown in Fig. 4. These two groups correspond to the two social-sciences clusters in the 7-cluster solution (cf. Section 4.4).

On the basis of the most important TF-IDF terms (see Table 6) we can assign clusters #2, #15 and #19 to geosciences, environmental science, biology and agriculture, which, in turn, form a larger group corresponding to the first of the three ‘‘megaclusters” in the 7-cluster solution.

These science clusters form two groups, #4, #20 and #5 form one group of chemistry, physics and engineering, while #8 and #18 form the third group comprising mathematics and computer science.

Here we have a biomedical and a clinical group. These two groups are in line with the hierarchical structure of the dendrogram in Fig. 4 but less clearly distinguished in the graphical network presentation (Fig. 6). Nonetheless, the terms provide an excellent description for at least some of the medical clusters: cluster #7 stands for the neuro- and behavioral sciences, #3 for bioscience, #10 for the clinical and social medicine, #13 microbiology and veterinary science, #12 non-internal medicine, #16 hematology and oncology and #17 cardiovascular and respiratory medicine. According to the dendrogram clusters 3, 13, 16 and clusters 7, 10, 12, 17 form one larger cluster each. On the basis of the best terms, we can characterise these groups as the bioscience–biomedical and the clinical and neuroscience group, respectively.

In this subsection we compare the structure resulting from the hybrid clustering with the ESI subject classification. This comparison is based on the centroids of the clusters and fields. The centroid of a cluster or field is defined as the linear combination of all documents in it and is thus a vector in the same vector space. For each cluster and for each field, the centroid was calculated and the MDS of pairwise distances between all centroids is shown in Fig. 7.

In Fig. 8, we use the Jaccard index to determine the concordance between our clustering solution and the ESI Scheme by comparing each cluster with every field, in order to detect the best-matching fields for each cluster. The darker a cell in the matrix, the higher the Jaccard index, and hence the more pronounced the overlap between the corresponding cluster and ESI field.

If clustering algorithms are adjusted or changed, one can observe the following phenomenon. Some units of analysis are leaving clusters they formerly belonged to and end up in different clusters. This phenomenon is called ‘migration’. We can distinguish between ‘good migration’ and ‘bad migration’.

‘Good migration’ is observed if the goodness of the unit’s classi- fication improves, otherwise we speak about ‘bad migration’. We can also apply this notion of migration to the comparison of clustering results with any reference classification. In the following we will use the ESI scheme as reference classification.

Out of 8305 journals under study, there were more than one third, namely, 3204 journals that were not assigned to the cluster which best matches their ESI field. As already mentioned above, we call these journals ‘migrated journals’.

‘Good migrations’ are observed if journals improved their Silhouette values after migration. Based on their titles and scopes (not shown), apparently they should indeed be assigned to the cluster to which they have moved.

Although the Silhouette and modularity values substantiate a more coherent structure of the hybrid clustering as compared with the ESI subject scheme, not all clusters are of high quality. Problems have been found, for instance, in clusters #1 and #12 where interdisciplinarity and strong links with other clusters distort the intra-cluster coherence.

Jeong, Y. K. & Song, M. (2016). Applying content-based similarity measure to author co-citation analysis. In Proceedings of iConference 2016.

Jeong, Y. K. & Song, M. (2016). Applying content-based similarity measure to author co-citation analysis. In Proceedings of iConference 2016.

本研究利用引用文獻出現文句內容的相似性來測量作者的主題相關性(topical relatedness)。傳統的作者共被引分析(Author co-citation Analysis, ACA)做法是利用參考文獻裡被引用作者的共被引頻率(White and Griffith, 1981)，然後利用Pearson相關係數 (Pearson correlation coefficient)或是 Salton提出的餘弦相似性測量作者的相似性，在書目計量學研究裡已經廣泛運用於確認與追蹤學科的知識結構(the intellectual structure of an academic discipline) (He & Hui, 2002)。然而這種做法並未考慮引用的內容，Jeong, Song, & Ding, (2014)與 Zhao & Strotmann (2014)則利用全文裡提到的作者並將有關的內容加入ACA的計算。

本研究認為累積被引作者出現的文句能夠代表作者的研究領域，因此利用JASIST的全文資料，剖析HTML，取出論文的後設資料(題名、作者姓名、出版年、DOI與摘要)、引用資訊(引用文句與參考文獻索引)以及參考資訊(作者姓名、出版年、題名與期刊)。在這個研究裡，共使用2003年1月到2015年6月的1910篇論文，合計77,408筆參考文獻。將引用文句與一般文句分開，連結文句內的參考文獻索引與參考資訊，選取100位最多被引用的作者，進行傳統的ACA以及本研究提出的新方法。本研究的新方法利用Mikolov et al., (2013)提出的Word2Vec 模型 (Word2Vec models)，根據參考文獻出現的引用文句，找出作者間的相似性。Word2Vec 模型以大量的文本為基礎，利用類神經網路方法( neural network approaches)，找出詞語之間的語意關係，將每一個出現於文句的詞語轉換成向量，使得這些向量之間的相似性能夠保持詞語在語意上的關係。本研究將被引用的作者姓名視為是引用文句中出現的詞語，測量作者間在研究主題的相似性與合作關係。

表2是傳統的ACA方法與本研究的方法分別找出的10組最相似的作者，本研究的方法找出10組最相似的作者中有一半是具有合作關係的作者。

另外，將兩種方法產生的作者關係分別繪製成網路圖，節點代表作者，利用PageRank決定的節點大小，節點的遠近由作者間的相似性決定，並且以Blondel, Guillaume, Lambiotte, & Lefebvre (2008)提出的社群偵測(community detection)方法進行分群。圖三與圖四分別是傳統ACA與本研究提出方法的結果。

圖三上可明顯地看到所有的作者分為兩群，依據社群偵測的分群結果，左邊的作者可再分為兩群：最左邊紅色的一群為研究資訊尋求行為(information seeking behavior)的作者，紫色的一群則與資訊檢索(information retrieval)研究有關，右邊綠色的一群則是研究書目計量學(bibliometrics)的作者。介於左右兩大群體的作者分別有兩位：Borgman和Salton。這兩位都是資訊科學領域傳統上會經常引用的作者。



在以Word2Vec方法產生的作者網路上，與資訊檢索有關的作者群組位於左方，包括上方的資訊尋求行為以及下方的文件檢索(document retrieval)兩個群組，書目計量學在圖四上則分為兩個有關的群組，一個主要包含作者分析(author analysis)，另一則是期刊引用分析(journal citation analysis)與評鑑指標(evaluation indicator)。與圖三不同的是，圖四上的群組彼此間都有連結，並且圖形上更具體地呈現次學科(sub-disciplines)以及重要的作者。

Unlike other ACA studies, we used citing sentences to reflect topical relatedness of authors.

In  our  research,  we extended  traditional  approaches by  adopting Word2Vec, one  of  deep learning methods, to measure author similarity.

We also conducted in-depth network analysis of author maps.

The results of Word2Vec-based author map revealed more specific sub-disciplines and the important authors in perspective of topical influence than traditional approach does.

