A Modified System for Weblog Topic Relevance Retrieval

Weblog is widely used, and the number of users is increasing rapidly. Weblog reflects every aspect of the society, such as politics, economy and culture, so the topic relevance retrieval research on Weblog becomes necessary. Because of a lot of noise in the corpus and it is usually difficult to obtain the appropriate query, the common methods sometimes fail to reach an acceptable precision. We design a Modified Topic Relevance Retrieval System (MTRRS) containing query formulation and a combination model. To design the query, manual adjustment and machine learning are used. During the machine learning processing, we define a center word list which helps to generate a novel distance feature. The result can be improved 22.97% on MAP by query formulation. The results of document retrieval model and passage retrieval model are combined. 33.55% increase on MAP can be received. Also by using the combination model, the retrieval result of the semi-machine learning query is closely approaching the manually adjusted result.