Combining Strategies For Xml Retrieval
Ning Gao,Zhi-Hong Deng,Jia-Jian Jiang,Sheng-Long Lv,Hang Yu
DOI: https://doi.org/10.1007/978-3-642-23577-1_30
2011-01-01
Abstract:This paper describes Peking University's approaches to the Ad Hoc, Data Centric and Relevance Feedback track. In Ad Hoc track, results for four tasks were submitted, Efficiency, Restricted Focused, Relevance In Context and Restricted Relevance In Context. To evaluate the relevance between documents and a given query, multiple strategies, such as Two-Step retrieval, MAXLCA query results, BM25, distribution measurements and learn-to-optimize method are combined to form a more effective search engine. In Data Centric track, to gain a set of closely related nodes that are collectively relevant to a given keyword query, we promote three factors, correlation, explicitnesses and distinctiveness. In Relevance Feedback track, to obtain useful information from feedbacks, our implementation employs two techniques, a revised Rocchio algorithm and criterion weight adjustment.