System and method to improve search result ranking through real-time analysis of user’s session behavior
Publication Date: 2010-Jul-29
The IP.com Prior Art Database
The field of this invention is related to information discovery. This invention is to improve user’s search experience with better ranked search result through real-time analysis of user’s search session behavior.
System and method to improve search result ranking through real -
Background and Problem:
Search engine is a common and powerful tool to find information using simple keywords. The challenge is that it is difficult for SearchEngine to provide users with most related information just based on simple keywords. Even for Google, a company started its business by ranking, how to find the most related information by understanding user's intention is still a challenge.
Two typical kinds of scenarios are given below:
1.1 Keywords has fuzzy meaning
For example, user searches with keywords "symphony", he is caring about music, not the IBM software product "Lotus Symphony". But state of the art service of search engine can't handle this.
(1) User input "symphony" in the search box
-time analysis of user's session behavior
time analysis of user's session behavior
(2) User click one search result about "music"
(3) User click "next page" for more search result about "music", but there are still many search result about "software" included that is not the result user wanted.
1.2 Search in vertical search engines
Vertical search engines have a lot of vertical domain knowledge, but just analyzing the keywords inputted by useris hard to get user's real search intention.
For example, user searched "mashup" in Google scholar. Without analyzing my selection of search results in the session, it can't give the most related results.
(1) User search paper by keyword "mashup"
User click through one paper about "security" of mashup.
(3) User click "next page" for more papers about "mashup". There are many papers not related to mashup security.
Prior arts and their drawbacks:
Known solutions about search result including:
Google Pagerank: http://infolab.stanford.edu/~backrub/google.html
Microsoft BrowseRank: http://portal.acm.org/citation.cfm?id=1390334.1390412 Yahoo ClickRank: http://portal.acm.org/citation.cfm?id=1557131
Personalized Rank: User can specify the interests in the profile and search engine will use the profile as the ranking factor when ranking the search result
Collaborative Filtering: Based on the analysis of massive user's history behavior, rank the search reslut.
Drawbacks of such known solutions
All the methods provided by the prior arts are ranking on the history data, which means user's behavior in the present searching session are not considered when search engine performs ranking. The solutions based on the analysis of history data have issues because:
History behavior can not reflect current user's...