Learning Recurrent Event Queries for Web Search

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 1129–1139, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Learning Recurrent Event Queries for Web Search Ruiqiang Zhang and Yuki Konda and Anlei Dong Pranam Kolari and Yi Chang and ZhaohuiZheng Yahoo! Inc 701 First Avenue, Sunnyvale, CA94089 Abstract Recurrent event queries (REQ) constitute a special class of search queries occurring at regular, predictable time intervals. The fresh- ness of documents ranked for such queries is generally of critical importance. REQ forms a significant volume, as much as 6% of query traffic received by search engines. In this work, we develop an improved REQ classi- fier that could provide significant improve- ments in addressing this problem. We ana- lyze REQ queries, and develop novel features from multiple sources, and evaluate them us- ing machine learning techniques. From histor- ical query logs, we develop features utilizing query frequency, click information, and user intent dynamics within a search session. We also develop temporal features by time series analysis from query frequency. Other gener- ated features include word matching with re- current event seed words and time sensitiv- ity of search result set. We use Naive Bayes, SVM and decision tree based logistic regres- sion model to train REQ classifier. The re- sults on test data show that our models outper- formed baseline approach significantly. Ex- periments on a commercial Web search en- gine also show significant gains in overall rel- evance, and thus overall user experience. 1 Introduction REQ pertains to queries about events which oc- cur at regular, predictable time intervals, most often weekly, monthly, annually, bi-annually, etc. Natu- rally, users issue REQ periodically. REQ usually re- fer to: Organized public events such as festivals, confer- ences, expos, sports competitions, elections: winter olympics, boston marathon, the International Ocean Research Conference, oscar night. Public holidays and other noteworthy dates: labor day, date of Good Friday, Thanksgiving, black friday. Products with annual model releases, such as car models: ford explorer, prius. Lottery drawings: California lotto results. TV shows and programs which are currently running: American idol, Inside Edition. Cultural related activities: presidential election, tax re- turn, 1040 form. Our interest in studying REQ arises from the chal- lenge imposed on Web search ranking. To illustrate this, we show an example in Fig. 1 that snapshots the real ranking results of the query, EMNLP, is- sued in 2010 when the authors composed this pa- per, on Google search engine. It is obvious the ranking is not satisfactory because the page about EMNLP2008 is on the first position in 2010. Ide- ally, the page about EMNLP2010 on the 6th position should be on the first position even if users don’t explicitly issue the query, EMNLP 2010, because EMNLP is a REQ. The query, “EMNLP”, implic- itly, without a year qualifier, needs to be served the most recent pages about “EMNLP”. A better search ranking result cannot be achieved if we do not categorize “EMNLP” as a REQ, and provide …

Content truncated to 500 words