Author co-citation Analysis (ACA), which was introduced by White and Griffith (1981), has been widely used in bibliometrics researches to identify and trace the intellectual structure of an academic discipline (He & Hui, 2002). In ACA, traditional approaches relied on the co-citation frequency of cited authors in the reference section.

Thus, one of the main topics in ACA was methodological discussion of what kind of measure is appropriate and relevant for calculation of author similarities (Leydesdorff, 2005; van Eck & Waltman, 2007). Existing approaches based on co-citation frequencies such as Pearson correlation coefficient and Salton’s cosine similarity, however, do not capture the citation content.

Thus, some recent researches used the full-text to obtain the topical relatedness between the cited authors (Jeong, Song, & Ding, 2014; Zhao & Strotmann, 2014). They analyzed the authors mentioned in the full-text and incorporated contents related with cited authors into ACA.

In that sense, cumulated citing sentences of cited authors are able to well represent the cited researches and cited authors’ research areas. In addition, these citing sentences are particularly useful for summarization of a research document.

Figure 1 shows the overall system flow of our approach.

For content analysis, however, we collected full-text research articles of JASIST in HTML format. Through the HTML parsing process, we extracted the metadata (title, author name, year, DOI and abstract), citation information (citing sentence, and reference id), and reference information (author name, year, title, and journal).

To compare our method to traditional ACA, we computed author-pairs in both approaches. In Word2Vec-based method, the full-text data, first, are splitting into sentences. In second step, matching the citing sentences with reference id in reference section, we separated the citing sentences and other general sentences. Then, citing sentences are preprocessed in the following steps: tokenization, POS tagging, lemmatization of the tokenized sentence, and stop word removal.

From these data, we trained Word2Vec model for calculating author similarity and generated author-author similarity matrix. To compare the previous research, traditional author counting approach, we also construct co-citation matrix based on citation counts. Since we preprocessed full-text including all reference information, these matrices considered all cited authors.

To evaluation, we selected top 100 authors which are highly cited in both methodology, and conduct network analysis through visualizing author maps.

The data was gathered from 1,910 full-text articles in the JASIST digital library over 12 years (from January 2003 to June 2015). The 1,910 collected documents have 77,408 references. We extracted elements from the full-text article: 1) citing sentences from the body of the article, 2) the references information, and 3) all cited authors. Table 1 shows the basic statistics of collected data.

Word2Vec models, one of the neural network approaches, are able to carry semantic meanings and turns text into a numerical form that deep-learning nets can understand (Mikolov et al., 2013). Based on a large amount of plain text, Word2Vec trains relationships between words automatically.

Word2Vec spatially encoded a word meaning and the relationship between words, which was originally applied to word clustering or synonym detection (Wolf et al., 2014). We applied Word2Vec into author similarity measure regarding cited author names as a word in plain text.

Since authors’ oeuvre was represented as the citing sentences in research articles, the Word2Vec-based method could consider various topics of the author.

In the proposed approach, however, the author names are also trained as words in a same citing sentence. Therefore, the similarity between two authors in the Word2Vec-based method reflects both topical relatedness and collaborations.

Table 2 shows top 10 pairs by the traditional ACA method (Pearson correlation based similarity) and the Word2Vec based approach respectively. About the half of pairs resulted from the Word2Vec approach are the co-author relationship.

This results imply that the proposed approach enables to detect wider range of author pairs in perspective of topical relatedness and grasp more diverse research fields of information science.

To examine whether there are structural differences in two measures of author similarity, we constructed two author networks with top 100 authors. For network visualization, we used PageRank (Brin & Page, 1998) to determine the node size and also adopted the modularity algorithm (Blondel, Guillaume, Lambiotte, & Lefebvre, 2008) for the community detection.

Figure 3 illustrates roughly two parts that consist of information retrieval and bibliometrics, two major research areas in JASIST. The author group of information retrieval (purple) along with information seeking behavior (red) is located at the left side, and the author group related with bibliometrics is located at the right side.

There are only two authors located between two groups (Borgman and Salton), who are traditionally cited authors in the information science field. Borgman studied various topics including information retrieval and scholarly communication and wrote the important books that had won the best information science book from ASIST. Salton’s works also received a lot of citations for a long time in the field of information science.

The author group related with information retrieval in the left side of the network is split into information seeking behavior (blue) located in the upper side of the network and document retrieval (yellow) located at the bottom side of the network. The group related to bibliometrics is also separated into two parts: (1) a cluster (green) including author analysis and (2) journal citation analysis and evaluation indicator (red).

Unlike Figure 3, the communities in the network are connected to each other. Brin is connected with both document retrieval and citation analysis communities. This may be attributed to the fact that the PageRank, developed by Brin and Page (1998), is used in information retrieval and also studied in network analysis to compute node centrality.

In bibliometrics, PageRank is adopted as one of the centralities in citation networks (Ding, Yan, Frazho, & Caverlee, 2009). Ingwesen, who is located between information retrieval and bibliometrics, studied information retrieval in earlier works, he extended the research area to network analysis such as webometrics.

It implies that the authors linked by citation are topically grouped in the Word2Vec-based author network.

Åström, F. (2007). Changes in the LIS research front: Time‐sliced cocitation analyses of LIS journal articles, 1990–2004. Journal of the American Society for Information Science and Technology, 58(7), 947-957.

Åström, F. (2007). Changes in the LIS research front: Time‐sliced cocitation analyses of LIS journal articles, 1990–2004. Journal of the American Society for Information Science and Technology, 58(7), 947-957.

scientometrics

本研究利用論文間的共被引分析探討1990到2004年間圖書資訊學(LIS)的研究前沿(research front)的改變，了解這個學科目前的處境與發展趨勢。分析資料為21種LIS期刊。將同時間內具有影響力的共被引文章定義為研究前沿(research fronts)，並且分為三個5年期間，分析領域的改變。研究結果發現LIS由兩個不同研究領域構成的穩定結構：資訊計量學(informetrics)和資訊搜尋與檢索(information seeking and retrieval)，由於分享研究興趣與方法，資訊檢索與資訊計量學有靠近的傾向。而網路為主的研究成為資訊計量學和資訊搜尋與檢索的主要研究則是這個領域的主要變化。

本研究採用的期刊來源為JCR (2003) 的Information Science & Library Science分類下的55種期刊。去除主要是被非LIS期刊引用的期刊以及評論性或商業性期刊後，選擇1990到2004年間有出版的期刊，如下表共21種。

論文的總數為13605筆，從中選取最高被引用的論文，建立共被引次數矩陣。以多維尺度演算法(multidimensional scaling algorithm, MDS)進行處理。
首先是研究基礎(research base)部分，從13605筆論文資料的221586次引用(150145篇參考文獻)中，選取被引用超過50次的文獻，共66筆進行分析。其共被引映射圖如FIG 1.:

與先前研究一致，圖書資訊學在圖形上分為兩個區域，圖形上半部為資訊搜尋與檢索相關論文，下半部則為資訊計量學文獻，此一結果和 Persson (1994) 與 White & McCain (1998)等研究相符合。另外在資訊計量學文獻右邊，還有一群文獻形成網路計量學(webometrics)叢集。網路計量學是利用連結、引用與叢集等資訊計量學方法進行網路本質與特性的分析。

