Affective adaptive retrieval: study of emotion in adaptive retrieval

Recommendation, retrieval and browsing techniques are used to address the information overload problem. Current stateof-the-art approaches in these domains define the relevance of documents based on their content and in some cases the social dynamics of the collaborating groups. Relevance, often implemented as the topical relevance, is the fundamental principle of selecting an item for access. From a user’s perspective, ‘Interest’ in an item is the basis of access. ‘Interest’ is regarded by Izard as one of the eight fundamental human emotions [2]. According to the affective-cognitive and emotion-pattern theories, interests can trigger and be triggered by cognition as well as emotions. In other words, emotion plays an important role in the development of interest in humans. From this perspective, the role of emotion in influencing users’ interests seems to be neglected in the current information systems. It is our objective to investigate the role of (both reader’s and writer’s) emotion to characterise user interest. A system that provides quantitative and qualitative information about the emotional value and feelings represented in documents would be useful for a broad range of users, such as those looking for information about movies, trying to plan their vacations, searching for specific articles for or against specific subjects, or even marketing managers looking for consumers’ feelings on new products. A system that goes one step further and provides quantitative and qualitative information about the emotional value and feelings based on personalised user profile –consisting of concepts and emotions– will be even more useful [1]. We believe that adding emotion as a new dimension would improve the performance of adaptive retrieval systems. This improvement in effectiveness would vary depending on the context: in scientific context, the effect in effectiveness might be small, while in the social context we expect to see a higher Copyright is held by the author/owner(s). SIGIR’09, July 19–23, 2009, Boston, Massachusetts, USA. ACM 978-1-60558-483-6/09/07. increase. Systems recommending movies, news, music, or books are good examples of systems that can benefit from utilising emotion as an additional measure of relevancy. In addition, emotion can be used as a means to help users navigate through their personal information space, which consists of documents, RSS feeds, blogs, music, web pages, etc. The research questions motivating this research are (i) what is the role of emotions in adaptive retrieval systems?, (ii) how can we incorporate emotion in adaptive retrieval techniques?, and (iii) what are the elements of emotion and consequently the emotional space in the context of adaptive retrieval systems? The relation between the defined emotion space and other spaces (e.g. semantic) will be investigated in retrieval, recommendation, and browsing. In sentiment analysis and opinion finding tasks, much research is dedicated to extract the sentiment from documents (writer’s emotions). In our recent study, we have investigated the effect of affective feedback in improving the performance of content-based recommender systems (reader’s emotions) [1]. We have also showed the performance of collaborative recommendation systems can be improved significantly by integrating semantic spaces [3]. Given these evidence, we will investigate the reader’s and writer’s emotion and their role in recommendation, retrieval, and browsing scenarios. The main problems in constructing methodologies for our research questions are: (i) Techniques used in extracting emotion from users are very limited in both scope and variation, and still in their infancy, and (ii) extracting emotion from textual documents is an emerging field of study. Finally, (iii) experimental user-studies expose users to lab conditions. This lacks the ecological validity of a naturalistic study and might deteriorate the accuracy of emotional feedback given by users. Therefore, designing and implementing scientifically-viable user studies is a major challenge.