information visualization
洞悉經常被認為是資訊視覺化的結果，但獲得洞悉的過程仍無法被了解，目前資訊視覺化的文獻將獲得洞悉的過程分為四類(提供全貌、調整、偵測樣式、比對心智模式)，這些過程提供了一些瞭解。Saraiya et al. [23]認為洞悉是觀察資料後的一個發現，可分為以下的四種情形：全貌(overview)、樣式(patterns)、群組(groups)以及細節(details)。North [16] 提出複雜(complex)、深入(deep)、質性(qualitative)、不可預期(unexpected)和相關(relevant)是洞悉的特性。本研究援引 Pirolli and Card [17]認為洞察為意義建構(sense-making)的一個部分，而意義建構如Klein et al. (p.71) [11]所定義之為有動機且持續的努力來了解人事間的連結，藉以預測它們進行的軌道(trajectories)與有效地行動。基於此一定義，意義建構有如下的現象：首先，意義建構的過程是周而復始的循環(cyclic and iterative)；其次，意義建構不只是發現的過程，更是創造的過程；最後，意義建構是回溯的(retrospective)，人們通常先建構一個架構，然後回溯地蒐集相關資料，並將它放到架構上，如果蒐集的資料與架構相符合，這個架構便獲得確定，否則人們會感到困惑，會捨棄、更正或取代原先的架構，來解釋新的資訊。Klein et al. [11]用Figure 1來說明上述的特性。
本研究以文獻分析探討資訊視覺化的研究者認為「人們如何透過資訊視覺化獲得洞悉？」的看法，歸納出四種不同但交融(intertwined)的方式：
1) 概觀(overview)：概觀描述人們對整個資料集合進行全盤了解的過程，能夠提供人們掌握已知與未知的事務、發現值得進一步探索的區域以及此一資料集合可以供新知的範圍。
2) 調整(adjust)：調整描述人們在探索資料集合時的過程調整抽象層級(the level of abstraction)與選取範圍(the range of selection)的過程。在探索大量資料時，運用選取功能可以進行過濾；群集(grouping)的功能可以聚集、簡化、組織和標示相關資料，使得大量資料能夠進行管理。
3) 偵測樣式(detect pattern)：偵測樣式表示發現資料集合內的特定分布、趨勢、頻率、離群(outliers)與結構。
4) 比對心智模式(match mental model)：資訊視覺化能夠降低了解時的認知負荷(cognitive load)，增強現有事物的再認知(recognition)，並且將呈現的視覺資訊與實際的知識連結起來。
循環交互地運用這些過程能夠提供對於資料集合的洞悉，可以對應到上述的意義建構理論。但由於這些過程得到的洞悉都相當抽象而高層次，資料本身的特質、使用者的興趣與背景知識對獲取洞悉有相當大的影響。
We found that: 1) Insights are often regarded as end results of using InfoVis and the procedures to gain insight have been largely veiled;
2) Four largely distinctive processes of gaining insight (Provide Overview, Adjust, Detect Pattern, and Match Mental Model) have been discussed in the InfoVis literature;
and 3) These different processes provide some hints to understand the procedures in which insight can be gained from InfoVis.
Saraiya et al. [23] have conducted insight-based evaluation studies in the domain of biology (analyzing biological pathways and micro-array data), and they define insight as “an individual observation about the data by the participant, a unit of discovery” (p. 444).
They (Saraiya et al. [23]) further group insights found in the context of bioinformatics into four different categories: overview (overall distributions of gene expression), patterns (identification or comparison across data attributes), groups (identification or comparison of groups of genes), and details (focused information about specific genes). (p. 445).
North [16] describes characteristics of insight as follows (p. 6):
• Complex. Insight is complex, involving all or large amounts of the given data in a synergistic way, not simply individual data values.
• Deep. Insight builds up over time, accumulating and building on itself to create depth. Insight often generates further questions and, hence, further insight.
• Qualitative. Insight is not exact, can be uncertain and subjective, and can have multiple levels of resolution.
• Unexpected. Insight is often unpredictable, serendipitous, and creative.
• Relevant. Insight is deeply embedded in the data domain, connecting the data to existing domain knowledge and giving it relevant meaning. It goes beyond dry data analysis, to relevant domain impact.
Sensemaking clearly begets insights as shown in the model of sensemaking (Information -> Scheme -> Insight -> Product) proposed by Pirolli and Card [17].
