With the decreasing cost and the increasing storage capacity, more web log data can be recorded nowadays. Compared with the data collected from experiments, the log data can more accurately reflect the user behavior with little bias. These data provide an opportunity to understand user behavior and help improve user experience. For example, by analyzing how users use search engines and why they are not satisfied with the search result, we can improve the usability of search engines, such as search personalization and search accuracy. However, analyzing log data is challenging. For instance, exploring the raw data is an essential step to formulate hypotheses and build models, but the log data size is large and increases over time. With the benefit that the human perceptual system...[ Read more ]

With the decreasing cost and the increasing storage capacity, more web log data can be recorded nowadays. Compared with the data collected from experiments, the log data can more accurately reflect the user behavior with little bias. These data provide an opportunity to understand user behavior and help improve user experience. For example, by analyzing how users use search engines and why they are not satisfied with the search result, we can improve the usability of search engines, such as search personalization and search accuracy. However, analyzing log data is challenging. For instance, exploring the raw data is an essential step to formulate hypotheses and build models, but the log data size is large and increases over time. With the benefit that the human perceptual system can process visual information rapidly, using visualization enables to express a large amount of information in a very efficient and intuitive way. Thus, in this case, visual analytics method, in which visualizations are the major components, can greatly help explore and analyze the log data.

In this thesis, we focus on two types of log data. The first one is the search log data, which record how users use different search engines to perform queries and is collected from a world wide distributed web browser. The second one is the learning log data from a Massive Open Online Courses (MOOCs) platform. In order to better understand the actual needs when analyzing the log data, we conducted several rounds of interviews with domain experts who are the end users of visual analytics systems. After that, we followed the user centered design and iteratively designed three analytics systems. In the first system, RankExplorer, we present a new visualization technique to intuitively show the ranking changes of queries in search log data. In the second system, LoyalTracker, we target on better understanding user loyalty and defection behavior in search log data. In the third system, VisMOOC, we focus on analyzing learning behavior through learning log data. All the three systems give domain experts new insights into user behavior. In order to validate the effectiveness and usefulness of proposed systems, we conducted case studies with domain experts and one user study for RankExplorer.