information visualization
許多研究都指出資訊視覺化的目的是提供洞悉(insight)，例如 Card, Mackinlay and Shneiderman [1]與Thomas and Cook [2]。然而目前大多數對於洞悉的定義卻是莫衷一是，例如North便有兩種不同但相關的看法。North [3] 認為洞悉的特徵包含複雜(complex)、深入(deep)、質性(qualitative)、不可預期(unexpected)以及相關(relevant)，這一類的看法與認知科學上的突發性洞悉(spontaneous insight)相近，將洞悉認為是靈光一現的剎那(a moment of enlightenment)，也就是問題從不知道如何解決忽然轉移到知道如何解決的一個過程，並且這類的看法需要注意的是這類的問題解決的過程並非依循尋常的方式，而是在僵局下，透過微弱的語意網絡，忽然激發較不清楚地相關資訊，所產生的典範轉移(paradigm shift)。但North與其同事[5]也曾提出另一種關於洞悉的看法，他們將洞悉定義為參與者對於資料的一種個別觀察(an individual observation)以及是一個發現的單位(a unit of discovery)，這種看法可以說是將洞悉視為是一種知識的進步(an advance of knowledge)或是一片段的資訊(a piece of information)，這種看法認為視覺化能幫助知識的建構，例如Yi et al. [6]以意義建構理論(sense-making theories)為基礎將視覺化能產生的洞悉分為四個不同但彼此交疊的過程：提供全貌(provide overview)、調整(adjust)、偵測樣式(detect patterns)以及比對心智模式(match mental model)。
由於自發性的洞悉來自於語意知識(semantic knowledge)不可預期的重組(reconfiguration)，要對一個問題產生自發性的洞悉必須具有相關知識。在另一方面，自發性的洞悉所引起的典範轉移(paradigm shifts)能夠讓人對於問題的瞭解產生新的結構與關係。因此，本研究認為這兩種看法在習得知識的循環上彼此相互支持，以Figure 2來表示。在僅有有限知識的一開始(0 to k1)，使用者並無法產生自發性的洞悉。當知識逐漸增加後(k1 to k2)，能夠產生自發性洞悉的可能性增加。最後(k2 to k3)，愈多的知識能夠產生自發性洞悉的可能性愈增加，但趨勢逐漸減緩。所以，應提供一個環境讓兩種看法的洞悉都能發生。
Many have argued that providing insight is the main goal of information visualization. Stuart Card, Jock Mackinlay, and Ben Shneiderman declare that “the purpose of visualization is insight,” [1] while Jim Thomas and Kris Cook propose in Illuminating the Path that the purpose of visual analytics is to enable and discover insight [2].
For example, Chris North categorizes insight to be “complex, deep, qualitative, unexpected, and relevant,” [3] which overlaps with the neurological definition.
However, North and his colleagues also define insight as “an individual observation about the data by the participant, a unit of discovery,” [5] which does not bear any clear relation to the strict aha moment of cognitive science. Instead, it implies a focus on knowledge-building not found in the cognitive definition.
We suggest that what the visualization community defines as insight actually has two parallel meanings: a term equivalent to the cognitive science definition of insight as a moment of enlightenment, and a broader term to mean an advance in knowledge or a piece of information.
The cognitive science community has used the term insight “to name the process by which a problem solver suddenly moves from a state of not knowing how to solve a problem to a state of knowing how to solve it.” [8]
In this tradition, spontaneous insight is a type of problem solving and differs from normal problem solving in several key ways.
First, spontaneous insight doesn't appear to be facilitated by gradual learning heuristics such as bottom-up inductive reasoning. In fact, researchers have observed that focused effort on normal problem solving often inhibits spontaneous insight. Spontaneous insight usually occurs when a person is in a relaxed state [9] (such as when taking a shower in the morning).
Second, whereas gradual problem solving requires no special inducement other than presenting someone with a problem, what precipitates spontaneous insight is still being discussed. One commonly held theory is that spontaneous insight often occurs when a person tries to solve the problem in a habitual way, fails, momentarily becomes frustrated (perhaps owing to incorrect assumptions or some other cognitive fi xedness), mentally reorganizes the pieces of the puzzle (perhaps by breaking through a failed thought paradigm), and “suddenly” sees the solution. [8]
Finally, in normal problem solving the path taken to the solution is conscious and logically clear to the problem solver; however, participants who experience a spontaneous insight often can’t describe the thought process that led to it, [10] indicating that this insight occurs subconsciously and isn't a process that can be directly controlled, manipulated, or repeated.
This indicates that normal problem solving involves a narrow but continuous focus on information highly relevant to the problem at hand. ... . This suggests that spontaneous insight occurs through sudden activation of less clearly relevant information through weak semantic networks, which corresponds to a participant’s paradigm shift following an impasse.
These findings suggest that spontaneous insight is qualitatively different from everyday problem solving. It involves a unique pattern of neural activity that corresponds with the unique sensation of
the “aha” moment that participants report.
