Study Of Related Information Of Specific Entity In Microblog Mining Algorithm

As a representative of Internet social applications rising with web2.0technology, microblog has gradually become an integral part of everyday life and this lead to the explosive growth of micorblog data. How to use the massive microblog data, how to meet the user’s query intention, how to mine relation information of specific entities has become focus of academic research.In this paper, we analyze the characteristics of microblog and propose a specific entity information mining system based on microblog data, called Weiyou system. There are three aspects studied, including Information retrieval in microblog environment, information mining of specific entity and recommendation system based on relation among entities. The main innovations and contributions of this paper are as follows.First, a query expansion algorithm is proposed, based on resistive network. In this algorithm, we use the concept of resistor in circuitry to simulate the relationship between different words in word space. The calculation of term relevance between complex words network can be effectively simplified with this algorithm. According to the result of TREC Microbolg Track, this query expansion method can meet user’s intention and improve the overall performance of information retrieval system.Second, this paper proposed a method to find the relationship between expansion words using word activation forces model. Using the concept of word affinity in WAF model we can easily calculate the relevance between expansion words so that we can find a disjunction of new search terms. This disjunction is used in the query reconstruction. It can be inferred from the experiment data that pairs of expansion words can reduce the information shifting caused by expansion words effectively while enhancing recall and precision of information retrieval system. Finally, a personalized recommendation system combining both user interest and contextual information is designed and implemented using word activation forces. This system achieves excellent result in TREC Contextual Suggestion Track, fully illustrating the effectiveness of WAF model in mining association among entities.