SPEAR: Search Personalization with Editable Profiles
MetadataShow full metadata
Search personalization is an important technique for improving search performance. Existing approaches work in a black box, where users have no clue on how it works and how to customize it. This lack of user control and flexibility can often be inconvenient and counter-productive. In this thesis we propose SPEAR, a transparent search personalization framework that enables full user control and manipulation. In SPEAR, a user can own multiple profiles and each can be modified arbitrarily. Profile terms can be manually entered or automatically generated from search history or social network feeds. Furthermore, the terms can be automatically expanded by adding their semantically related derivatives. Negative terms are allowed for specification of negative preferences, which can be very useful in filtering out undesirable results. The in-use profile will help re-rank search results based on how consistent they are with respect to the profile. We implement SPEAR in the context of a Web search using Google Web search API and Facebook Graph API, demonstrating the promise and potential of the approach.