Recently, Yi and his colleagues provided a comprehensive survey on information visualization literature that considered insight as a goal or a measurement [6]. On the basis of sense-making theories, they concluded that four distinct but intertwined processes in visualization can lead to insight: provide overview, adjust, detect patterns, and match mental model.
In the visualization community, researchers often talk about discovering insight, gaining insight, and providing insight. This implies that insight is a kind of substance, and is similar to the way knowledge and information are discussed.
In the cognitive science community, researchers more often discuss experiencing insight, having an insight, or a moment of insight. In this context, insight is an event.
On the basis of the cognitive definition of insight, this statement restricts visualization into considering only a specific mode of problem solving that produces results that, although measurable, aren't easy to track.
On the other hand, considering insight only as knowledge or information limits visualization’s potential to structured knowledge building and information display.
If spontaneous insight comes from the unexpected reconfiguration of semantic knowledge, [10] then relevant knowledge about a problem must be necessary for spontaneous insight to arise. ... Conversely, the major paradigm shifts associated with spontaneous insight can create new structures and relationships in a user’s understanding of a problem, which can then serve as the schematic structures needed for generating future knowledge-building insights.
Together, the two types of insight support each other in a loop that allows human learning to be both flexible and scalable.
As Figure 2 shows, when the user has only a limited amount of knowledge (0 to k1), spontaneous insight won’t likely occur.
As the amount of knowledge increases (k1 to k2), the probability of spontaneous insight increases sharply.
Finally, after a certain point (k2 to k3), further increase of knowledge increases the probability in only a limited fashion until it’s asymptotically close to a spontaneous insight occurring.
On the other hand, a reduction in the probability of gaining a spontaneous insight undoubtedly occurs, at least for a while, if the user is distracted from this freer knowledge association.
But whatever model is chosen, our main point is that spontaneous and knowledge-building insights should be considered distinct because the best approaches to gain one or the other are different.
For spontaneous insight, we can evaluate exploratory, “prequery” approaches that keep one “in the cognitive zone” or “in the flow,” and quantitatively identify when a spontaneous insight occurs through an EEG or fMRI.
For knowledge-building insight, we can evaluate detailed knowledge-gathering methods and look to appropriate user studies to measure how much knowledge a user gains.
Using these combined approaches, we can not only more accurately determine visualization tool’s effectiveness, but also provide cognitive scientists with more complex problem-solving artifacts (they have few available) and shed light onto how to promote the two types of insight through visualization tools to solve real-world problems.
Recently, Yi and his colleagues provided a comprehensive survey on information visualization literature that considered insight as a goal or a measurement [6]. On the basis of sense-making theories, they concluded that four distinct but intertwined processes in visualization can lead to insight: provide overview, adjust, detect patterns, and match mental model.
In the visualization community, researchers often talk about discovering insight, gaining insight, and providing insight. This implies that insight is a kind of substance, and is similar to the way knowledge and information are discussed.
In the cognitive science community, researchers more often discuss experiencing insight, having an insight, or a moment of insight. In this context, insight is an event.
On the basis of the cognitive definition of insight, this statement restricts visualization into considering only a specific mode of problem solving that produces results that, although measurable, aren't easy to track.
On the other hand, considering insight only as knowledge or information limits visualization’s potential to structured knowledge building and information display.
If spontaneous insight comes from the unexpected reconfiguration of semantic knowledge, [10] then relevant knowledge about a problem must be necessary for spontaneous insight to arise. ... Conversely, the major paradigm shifts associated with spontaneous insight can create new structures and relationships in a user’s understanding of a problem, which can then serve as the schematic structures needed for generating future knowledge-building insights.
Together, the two types of insight support each other in a loop that allows human learning to be both flexible and scalable.
As Figure 2 shows, when the user has only a limited amount of knowledge (0 to k1), spontaneous insight won’t likely occur.
As the amount of knowledge increases (k1 to k2), the probability of spontaneous insight increases sharply.
Finally, after a certain point (k2 to k3), further increase of knowledge increases the probability in only a limited fashion until it’s asymptotically close to a spontaneous insight occurring.
On the other hand, a reduction in the probability of gaining a spontaneous insight undoubtedly occurs, at least for a while, if the user is distracted from this freer knowledge association.
For spontaneous insight, we can evaluate exploratory, “prequery” approaches that keep one “in the cognitive zone” or “in the flow,” and quantitatively identify when a spontaneous insight occurs through an EEG or fMRI.
For knowledge-building insight, we can evaluate detailed knowledge-gathering methods and look to appropriate user studies to measure how much knowledge a user gains.
Using these combined approaches, we can not only more accurately determine visualization tool’s effectiveness, but also provide cognitive scientists with more complex problem-solving artifacts (they have few available) and shed light onto how to promote the two types of insight through visualization tools to solve real-world problems.
沒有留言:
張貼留言