Analytic provenance for sensemaking behavioural markers in visual analytics

Studying how analysts use interactions in visualization systems is an important part of evaluating how well these interactions support analysis needs, making sense of data or generating insights. As sensemaking is inherently a fluid activity involving transitions between mental and interaction states, lack of accuracy or precision into adopted visualization techniques can create a gap between cognitive constructs and manipulations or interactions humans apply to think or reason about the data. To tackle the problem, this thesis proposes ‘Behavioural Markers (BMs)’ which are representatives of the action choices that analysts make during their analytical processes as the bridge between that gap. Appropriate tools, techniques are required to log individual processing activities and utilize those to complement the information entailed with transparent processing operations.

As a first step to achieve the goal of bridging between human cognition and analytic computation through interactions at micro-analytic level, this thesis contributes to an extensive research with groups of real police intelligence analysts for designing and developing a visual judgemental system named as ‘PROV’ according to W3C standard. Secondly to explain how human cognition leads to interactions and vice versa, it contributes to development of an exhaustive list of behavioural constructs and detection of those ingrained cognitive constituents through interaction network graph analysis and translate those by theories of psychology for externalizing human thought processes. Recovering cognitive reflection on analytic reasoning processes from extended log data or only by observing is a difficult task. Due to cognitive and perceptual variances, conventional clustering or pattern mining techniques for user behaviour modelling, task identification, clickstream modelling don’t fit very well with this purpose. To overcome these limitations as third step, this research proposes ‘BreakPoints (BPs)’ as the way to pinpoint internal transitions in perception and cognition which are nipped into analytic interactions. This research has contributed to development of machine learning models to contextualize those streams of actions, infer cognitive transition points into both known and unknown task scenarios. Proposed approaches have significantly improved results compared to existing techniques. Finally for transparent validation of all computational outcomes in terms of reliability, accuracy, relevance and to build human trust on those results, this research has presented visual explanations of machine produced results by unfolding blackbox calculations.

The major research results reported into this thesis have contributed to the project VALCRI (Visual Analytics for Sensemaking in Criminal Intelligence), Analysis which has received funding from the European Union Seventh Framework Programme FP7/2007-2013 through Project VALCRI, European Commission Grant Agreement No- FP7-IP-608142, awarded to Middlesex University and partners.