A Research On Query Expansion Of Entity Information In Object Retrieval

In this paper, a topic is deeply discussed that a research on query expansion of entity information in object retrieval. Nowadays, the demand of information retrieval has begun to change, from traditional web search to object retrieval. It has enhanced the technology of entity information extraction to a high significance whose one of the most important part is query expansion of entity information. The purpose of entity information extraction is to build up an entity knowledge base automatically. There are two functions of query expansion of entity information:one is to enrich the information of entity query, and the other one is to obtain the specific attribution, alias et al., of the entity as to achieve the linking of the entities possessing the relation of coreference.The main content of this paper are listed, as follows:To start with, the characteristics of object retrieval is compared with that of traditional information retrieval in certain aspects, like pre-processing, word detection, relevance calculating, et al. And on this basis, the main research task is decided to be entity information expansion based on statistics and syntactic knowledge, respectively.Then, to solve the problem of obtaining the terms highly relevant to entity query, a method of entity information expansion based on statistics is proposed. Using relevance feedback and hierarchical clustering, the relevant terms can be acquired according to co-occurrence similarity between entity query and terms in the documents. Based on this model, more than two thousand entities have been expanded with the result of top five relevant terms. And the result in ad hoc task of TREC2012Microblog has proved the effectiveness of this method.Eventually, to extract the alias and synonyms of the entities, a model of entity information expansion based on syntactic knowledge is proposed. With lexical analysis and grammar matching, the coreference resolution of the entities in the documents can be achieved, as the result that the semantic information successfully obtained. The effectiveness of this model has been proposed when exploited in two subtasks of TAC2012KBP.