Analyzing online search behaviors has become a major topic of interest in Information Retrieval and Information Seeking in order to improve search and retrieval capabilities. Since the way people search varies depending on information need, demographics, previous search experience, knowledge of the topic being searched on, cognitive abilities, and personal traits, understanding the user is a vital first step towards providing a personalized search experience. Although previous work has been done to build interesting user behavioral models, there is a lack of research regarding developing models for exploratory search tasks. Given that exploratory search is open-ended, multi-faceted, and lacks clear boundaries, user behaviors during an exploratory search task are highly variable. The importance of search assistance applicability and predictions to improve the user experience and user goal achievement also becomes paramount.
We have shown how user search behavior can be represented as patterns of actions sequences and how these patterns could provide insights into different stages of the search process. By understanding these patterns we could then provide recommendations that improve users’ search performance. We have also shown how to predict user search performance successfully using a time series of implicit features extracted from the search process. Using sequences of actions to represent user behavior with unsupervised learning, we have also shown how user activity sequence patterns can be used in identifying different search task types effectively.
We are also building a framework to evaluate the search performance of individuals who are performing exploratory search tasks. We incorporate factors that define exploratory search such as exploration, creativity, and knowledge discovery in evaluating the user search process over time and integrate user search behavior patterns in understanding the underlying process that guides exploratory search. Since the underlying search process is unique for each individual, this approach tries to provide a more personalized understanding and evaluation of user search performance.