資訊搜尋與檢索從早期的系統導向資訊檢索(systems-oriented information retrieval)發展到使用者-系統互動研究(user-system interaction studies)和資訊行為(information behavior)。

除了網路計量學以外，資訊計量學以書目計量映射(bibliometric mapping)為中心，周圍的部分是書目計量分布(bibliometric distributions)。

為了進一步了解與核對共被引分析的結果，將共被引資料輸入叢集分析。叢集分析所產生的8個叢集符合映射圖的結構，各叢集如TABLE 2。

從TABLE 2各叢集出版年度的中位數，可以將八個叢集分為四個時期：第一個時期圖書資訊學的研究包括實驗性資訊檢索(experimental information retrieval)、書目計量映射以及書目計量分布；第二個時期開始對於資訊檢索的使用者端產生興趣，增加了搜尋過程與認知面向的資訊檢索研究；隨後是在1990年代早期進行的相關性(relevance)研究，同時也傾向於一般的資訊行為；1990年代末期則受到網路科技的影響，開始進行網路以及網路計量學的研究。

接下來的共被引分析，被引用的參考文獻僅限於也在13605篇論文裡的論文，來了解具有影響力的論文，做為研究脈絡(research context)。選取被引用次數超過25次的論文，共65篇。呈現的圖形大致上仍然可明顯的看出分為上半區域的資訊搜尋與檢索和下半區域的資訊計量學。但資訊檢索的研究以認知性資訊搜尋與檢索、相關性和資訊行為為主要，實驗性資訊檢索研究則成為邊緣。

相較於研究基礎，在研究脈絡上可以發現資訊計量學的結果較為分散，包含三個部分：研究合作(research collaboration)、書目計量映射與網路計量學，並且以網路計量學最為主要。

以TABLE 3的叢集結果來看，在研究脈絡中雖然實驗性資訊檢索與書目計量分布消失了，但增加了兒童的資訊行為研究和對於研究合作的資訊計量學分析。雖然這些研究依然存在，但本身並沒有形成叢集，而是歸入其他的叢集中，如IR/Search。

對三個5年的時期進行研究前沿分析，第一個時期1990-1994年，共有3401篇論文，彼此間有1581次引用，39篇論文獲得5次以上的引用。這個時期以ISR為主，特別是使用者觀點的ISR研究；資訊計量學由兩個小叢集組成：一為研究合作，另一聚焦於映射。

第二個時期1995-1999年，包含3318篇論文，彼此間的引用共有2117次，獲得5次以上引用的論文共有52篇。這時期ISR的聚集相當明顯，除了聚焦在資訊科技(information technology)和實驗性資訊檢索(experimental IR)的兩個叢聚外。在一般的資訊計量學之外，另外還有研究成效(research performance)的叢集。

第三個時期2000-2004年，有4147筆論文，彼此間有2926次的引用，62篇論文的引用次數超過7次。在這個時期，可以看出資訊計量學較前面兩個時期緊密連接，主要聚焦在網路計量學，而ISR則較前兩個時期變得較為分散，可分為三個叢集：ISR、兒童的資訊行為(children's information behaviors)以及健康資訊學(health informatics)。

本研究發現LIS有相當穩定的結構，主要為ISR及資訊計量學所構成。另外，從研究基礎上發現，大多為理論或方法學的文獻，但研究脈絡與前沿上的文獻卻以實務性的論文為主。就三個時期的研究來看，1900-1994年以圖書館與資訊服務(library and information service)為主，第二個時期則是線上資料庫與資訊尋求；第三個時期受到WWW影響，主要的研究從群體利用WWW搜尋資訊的方法到發展分析網站影響因素的方法。最後，本研究發現ISR與資訊計量學有愈來愈接近的趨勢，其原因是因為兩者都需要測量文件(或搜尋問題)之間的關係強度，並且也都對將資訊視覺化有興趣，因此彼此引用整合的機會增加。

Based on articles published in 1990–2004 in 21 library and information science (LIS) journals, a set of cocitation analyses was performed to study changes in research fronts over the last 15 years, where LIS is at now, and to discuss where it is heading.

The results show a stable structure of two distinct research fields: informetrics and information seeking and retrieval (ISR). However, experimental retrieval research and user oriented research have merged into one ISR field; and IR and informetrics also show signs of coming closer together, sharing research interests and methodologies, making informetrics research more visible in mainstream LIS research. Furthermore, the focus on the Internet, both in ISR research and in informetrics—where webometrics quickly has become a dominating research area—is an important change.

The nature and intellectual organization of LIS has been thoroughly investigated in analyses describing the general traits of LIS research, as well as mapping how LIS has been organized in different research themes (Persson, 1994; White & Griffith, 1981; White & McCain, 1998).

My approach centers on the following questions. What research topics have dominated LIS during the period 1990–2004? What changes can be observed in the topics addressed over the last 15 years? Can these changes can be used to tell us something about where LIS is heading?

Most definitions of “research fronts” explain them as groups of citing articles being clustered through bibliographic coupling (e.g., Persson, 1994), and their relations to the cited documents clustered by cocitation analysis (Garfield, 1994; Morris et al., 2003; Price, 1965). Although Persson sees the current (citing) articles as the research front and the cited documents as the research base, Garfield, for example, also includes the clusters of cocited core articles into the research front.

In addition, by analyzing the co-occurrence of highly cited documents, we also get an indication on the impact of the articles, thus expanding the definition of research fronts as including influential, as well as current research.

To identify LIS research, and to select journals for the analyses, the Journal Citation Reports: JCR Social Sciences (Thomson ISI, 2003) was used. To defining LIS research, JCR’s Information Science & Library Science classification, covering 55 journals, was used.

To limit the definition, all general LIS journals were identified and the specialized ones were excluded. This was done using the “Citing Journal” field in JCR: If the journal primarily was cited by non-LIS publications, it was excluded from the study.

The analyses were done on a document level, as opposed to an analysis on the author level. Although an author analysis provides more of an overview, the document analysis is more detailed, e.g., by not grouping documents on different topics by the same author.

The result reflects contemporary and influential research within a specific field of research, i.e., the research front.

The research base was based on the 13,605 journal articles published from 1990–2004 and their 221,586 references to 150,145 unique documents. The 66 most-cited documents that received 50 citations or more were selected for further analysis (Figure 1).

The map shows two main areas consistent with the structures found in earlier analyses on LIS (e.g., Persson, 1994; White & McCain, 1998). On the top half of the map, a group of information-seeking and retrieval (ISR) related literature is featured and on the bottom half, a group of informetrics literature. However, on the right side of the informetrics field, a group of webometric studies has formed a cluster. Webometrics is the study of the nature and properties of the World Wide Web, using informetric methodologies such as link, citation, and cluster analyses (Björneborn & Ingwersen, 2001).

In the ISR section of the map, there is a thematic shift from right to left. Systems-oriented information retrieval (IR) literature is on the far right, followed towards the left by user-system interaction studies and information behavior. In comparison to Persson (1994), the “soft” part of the IR-field has increased its impact compared to the “hard” systems-oriented IR research.

Apart from the webometric group on the far right, the informetrics field is centered on bibliometric mapping, surrounded by documents concerning bibliometric distributions.