Sensemaking simply can be defined as “making sense of things” or, drawing from Klein et al. (p.71) [11], more comprehensively described as “a motivated, continuous effort to understand connections (which can be among people, places, and events) in order to anticipate their trajectories and act effectively.”
First, sensemaking procedures are cyclic and iterative. Russell et al. [22] describe the sensemaking procedure using Learning Loop Complex theory, which consists of 1) search for representations (generation loop), 2) instantiate representations (data coverage loop), 3) shift representations. As the name of their theory illustrates, the process of sensemaking is iterative in collecting data and re-generating a representation or scheme.
Second, sensemaking is not only a discovery procedure, but also a creation procedure. Weick [32], an organizational theorist, emphasized this notion of creation by providing a very clear distinction between interpretation and sensemaking. Weick argues that interpretation is a component of sensemaking, and states “the act of interpreting implies that something is there, a text in the world, waiting to be discovered or approximated. Sensemaking, however, is less about discovery than it is about invention” (p.13).
Third, sensemaking is retrospective. Literature in various contexts (e.g., [7,32]) supports that people often do not make sense of things after collecting information first. Instead, people often construct a framework first and retrospectively collect the relevant information and place it into the framework. If the collected information fits well with the framework, the framework is confirmed. However, if it does not, people become puzzled, and the framework could be discarded, updated, or replaced to explainthe new information.
These characteristics are well summarized by the Data/Frame Theory of sensemaking of Klein et al. [12].
Insight is not only an end result or simple discovery of hidden truth, but also an intermediate state in the iterative and cyclic procedure of sensemaking and invention. Insight could be a framework that needs to be created first in a person’s mind to draw the boundary of a problem and collect and understand information.
Hence, we conducted an extensive InfoVis literature review, specifically focusing on the following question: “How do people gain insight through InfoVis?” We reviewed 4 books, 2 book chapters, and 34 papers published in major venues in InfoVis. In order to consider the bigger picture, we initially focused on books and articles that survey the benefits of InfoVis, and latter reviewed case studies and evaluation studies to find some procedural aspects of insight.
In this section, we introduce four largely distinctive processes through which people gain insight while using an InfoVis system. ... The processes we have identified are: 1) Provide Overview, 2) Adjust, 3) Detect Pattern, and 4) Match Mental Model.
Provide Overview characterizes processes through which a person comes to understand the big picture of a dataset of interest. Even though observing an overview may not directly help a person gain insight, it appears to play an important role by helping people make sense of and find which areas they need to investigate more, thereby promoting further exploration of the dataset.
That is, Provide Overview allows people to grasp what they know and do not know, which areas are available for further investigation, and to what extent they could gain new knowledge from the dataset.
Adjust refers to a process through which people explore a dataset by adjusting the level of abstraction and/or the range of selection. Being able to flexibly change perspective on the dataset allows people to make sense of various aspects and test different hypotheses they have generated.
Selecting the range of a dataset to display by using a filtering interaction technique is a way to help explore a large amount of data.
Grouping is also an effective way to explore data by abstracting huge datasets into more manageable pieces. ... Through the process of grouping and aggregating, relevant information is gathered, simplified, organized, and labeled.
Detect Pattern means to find specific distributions, trends, frequencies, outliers, or structure in the dataset. ... A pattern itself could be an insight and further a person can cast a new questions and hypotheses by understanding patterns.
One of the benefits of InfoVis is that a visual representation of data can decrease the gap between the data and user’s mental model of it, thereby reducing cognitive load in understanding, amplifying human recognition of familiar presences, and linking the presented visual information with real-world knowledge.
Instead, these four different processes are intertwined and often used together to generate insights. For example, Provide Overview, as mentioned previously, often precedes further Adjust, and Adjust and Detect Pattern are often used together to gain deeper insight. This aspect clearly mirrors the cyclic and iterative characteristic of sensemaking.
More specifically, some of insights found in the literature are somewhat abstract and higher level, so that linking them with any of the categories is questionable.
Additionally, one of most important factors to help users gain insight might be the degree of users’ engagement into the dataset. ... The nature of data and users’ interests and background knowledge can also heavily affect the insight gaining procedure.
Usability seems to be another important aspect to promote the insight gaining process.
Clutter and occlusion are also examples of barriers for insight acquisition that need to be addressed. Sometimes too much data on a limited screen results in visual clutter and occlusion, which in turn diminishes the possibility of uncovering patterns and trends. Consequently, researchers have sought to reduce clutters in various ways [6].
