Learning to Rank Documents Using Similarity Information between Objects

摘要

Most existing learning to rank methods only use content relevance of objects with respect to queries to rank objects. However, they ignore relationships among objects. In this paper, two types of relationships between objects, topic based similarity and word based similarity, are combined together to improve the performance of a ranking model. The two types of similarities are calculated using LDA and tf-idf methods, respectively. A novel ranking function is constructed based on the similarity information. Traditional gradient descent algorithm is used to train the ranking function. Experimental results prove that the proposed ranking function has better performance than the traditional ranking function and the ranking function only incorporating word based similarity between documents.