To enhance the results of the cocitation analysis, a cluster analysis (Persson, 1994) was performed, resulting in eight clusters (Table 2). The clusters support the structures identified in the map, and reveal a division of the soft IR-research: from search- and relevance-focused documents, over cognitive IR and information seeking, to information behavior.

The publication years of the clustered documents shows four generations of research orientations, a trait also visible in the IR part of the map. The first generation of LIS research includes experimental IR, bibliometric distributions, and bibliometric mapping. The second generation of research, with references published from the early 1980s marks the increasing interest in the user side of IR, incorporating the search process and the cognitive perspective into IR and LIS research. This is followed by the relevance studies in the early 1990s; and a contemporary trend to focus on general information behavior. The most recent trend in the LIS research base is studies on World Wide Web and webometrics, dating back to the late 1990s.

The results of the second analysis show influential research areas during the period 1990–2004. It is still the same 13,605 articles providing the material, but only the 18,615 citations to articles present in the set of citing documents are analyzed. Here, as well as in the following time-sliced analysis, the self-citations were removed. Out of the 5024 unique-cited documents, the 65 articles being cited 25 times or more were selected and analyzed (Figure 2).

The general structure of the map is the same: with informetrics on the lower half and ISR on the top half. There are some differences, however. In the top half, a center has developed around “Kuhlthau, 1991” and “Ingwersen, 1996,” focusing on cognitive ISR, relevance, and information behavior, while experimental IR research has become peripheral. Different perspectives on the user-oriented research has dominated the information-seeking and retrieval field; and has together with the wider information behavior field formed a strong research area of different variations on information-seeking research.

At the same time, the informetrics field has become more dispersed, with three clearly defined subfields: research collaboration to the left, bibliometric mapping in the middle, and webometrics on the right side. In comparison with the research base, webometrics has become the dominating research area within the informetrics field.

Wouters, P., & Leydesdorff, L. (1994). Has Price's dream come true: Is scientometrics a hard science?. Scientometrics, 31(2), 193-222.

Wouters, P., & Leydesdorff, L. (1994). Has Price's dream come true: Is scientometrics a hard science?. Scientometrics, 31(2), 193-222.

本研究利用Scientometrics期刊論文以及其參考文獻為研究資料，根據多種資訊判斷科學計量學領域是否已經是硬科學，並分析這個領域的其他特性。本研究所使用的科學計量學資訊有引用文獻的相對年齡(relative age of the cited literature)、論文作者間的關係、論文題名的詞語模式(patterns of words  in the titles of these articles)等。

根據Price的知識增長理論(theory of knowledge growth)，科學家會引用本身領域的文獻，因此，如果有研究前沿(research fronts)存在於這個領域，便會產生立即效應。Price指標(Price index)可以測量立即效應(immediacy effect)，Price指標較大表示引用文獻的相對年齡較低，例如Price(1970)測得生物化學和物理的Price指標值約在60%到70%，社會科學大約在42%附近。Crane(1972)則認為科學會形成作者間彼此緊密相連的社群，因此本研究分析作者間的合著關係和引用的關係，並且利用網絡分析技術探討科學社群的凝聚程度，並且測量作者在網絡的位置連結性以及結構的相似性，根據這些資訊進行叢集，集結彼此間連結性強的作者形成一個叢集，或是形成位置相似的作者叢集。另外，Rip and Courtial (1984)和Leydesdorff (1989a)
指出題名上的詞語可視為是出版品的認知訊息(cognitive message)的指標，詞語在題名上的共現可視為是詞語間關係存在的紀錄，因此本研究也利用網絡分析技術探討詞語的共現網絡。

研究結果發現分析的779筆Scientometrics期刊論文資料，除了前三年快速的增加外，平均每年增加3.5筆，並且這些論文資料共包含12341筆參考文獻，平均每篇論文有15.8筆參考文獻。各項指標都相當穩定。Price指標的平均值為43.0%，若以每年的Price指標的平均值在34.0%到51.4%之間。

以作者資料來看， 779筆論文資料共計由669位不同的作者完成，有接近3/4的作者(488位)僅出現在一筆論文資料上，每位作者平均出現在1.8筆論文資料上，其作者生產力符合Lotka分布，並且大部分(61%)的論文是單一作者，平均每篇論文有1.6位作者。合著作者的論文資料中，大多數的作者都僅和一到兩位同事合作，合著網絡相當破散，但幾個較大的網絡與作者在同一機構任職、參與同一研究計畫或者具有共同的研究興趣有關。

Scientometrics期刊論文的作者引用網絡則呈現高度凝聚的狀態。779筆論文資料中有441筆被其他Scientometrics期刊論文引用，每筆Scientometrics期刊上的論文引用的論文平均有19.4%同樣是Scientometrics期刊的論文。發表超過1篇以上論文的作者共有181位，其中的130位作者有引用其他129位作者的資料，利用作者之間的彼此互相引用關係，發現形成的集團(clique)大多與作者任職機構有關，也有一個集團是成員間曾彼此辯論(debate)而產生。最後，題名上的詞語共現網絡也同樣有高度凝聚的情形。從上面的資訊可以判斷科學計量學領域已經由多種的學科背景在認知與社會性上整合而成，但並沒有發現研究前沿的現象。

In more than one respect, Scientometrics displays the characteristics of a social science journal. Its Price Index amounts to 43.0 percent, and is remarkably stable over time.

The majority of the published items in Scientometrics has been written by a single author. Moreover, the network of co-authorships is highly fragmented: most authors cooperate with no more than one or two colleagues.

Both the citation networks of the authors and the network of title words indicate that the field is nonetheless highly cohesive.

The characteristics of the publications in this journal, and the patterns of the bibliometric relations among them, may therefore indicate the type and extent of the cognitive and social integration of the various disciplinary backgrounds into scientometrics as a field.

The question, in other words, is how "hard" scientometrics is, and how strongly its knowledge is codified. These properties can be measured in terms of:
a) the relative age of the cited literature, the so-called "Price Index",
b) the relations among the authors of articles published in Scientometrics; and
c) the pattern of words in the titles of these articles.

According to Price's theory of knowledge growth (Price 1965), science distinguishes itself from other fields of study by the way scientists refer to their literature (Price 1970). The existence of "research fronts" in science supposedly leads to an "immediacy effect", which can be measured in terms of the so-called "Price Index".

The Price Index is defined as "the proportion of the references that are to the last five years of literature" (Price 1970). Price estimated that this index would vary between 22 and 39 percent if no immediacy effect were present. [1] A field that was all research front and with no general archive might have a Price Index of 75 to 80 percent.

From his analysis of 162 journals, Price (1970) concluded: "Perhaps the most important finding I have to offer is that the hierarchy of Price's Index seems to correspond very well with what we intuit as hard science, soft science, and nonscience as we descend the scale." Biochemistry and physics are at the top, with indexes of 60 to 70 percent, the social sciences cluster around 42 percent, and the humanities fall in the range of 10 to 30 percent.

Science is, on the whole, practised in tightly knit communities in which the authors address one another (Crane 1972).

