A new video search approach based on user's preference and query ambiguity

Common re-ranking algorithms such as PageRank and HITS are not suitable for text-based video search approaches. Methods that utilize addition information (such as video length, release time) require too much user participation and cannot meet the different need of the users with different preference. Personalized search that collect users' preference information implicitly and re-rank the search result automatically can be used to re-rank search results for the text-based video search approach. Most existing personalized approaches only utilize users' long term preference or short term preference. They ignore the ambiguity difference among queries and suffer from adverse affects brought by improper personalized strategy over queries with certain ambiguity. We propose a new search approach and develop a system called CMSearch. CMSearch utilizes both users' long term preference and short term preference to capture us.ers' preference and makes use of the query ambiguity to adjust the utilization of the preference. Performance tests show that the system archives an average Ranking Score improvement of 37.52%.