KEYPHRASE BASED USER PROFILES IN SHORT-TERM AND SESSION-TERM QUERY LOGS

Personalization depends on prior knowledge about information retrieval and web search and aims to build accurate and detailed user models. Thus, in the first step, it has to present a definition of a user model and in the next step, the contexts of what type of data is used in user-profiles and the models to represent them have to be provided. The structure of the user profile always is an important issue because of its impact on ranking performance. It is clear that if the algorithms used in the user model are more accurate and robust, the user model and personalized services will result in better efficiency and quality. Therefore, creating an efficient user profile is a challenge. Our motivation is to develop a keyphrase-based profile that operates on documents to improve personalization. These profiles are created using the keyphrase based models on the query log, as long-term, short-term and session-term to consider user interest in different time intervals to compare efficiency. Besides, we conduct comparative research on topic-based user profiles, intending to compare keyphrase-based and topic-based profiles in the personalization process. The results obtained more accuracy in session-based models by 13% in mean reciprocal rank and 14% in normalized discounted cumulative gain than long-based models.