Co-authorship relations can be considered as indicators of co-operation. [3]

The meaning of citation relations is less clear, given the ongoing citation debate (MacRoberts and MacRoberts 1989; Cozzens 1989; Luukkonen 1990; Leydesdorff and Amsterdamska 1990; Woolgar 1991). But whatever the precise meanings of citations may be, citations can be considered as sociometric data, and the resulting network can accordingly be analyzed (cf. Shrum and Mullins 1988).

We analyzed the extent to which the authors are connected to one another, i.e. the cohesiveness of the network, as well as the pattern displayed by each author in relation to all other authors, i.e. the position of authors in the network.

We also analyzed the similarities among authors in both these dimensions of the matrices, i.e. we clustered strongly connected authors as well as authors in similar positions. Direct as well as indirect linkages between the authors are involved in this analysis.

Strong cliques are sets of authors connected by relations in such a way that all members of the clique are connected to one another, and anyone for whom this holds is included in the clique. The inclusion criterion is less strong for weak cliques, in which all pairs within the clique must have relationships with all other pairs, and anyone with a relation to or from a member of the clique is included.

Strong structural equivalence clusters are sets of authors with completely identical positions in the network (the distance dP between them is zero). Weak structural clusters are sets of authors with a significant similarity in their patterns of relations (the distance dP is small).

As noted, we wished to know whether a structurally codified semantics of scientometrics exists or whether, on the contrary, the articles in Scientometrics use the different terminologies of the various disciplines surrounding scientometrics.

Given the functions of titles of articles, the words in these titles can be considered as indicators of the cognitive message of the publication (Rip and Courtial 1984; Leydesdorff 1989a). The co-occurrence of words in titles can be considered as an indication of the existence or non-existence of relations between these words (Callon et al. 1983).

Since 1978, 779 items have been published in Scientometrics. They contain 12,341 references to the scientific literature. [10]

The number of publications per year in the journal increases in a linear way (Fig. 1). [11] After a steep growth during the first three years, the number increases by 3.5 publications per year.

The number of references per year shows a comparable pattern, although somewhat more irregular. Every publication contains on average 15.8 references (cf. Yitzhala, 1991). Since 1986, this number has become stable at an average of 15 references per publication (Fig. 2).

The publications in Scientometrics were written by 669 different authors. On average, every author published 1.8 times and every paper was written by 1.6 authors. Nearly three-fourth of the authors (488 or 73 percent) published only once in Scientometrics. The distribution of productivity among the authors is a Lotka distribution (Fig. 4).

The average Price Index of Scientometrics is 43.0 percent. ... The Price Index varies between 34.0 and 51.4 percent (Fig. 5). The regression line is not significant. [12] Apparently, the index displays neither rise nor fall since 1978.

Recently, Schubert and Maczelka (1993) concluded from an analysis of Scientomettics in 1980-81 and 1990-91 that the journal has moved slightly from the "soft" (social) towards the "harder" (natural) sciences. They drew this conclusion from the rise of the Price Index from 35 percent to 42 percent between these measurement points. This observation is, however, based on only two measurements. Because of the statistical fluctuations in the value of the Price Index over time, any conclusion can be drawn regarding the development of the Price Index if one restricts oneself to only two measurement points.

In accordance with Price's theory, the number of references to literature of a specific age rises until the cited literature is two years older than the citing literature, and then falls off (Fig. 7). Note that this decline is gradual. Apparently, only a small "immediacy effect" is visible in scientometrics.

A general phenomenon in science is the growth of the number of co-authored scientific articles, relative to the total scientific production (Luukkonen et al. 1992; Abt 1992).

In Scientometrics, however, 61 percent of the articles have been written by a single author. This share is stable over time.

The network of co-authorships is highly fragmented. ... With the exception of three subgroups, most co-authors cooperate with no more than one or two colleagues.

Comparison of the composition of the weak structural equivalence clusters with the relational cliques reveals that two clusters are identical: a group of authors from Leiden (Van Raan et al.) and a group of authors with various institutional affiliations, probably best characterized as the "co-word analysis group". So, these two groups have distinct identities, with respect both to their relations and to their positions in the network.

Some clusters seem constituted by the institutional affiliations of the authors. This holds for the Leiden group and for the authors around ISI (cluster 3). In other cases, nationality appears to be the binding force. This holds for the group in Hungary (cluster 5), the Belgian informetricians (cluster 10) and the Spanish scientometricians (cluster 6). However, cluster 1 can best be characterized by its research program (co-word analysis). Cluster 2 seems to consist of authors from Sussex together with CHI Research Inc. Thus, co-author relations are not only institutionally defined; shared interests and common intellectual goals play a role as well.

To sum up, scientometrics is a fragmentary field of co-authorships. The authors are highly selective in their co-authorship relations with one another. Co-authorships are defined neither exclusively by social nor only by intellectual factors. Both dimensions shape the pattern of co-authorships.

With respect to the number of solitary authors and the large number of isolated small clusters, scientometrics exhibits the pattern of a social science.

Of the 779 articles published in Scientometrics, 411 were subsequently cited one or more times in Scientometrics. The share of references to Scientometrics (as a percentage of all references) has stabilized around an average of 19.4 percent since 1987.

Of the 181 authors in the core set, 130 authors cite one another. So, 51 (or 28.2 percent) of the authors publishing more than one article in Scientometrics from 1978 till 1993 are neither citing nor cited within this group of authors.

The core set of authors in Scientometrics is found to be highly cohesive in terms of their mutual citation relations. All these authors are members of one single weak clique. Moreover, a majority of these authors (88) also belongs to one strong clique (Table 5).

The picture is different if we exclude all indirect relations from the analysis. This "fine structure" of the citation matrix is shown in Table 6, where 13 strong cliques and 6 weak cliques are revealed. Most strong cliques seem to coincide with shared institutional affiliations. The exception is clique 9, which indicates the existence of a debate among the members of this clique.

The most striking feature of the network of title words of articles published in Scientometrics is its cohesiveness. All words cluster together in a single strong component clique (Table 8). If only direct relations are included, all words cluster together in a single weak component clique. This means that all words are either used together in a title or share a common co-word.

Thus, the language of scientometrics is both strongly unified and weakly codified. This strong cohesiveness is a stable characteristic of the titles in Scientornetrics, from the very start of the journal. Perhaps a distinct discourse already existed before the journal was founded. In any case, it constitutes a textual identity of scientometrics as a field, one probably different from the various mother disciplines. Thus a process of de-differentiation seems to have occurred not only in the patterns of citing (and being cited) but also at the cognitive level.

The interpretation of the Price Index is complicated because of these variations within disciplines. If we, nevertheless, take the Price Index preliminary as an indicator of "hardness", scientometrics belongs to the group of relatively hard social sciences. At the same time, it stays unequivocally within the social science range. Taken literally, Price's dream has therefore not come true, since he postulated the emergence of a completely new type of social science with a natural science character. But if we reformulate his goal a posteriori in a more modest way, as the building of a relatively hard social science, it did come true.

The value of the Price Index appears stable over the years. Since a number of other indicators also exhibit stability, this seems to suggest the existence of some scientometric identity. For example, the journal expands at a regular rate, while the percentage of co-authored papers increases only very slowly. The origin of this stability can best be explained by the finding that the community of researchers who have published more than once in Scientometrics acts as a tightly knit network.

