APPLYING SYNTACTICAL INFORMATION IN WEB SEARCH

Abstract

Web search could be much facilitated if we can better relate the user intention with the meaning of the web content. In this paper, we first survey the various existing methods, focusing on the dilemma that obtaining high accuracy results usually sacrifices the response time. We then propose a novel information retrieval framework to combine keyword-based search and search based on syntactical information. In particular, we design a sequential structure called LSC (Language Sequential Component) to encode syntactical information. Given a sentence, LSC provides a bridge from its syntactical representation and semantic meaning. We also propose a learning algorithm to obtain the LSCs from a training set, a classification algorithm to find the relevant LSCs from a user query to interpret the intentions of the user, and a search framework (called Semantic Search Engine) to incorporate syntactical information into a keyword based search system. Our experiments show the Semantic Search Engine outperforms the keyword-based approach significantly.