Personalization Approaches for Ranking: A Review and Research Experiments

Personalization Approaches At Querying Stage

Query logs record user’s search queries and related actions; which is an important source to learn user’s search behavior implicitly. Web Search technology may collect information about user in general and then identify groups of user to improvise ranking. Users may explicitly specify interests, demographics or cognitive characteristics. One disadvantage of such explicit feedback is that it is difficult to collect and maintain it for temporal personalization. The user models can also be built implicitly based on content the user has clicked on or the web history of interaction. These models are suitable for personalization using query adaptations or query modification where query is improvised to better represent the information need based on the user model. Short queries can be augmented with additional words to minimize the vocabulary problem, such as polysemy and synonymy, which are prevalent in keyword-based search. Alternatively, if the query retrieves a smaller number of resources than expected, it is possible to expand it using words or phrases with a similar meaning or applying some other statistical relations to the set of relevant documents [Micarelli et al.2007]. The major advantage of query modification approach is that the amount of work required to retrieve the results is the same as in the non-personalized search. On the other hand, query modification approach is less likely to affect the result lists as it does not affect all stages of ranking process.