Abstract:HITS is one of the classical web link analysis algorithms, the main problem of it is the topic drift and mutual reinforcement. According to these problems, we propose an improved algorithm T-HITS, using a network structure to map spam collection with corresponding website and exclude the spam links with link text. At last we use the model of credibility to fix the results, and the experimental data shows that the improved algorithm has a big improvement about the degree of correlation of the results, and can decreases the probability of the topic drift.