(Not Too) Personalized Learning to Rank for Contextual Suggestion
Abstract
In this work, we emphasize how to merge and re-rank contextual suggestions from the open Web based on a user's personal interests. We retrieve relevant results from the open Web by identifying context-independent queries, combining them with location information, and issuing the combined queries to multiple Web search engines. Our learning to rank model utilizes three types of profiles (a general profile, a city profile, and a personal profile) to re-rank and merge the results retrieved from the Web. We find that the learning model generates better results when the user profiles' weights are biased heavily towards major personal interests. The detections of major, minor and negative personal interests are done by statistical analysis across users, examples, and context-independent query types. For user interests detected by query types, we call the interests "macro-level interests", while for user interest detected by examples, we call them "micro-level interests". In our experiments, we find that "micro-level interests" effectively avoid favoring too much towards rare query types such as spa and game, and thus yields more balanced rankings. Finally, for the top ranked suggestions for each user and context, we generate result descriptions by learning to rank favorable Yelp comments and using a natural language generation algorithm to generate positive comments.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2012
- Accession Number
- ADA579320
Entities
People
- Andrew R Yates
- Dave Deboer
- Hui Yang
- Nazli Goharian
- Ophir Frieder
- Steve Kunath
Organizations
- Georgetown University