Real-Time Query Expansion Based On Dependency Relation Network

With the high-speed development of the Internet, search engines which can do efficient information retrieval have become an indispensable tool for everyone. By inputting some words into the search box, we can get the search results offered by search engine which might contain the information we need.But the queries are usually very short and ambiguous, so sometimes the search engines will misunderstand the true intentions of people and return unrelated information.Real-time query expansion is a kind of query expansion method which offers expansion words when the users are inputting the terms. It can help the users complete their queries, save keystrokes and disambiguate the intentions by recommending new query words. Traditional real-time query expansion techniques are mostly using keyword frequency methods based on query logs.First, this paper proposes a complete query intention representation method using the structure of "verb+attributes+noun". Then a dependency relation network which contains more than50,000nodes is built by analyzing a large-scale corpus which size is1.15G with more than915,600articles.Second, a real-time query expansion method is built using this network. Experiments shows that the successful expansion rate is84%using this method. And the keystrokes needed is significantly reduced.Finally, a complete search engine system is built. This system combines the real-time query expansion method described above and a string-based word completion method to offer real-time expansion words for users. Experiments shows63.75%percent of user keystrokes and clicks are saved and the nDCG score performance for search results is88.95%. The comparison with Bing proves this system has more stable expansion ability when the word order of user input changes.