Research On Fine-grained Retrieval Technology Via Structuring Literature Corpus

The use of artificial intelligence and natural language processing to enhance the intelligence and naturalness degree of search engine,which is the driving force behind the development of modern search engines.This paper focuses on discovering user's potential query intents to provide more granular and accurate search results.Specific to the literature field,this paper improves the intelligence level and user experience of the search engine by using two query intent templates: problem and method,to identify the problem retrieval,the method retrieval and the problem and method retrieval.This paper works are as follows:1)Problem-method-oriented query model: In order to provide more intelligent search results,this paper propose a new query model,which summarizes user's information needs into problem needs and method needs.This query model can describe problemcentric,method-centric,and problem and method's hybrid query.2)Query intent analysis and matching: Firstly,use the named entity recognition technology to extract some specific types' entity,indicating the query's intent.Secondly,use the Markov Random Field model to calculate the joint probability of the query and documents,which can capture the intrinsic dependencies among query entities,to implement the query model defined by(1).3)Query expansion based on knowledge graph: In order to solve the problem of incomplete user query information,this paper proposes a query expansion model based on knowledge graph representation learning.First,construct the academic knowledge graph and then use knowledge representation learning to map the symbolized entity to a low-dimensional dense entity embedding.Last,this expansion method searches nearest neighbor entities to extend the query,which provides semantic approximation matching ability and alleviates the problem of query mismatch and incomplete information.4)Experiment and evaluation: The effectiveness of the proposed method is verified by a comparison experiment on the benchmark dataset(ESR dataset).Experiments show that the retrieval model proposed in this paper can effectively use the entity and entity type information to improve retrieval performance.The query expansion method proposed in this paper can efficiently alleviate the problem of incomplete user query information.