In addition to the co-authorships within various institutes, and partly overlapping with this structures, there are national co-authorship relations, like those among the Belgian informetricians, and programmatic co-authorship relations, like those among the users of the French co-word instrument. In general, co-authorship relations are firmly embedded in existing social structures, both at the national and at the community level.

These various strong graphs of co-authors, however, are structurally embedded in the communication structure as indicated by textual indicators. Both in terms of citation relations and in terms of title-words the network is very cohesive, while the structural dimensions of codification are less clear.

In summary, the community of authors publishing in Scientometrics is well integrated, while there are no indications of an exclusive paradigm or a research front.

Glenisson, P., Glänzel, W., Janssens, F., & De Moor, B. (2005). Combining full text and bibliometric information in mapping scientific disciplines. Information Processing & Management, 41(6), 1548-1572.

Glenisson, P., Glänzel, W., Janssens, F., & De Moor, B. (2005). Combining full text and bibliometric information in mapping scientific disciplines. Information Processing & Management, 41(6), 1548-1572.

本研究以詞語共現分析(co-word analysis)，將Scientometrics期刊2003年發表的論文，歸類為六個叢集。為了瞭解叢集結果的有效性，將這個結果與專家歸類的結果進行比較，同時也利用書目計量指標分析各個叢集。

Braam, Moed, and Van Raan (1991)建議利用詞語分析(word analysis)評估共被引叢集分析的結果，這些詞語利用書目紀錄裡的索引詞(indexing terms)和分類碼(classification codes)作為基礎。

在專家歸類方面，本研究援引Schoepflin and Glänzel(2001)研究的六個類別：數學模型與資訊計量學法則(Mathematical models/informetric laws)、個案研究(Case studies)、科學計量學的進展(Advances in Scientometrics)、指標工程(Indicator engineering)、社會學方法(Sociological approaches)與政策相關議題(Policy relevant issues)。加上近年興起的網路計量學(Webometrics)後，本研究用來進行專家的歸類的類別為：科學計量學的進展(Advances in Scientometrics)、實務論文與個案研究(Empirical papers/case studies)、數學模型(Mathematical models)、政策議題(Political issues)、社會學方法(Sociological approaches)以及資訊計量學與網路計量學(Informetrics/Webometrics)。下表是共詞分析與專家歸類的比較結果：

除了較大的A與E類別分布在多個叢集外，較小的類別大多集中內一個或兩個叢集上。

從六個叢集上的論文在專家以及它們的詞語網絡，可以將這些叢集分別：叢集1是書目計量學指標的方法學研究(methodological indicator research)，這些指標用來測量發表活動(publication activity)以及引用影響(citation impact)的研究；叢集2大多為有關於國家和機構方面或科學領域的個案研究(case studies)與實務性的論文(empirical papers)；叢集3和叢集1同樣是理論與方法學問題相關的論文，但更著重在資訊計量學法則(informetric laws)、頻率分布(frequency distributions)與多變化統計(multivariate stattistics)等先進方法學技術。叢集4是網路計量學和其他網路相關議題；叢集5是論文數較少的叢集，總共僅包括3篇論文，這些論文與共被引分析以及其他引用統計的分析有關；叢集6則是最大的叢集，包含的面向相當廣泛，從社會學、政策到科技等許多相關主題。從上述的分析，可以了解科學計量學目前主要的兩個面向是基於科學計量學標準技術的方法學研究和擴展傳統書目計量學範圍的實務研究。

接著利用平均參考文獻年齡(mean reference age)和連續出版品所占部分(share of serials)等書目計量學特徵分析上述的叢集結果。如下圖所示

在各個叢集裡，網路計量學具有低參考文獻年齡的特徵，並且連續出版品所占部分為中到高。政策議題相關的論文大部分具有相對低的連續出版品所占部分，但另有一群論文的連續出版品所占部分則明顯地高，因此相關的論文在圖形上分成兩個子叢集。至於科學計量學的先進方法與技術，除了少數例外，大部分的論文的平均參考文獻年齡在5到15年間，連續出版品所佔的部分則是在50%到90%間。實務性研究的論文在連續出版品所占部分的特徵分為兩群，一群的連續出版品所佔部分較低(<=55%)，另一群則較高(>=67%)，較低的一群與政策相關研究具有類似的特徵。

The question how bibliometric measures can, in turn, be assumed to reflect formal characteristics of documented scientific communication that might supplement results obtained from content-based analyses could also be answered in a positive way. Reference-based citation measures can help to fine-structure clusters determined on basis of co-word analysis.

Braam, Moed, and Van Raan (1991) suggested combining co-citation with word analysis in the context of evaluative bibliometrics to improve efficiency of co-citation clustering. The word analysis by Braam et al. used publication ‘‘word-profiles’’ that were based on indexing terms and classification codes.

Not much later, Noyons and Van Raan (1994) and Zitt and Bassecoulard (1994) demonstrated the appeal of plunging into contents by using keywords from both patent—and scientific literature to characterise the science-technology linkage.

The study by Schoepflin and Glänzel aimed at monitoring and characterising structural changes in the research profile in bibliometrics in the period 1980–1997. The authors created five categories, Mathematical models/informetric laws, Case studies, Advances in Scientometrics, Indicator engineering, Sociological approaches and Policy relevant issues. The term Webometrics did not yet appear in this scheme since at that time it was not yet established as a sub-discipline of scientometrics/informetrics.





We see classes S, M, I and P, admittedly all of smaller size, moderately to well conserved in the text-based cluster structure. Conversely, papers assigned to the larger classes A and E are heavily shifted around the text clusters.

The map in Fig. 6 represents the content structure of cluster 1 with altogether 9 papers. This cluster represents publications that are concerned with methodological questions related to bibliometric indicators. Indicator-related terms such as indicator names and terms relevant in the context of measuring publication activity and citation impact are close to the centre, and strongly interlinked. ... One could consider this cluster representing methodological indicator research.



Cluster 2 is dominated by empirical papers and case studies (cf. Table 3). ... The terms in this map are presented in Fig. 7 and relate above all to national and institutional aspects as well as to science fields. This is the cluster of case studies and traditional bibliometric applications.



Cluster 3 is a second theoretical/methodological cluster. Unlike the first one, this cluster relates to more advanced methodological techniques, such as informetric laws, frequency distributions and multivariate statistics. This cluster could be characterised as theoretical and mathematical issues in bibliometrics. The term structure is presented in Fig. 8.




Cluster 4 presented in Fig. 9 clearly represents webometrics and network-related issues. All terms are strongly interlinked. This cluster corresponds by and large to the category of Webometrics/Informetrics.



Cluster 5 with 3 papers is the smallest one. Co-citation analysis and the analysis of other citation statistics are the topic of these papers. The term structure (cf. Fig. 10) reflects the statistical vocabulary used in these studies. This cluster covers specific applications of statistical methods.



The last cluster with 30 papers (see Fig. 11) is by far the largest one. It comprises technology and innovation related studies, the science-technology interface and almost the complete Triple Helix issue can be found here (cf. Table 3). Also the sociological approaches are covered by this cluster. This cluster can be considered a borderland of classical scientometrics, namely the interdisciplinary approaches such as sociological, policy relevant and technology related issues.



