Research On Short-text Retrieval Query Expansion And Sorting Method For Microblog

With the further development and popularity of the Internet,microblog as a powerful network platform and social media is increasingly popular.Currently Twitter is popular in the worldwide and other domestic microblog such as Sina microblog and Tencent microblog both have huge user base,while generating hundreds of millions of content every day.Because microblog messages do not exceed the 140-character length limit,as well as writing free and mixed with many network terms and emoticons.With the rapid growth of microblog data,how to retrieve the valuable real-time information from the short text of the microblog is becoming more and more important.There are still many deficiencies in the solution of these problems with traditional information retrieval technology.In order to solve the above problems,based on previous studies,microblog as the research object,microblog short text retrieval related technology in-depth study.In this thesis,we start from relevance and real-time two sides,making the search results as relevance as possible and the search results relatively new.Firstly,this thesis introduces the query expansion technology for short text retrieval,including global query expansion and query expansion based on query.In this thesis,we introduce the global query expansion method based on word activation combine with the context and semantics to simulate the human brain to build the human brain knowledge model,and increase the breadth of query expansion.In this thesis,we introduce the improved pseudo-correlation feedback query expansion of the related model.We need to retrieve the query twice.In the first TOP K document(the default is the related document),we calculate the correlation with the original query term,find out the most relevant Query extension,improve the depth of the query to select the extension.This thesis introduces an improved pseudo-correlation feedback query expansion method that uses word activation global model and relevance model to query extension words under the language model,and improves the precision of selecting extended query words in terms of global and local,breadth and depth.After extracting query extension words,the second search is performed,and the search result is presented to the user as the final result.In the second retrieval,we use the way of weighting the original query words and extension words,calculate the correlation degree with the document,and combine the real-time features of microblog to re-sort the results of the second search.At last,we design a series of experiments,compared with the original models,and compared with the models without reordering function,this thesis shows that the improvement of the retrieval model in this thesis has improved the accuracy of query expansion and retrieval results.The related technology proposed in this thesis can effectively improve the retrieval efficiency and improve the user satisfaction.