Similar Document Search and Recommendation

Abstract

Query formulation is one of the most difficult aspects of search, especially for a novice user. We propose a new search interaction where the user searches with a reference document and the system learns from the user inputs over a period of time to “push” relevant and new content without additional user interaction. Our method is based on identifying key phrases from the input document. The key phrases are used to query a search engine and the results are evaluated for similarity to the original document. By caching documents received from a user over a period of time, a user profile is built. The profile is then used to provide recommendations to the user. Evaluations show that this method has a good precision in finding documents of interest to the user. Also our key phrase extraction method has good recall in retrieving the input document. Additional experiments reveal that our recommendation system is of help in exploring documents of interest to the user.