The two large categories A and E covering 65% of all papers proved heterogeneous. Category A has (jointly with category M) three sub-clusters, namely, Cluster 1, 3 and 6, whereas Category E falls apart into three other sub-clusters: Cluster 2, 5 and 6. Policy relevant issues are also covered by clusters 2 and 6. Only Category I is represented by a corresponding co-word cluster, namely cluster 4.

The full text analysis substantiates that both methodological and empirical research have nowadays at least two different main focuses each, one is based on scientometric standard techniques such as classical indicators, the other ones are clearly broadening the scope of traditional bibliometrics.



As already seen in the pilot study, Webometrics is characterised by low reference age and medium–high share of serials (cf. Glenisson et al., 2005).

Most of the policy related issues are characterised by relatively low share of serials. Nevertheless, there is a group of papers with clearly higher share, too. This confirms the results of the full text analysis, namely that this category practically forms two sub-clusters.

The category Advances in Scientometrics proves strikingly homogeneous with several outliers only. Most of the A-class papers have, however, a mean reference age ranging between 5 and 15 years, with medium–high share of serials ranging between 50% and 90%.

The empirical groups proved heterogeneous, indeed. Regarding the share of serials this class forms two distinct sub-classes, particularly, one with low share (<=55%) and one with relatively high share (>=67%). The class with lower share has similar characteristics as the policy relevant class.

The question how bibliometric measures can, in turn, be assumed to reflect formal characteristics of documented scientific communication that might supplement results obtained from content-based analyses could also be answered in a positive way. Reference-based citation measures can help to fine-structure clusters determined on basis of co-word analysis.

Chen, Y., Börner, K., & Fang, S. (2013). Evolving collaboration networks in Scientometrics in 1978–2010: a micro–macro analysis. Scientometrics, 1-20.

Chen, Y., Börner, K., & Fang, S. (2013). Evolving collaboration networks in Scientometrics in 1978–2010: a micro–macro analysis. Scientometrics, 1-20.

科學計量學(Scientometrics)利用數學、統計與資料分析方法與技術，蒐集、處理、解釋與預測學術傳播、成效、發展與動態等科技的特徵，對科技進行量化研究。就實務的技術而言，科學計量學利用書目計量的概念測量文本與資訊，並且利用科學地圖(science map)展現結果 (Börner 2010; Börner et al. 2003)。本研究利用網絡分析技術，從巨觀(國家)、中觀(機構)與微觀(作者)三種層次，探討Scientometrics期刊1978-2010年發表的2541筆論文上的合作情形。

過去對於Scientometrics有以下的相關研究，Schoepflin and Glänzel (2001) 利用1980、1989和1997三年出版的Scientometrics論文，發現科學政策(science policy)與科學社會學(the sociology of science)等主題相關的論文比率減少。Peritz and Bar-Ilan (2002) 以1990和2000年的Scientometrics論文，確認Research Policy和Social Studies of Science分別是第三和第四最常引用的期刊。Hou, Kretschmer, and Liu (2008)從Scientometrics2002到2004年論文上的作者合作網絡上發現一半以上的作者有合作的經驗，但是網絡的連結並不強而且疏鬆。Dutt, Garg, and Bali (2003)使用大量論文資料進行，研究資料期間為1978到2001，發現機構的平均論文數偏低，顯示研究的產出相當分散，而且以單一作者的論文為主，雖然多位作者的論文正蓄勢待發。

研究結果發現：
(1) 論文生產力較大的國家有美國、比利時、英國、荷蘭和西班牙。
(2) 機構與作者數隨時間增加，但是機構的平均論文數成長緩慢，近年的作者平均論文數則減少。
(3) 具有高中心性及中介性的一些機構可視為是合作網絡上的守門人(gatekeepers)。
(4) 近期較高生產力的作者取代了早期的重要作者。

Scientometrics的論文、作者和機構平均每年增加率為20%，顯示這個領域吸引愈來愈多的研究人員和機構加入。另外，Bettencourt et al. (2009) 以下面的公式指出當領域成長時，它的合作網路將會變得更為稠密。

而本研究三種層次的合作網絡，國家合作網絡的α值為 2.9533，機構與作者則分別為1.5222和1.2353。明顯的可以看出國家合作網絡相當快速地變得稠密，但由於單一作者論文的增加以及許多合作仍然是同一國家內的機構或同一機構內的作者彼此間的合作，使得機構合作網絡和作者合作網絡的α值較小。
在網絡的直徑(diameter)，也就是網絡上最長的路徑方面，國家合作網絡的直徑在1989到1998年是增長的情形，但在近十年則是減短；反之，機構合作網絡和作者合作網絡到2010年仍在持續增長。
測量三種層次合作網絡的節點的連結程度(connection degree)，其分布情形都符合冪次法則(power law)。節點的重要性可以從它們的程度中心性和中介中心性來推測，早期和近期的重要作者有很大的不同，在國家與機構方面的差別相當小。
比較三種層次的結果可以發現，較低層的結果會影響到上面的層次。例如有些作者的高排名不僅影響機構的排名，同時也會主導國家的排名；當具有高生產力的作者移動後，會牽動機構網絡的結構性變化。

Specifically, we would like to understand if and how collaborations at the author (micro) level impact collaboration patterns among institutions (meso) and countries (macro).

All 2,541 papers (articles, proceedings papers, and reviews) published in the international journal Scientometrics from 1978–2010 are analyzed and visualized across the different levels and the evolving collaboration networks are animated over time.

(1) USA, Belgium, and England dominated the publications in Scientometrics throughout the 33-year period, while the Netherlands and Spain were the subdominant countries;

(2) the number of institutions and authors increased over time, yet the average number of papers per institution grew slowly and the average number of papers per author decreased in recent years;

(3) a few key institutions, including Univ Sussex, KHBO, Katholieke Univ Leuven, Hungarian Acad Sci, and Leiden Univ, have a high centrality and betweenness, acting as gatekeepers in the collaboration network;

(4) early key authors (Lancaster FW, Braun T, Courtial JP, Narin F, or VanRaan AFJ) have been replaced by current prolific authors (such as Rousseau R or Moed HF).

Comparing results across the three levels reveals that results from one level might propagate to the next level, e.g., top rankings of a few key single authors can not only have a major impact on the ranking of their institution but also lead to a dominance of their country at the country level; movement of prolific authors among institutions can lead to major structural changes in the institution networks.

Scientometrics is a distinct discipline that performs quantitative studies of science and technology using mathematical, statistical, and data-analytical methods and techniques for gathering, handling, interpreting, and predicting a variety of features of the science and technology enterprise, including scholarly communication, performance, development, and dynamics.

In practice, scientometrics often requires the use of bibliometrics, the measurement of texts and information, and results might be presented as science maps (Börner 2010; Börner et al. 2003).

The study presented here uses papers that appeared in Scientometrics, the flagship journal of the field (Chen et al. 2002) publishing a major percentage of works in scientometrics as well as in the field of informetrics (Bar-Ilan 2008) over the last 33 years.

For example, Schoepflin and Glänzel (2001) used papers published in Scientometrics for the years 1980, 1989, and 1997 to identify a decrease in the percentages of both the articles related to the subjects of science policy and to the sociology of science.

