Incorporate User Behavior Information to Improve Web Search Ranking
Every day, millions of users interacted with internet search engines, issuing queries, following some of the links in the result, clicking on ads, spending time on pages and performing other actions. Actually, these actions could be served as valuable resource and information for tuning and improving web search ranking .
Implicit relevance feedback for ranking and personalization has become an active area of research. Lots of works has been done in this area and shown the value of incorporating implicit feedback into ranking process. However for user feedback information, a model need to be set cause user feedback actually is noisy.
To accurately interpret noisy user feedback obtained, Microsoft researchers, Eugene Agichtein, Eric Brill and Susan Dumais characterized this problem in detail (the first link bellow), where they motivated and evaluated a wide variety of models of implicit user activities. The general way is to represent user actions for each search result as a vector of features, and then train a ranker on these features to discover features values indicative of relevant search results. They first briefly summarize our features and model, and the learning approach in order to provide sufficient information to replicate their ranking methods and the subsequent experiments.
In the paper below (the second link), they used machine learning to derive a user feedback model using user actions as features, and compared their ranking methods over a random sample of 3,000 queries from the search engine query logs and more than 12 million user interactions with a major search engine. Finally, they established the utility of incorporating “noisy” implicit feedback to improve web search relevance.
Web ranking basically is oriented to people, and user interaction in some extent could reveal user’s preference and focus, which is a valuable resource for improving web rankings. I believe using user interactions to improve web search ranking would be helpful and valuable to establish a truly intelligent search engine.
http://delivery.acm.org/10.1145/1150000/1148175/p3-agichtein.pdf