Kernel-based Similarity and Discovering Documents of Similar Interests

One of the continuing problems in information retrieval is searching documents of similar features. A number of methods have been developed for solving such a problem using latent topic analysis or its improvements. Anyhow, measures of similarity are crucial and play important role in finding out feasible solutions. It is dealt with paper a proposed method using diffusion kernel of term-network to set up a similarity measure and searching in a given corpus for documents that meet some specified similar features. In doing so, it is recognized some properties of similarity based on kernel in comparison with others, especially with measures of similarity based on an adaptive model of latent topic analysis named hk-LSA. Numerical experiments and statistical comparison are used to show evidently results of the proposed method.