Peritz and Bar-Ilan (2002) used papers published in Scientometrics for the years 1990 and 2000 and confirmed that Research Policy and Social Studies of Science are the third and fourth most frequently referenced journals in articles published in Scientometrics.

Hou et al. (2008) analyzed the structure of scientific collaboration networks in scientometrics at the micro level (individuals) by using bibliographic data of all papers published in Scientometrics from the years 2002–2004. They found that although half the authors had co-authored with each other, the network was not strongly connected and the collaborative network in the field of scientometrics was very loose.

Dutt et al. (2003) analyzed Scientometrics papers published during 1978–2001, examining the distribution of countries and themes and comparing institutions and coauthors to show that the research output is highly scattered, as indicated by the average number of papers per institution and dominated by single-authored papers; however, multi-authored papers are gaining momentum.

Chen et al. (2010) introduced a multiple-perspective co-citation analysis for characterizing and interpreting the structure and dynamics of co-citation clusters of the field of information science between 1996 and 2008. He showed that the multiple-perspective method increases the interpretability and accountability of both author-citation analysis (ACA) and document- citation analysis (DCA) networks.

Wagner and Leydesdorff (2005) applied network analysis to map the growth of international co-authorships, and they found that international co-authorships can be explained based on the organizing principle of preferential attachment, although the attachment mechanism deviates from an ideal power-law.

Samoylenko et al. (2006) visualized the scientific world and its evolution by constructing minimum spanning trees (MSTs) and a two-dimensional map of scientific journals using the Science Citation Index from the Web of Science database for 1994–2001 and showed a linear structure of the scientific world with three major domains: physical sciences, life sciences, and medical sciences.

Perc (2010) studied the evolution of Slovenia’s scientist collaboration network from 1960 to 2010 with a yearly resolution and showed the network had a ‘‘small world’’ pattern and its growth was governed by near-linear preferential attachment. This paper will advance the existing works by studying the evolution of scientometrics at three different network levels.

Figure 1 shows the growth (annual and cumulative) of the number of papers, countries (or regions), institutions and authors from 1978 to 2010. By counting the annual numbers in each figure, we obtain average annual growth rates, which are 20.4 % (papers), 9.4 % (countries), 19.6 % (institutions), and 20.1 % (authors).

As Bettencourt et al. (2009) pointed out, when fields grow, their collaboration networks densify—i.e., the average number of edges per node increases over time. They found that the relation between the number of nodes and edges followed a simple scaling law with scaling exponent (α > 1):



Figure 2 shows that the scaling exponent a equals 2.9533 at the macro-country, 1.5222 at the meso-institution, and 1.2353 at the micro-author levels. It has the highest value for countries—i.e., the country collaboration networks densify rather quickly, which is also due to the fact that this is the network with the fewest nodes. However, a large number of within-country or within-institution collaborations or an increase in single-authored papers would also result in smaller α values.

The diameter of a collaboration network has major implications for information diffusion—the shorter a pathway of coauthor linkages that connects an author pair, the more likely knowledge diffuses.

Over the 33 years, the country collaboration network diameter grew from 1989 to 1998 (there were no edges before 1989), achieves the highest value in 1998, and decreases in the last 10 years. This might be due to the rather limited number of countries that perform scientometrics research.

The diameters of the institution and author collaboration networks increase continually and both reach a diameter d = 15 in 2010.

A closer look at the density of the three networks (the ratio of the number of actual edges to all possible edges in a fully connected graph with the same number of nodes) shows that both the meso and micro networks’ densities decrease over time while the macro network, which experienced a topological transition from large to decreasing diameter, shows an increase in density.

In an attempt to understand the structure of the 1978–2010 networks, the degree for each node in the network was determined and the node degree distribution p(k) plotted in Fig. 3. ... All three networks exhibit power law degree distributions.

To understand which countries, institutions, and authors play key roles in the three networks, the degree centrality (the number of links a node has) and betweenness centrality (nodes that have a high probability to occur on a randomly chosen shortest path between two randomly chosen nodes have a high betweenness) (Freeman 1977) values for each node were calculated. The resulting TOP-5 countries, TOP-10 institutions, and TOP-10 authors calculated for every 6 years (cumulatively from 1978) are listed in Tables 1 and 2.

In addition, the last table column shows the TOP-10 countries, institutions, and authors if only 2001–2010 data is considered. While the differences are minimal for countries and institutions, the list of TOP-10 authors changes considerably if only recent works are considered.

Figure 4 shows that, by the end of 2010, Belgium, USA, England, Germany, the Netherlands, China, and France are central network nodes with a large number of papers. These six countries not only link to each other but also to outside countries—e.g., Belgium and Germany have strong links to Hungary, and Belgium and England have strong links to Finland.

When analyzing the evolving institution collaboration networks, it becomes clear that a few key institutions manage to stay in the TOP-10 list—among them are the Univ Sussex, KHBO, Katholieke Univ Leuven, Hungarian Acad Sci, and Leiden Univ.

During the evolution of the co-author networks, early authors are replaced by current authors. Most TOP-10 authors from 1980 and 1986 are missing in the later years. Key authors listed in the TOP-10 lists around 1986 decline in ranking or are replaced by other authors.

One might assume that rankings on the author (micro) level impact the ranking of institution (meso) and country (macro) levels. While author rankings impact institution rankings; institution rankings are less predictive of country rankings, as exemplified below.

As can be seen in Table 3, USA ranks first in the number of institutions and the number of papers over the 33 year time span. However, the average number of papers per institution was low for the USA, especially when compared with Belgium, Netherlands, and Hungary. ... Similarly, while no single author in the USA appears in the TOP-10 lists, the number of all authors combined and the number of their papers results in a high country ranking.

Can one single author impact the ranking of an entire institution or country? The answer is yes. ... The 155 papers of the Hungarian Academy of Sciences were co-authored with 30 institutions, 22 of which were contributed by papers authored by Glänzel W. As for the 93 papers by the Katholieke Universiteit Leuven, 13 of 51 institution links were added by Glänzel W.

Over the 33 years, the number of countries grew steadily with a linear growth feature with USA, Belgium and England leading in terms of centrality and betweenness. ... As their share increases, they have a stronger impact on the evolution of scientometrics. Over time, more and more collaboration links are generated and the average node degree and network density increase as well (see Table 4).

It is important to point out that some top-ranking countries have a small number of top-ranking institutions (e.g., Katholieke Univ Leuven in Belgium) while other countries (USA) have a large number of contributing institutions.

Similarity, some top-ranking institutions have one or two top-ranking authors, e.g., Glänzel W and Rousseau R

That is, single authors can not only have a major impact on the ranking of their institution but also of their country.

At the same time, the growth rate of institutions, authors and papers for each year were similar about 20 %. It suggested that this field had been attracting more and more institutions and authors to join the field of scientometrics.

The co-author network analysis showed that many new authors joined the field of scientometrics, especially in the recent 8 years. The diameter, average degree, and density of the network show the same trends as those calculated for institutions.

While co-author networks experience the departure of senior and the arrival of young researchers, the institution and country networks seem to have a comparatively stable structure of key nodes.