Abstract: Taking into account that the case cannot be quantified, the limited experience of business judge, the lack of unified standards for the identification of similar cases and other factors, a typical model of court texts based on topic model is proposed in this paper. The recommendation is given by regular and weighted method according to the number of years of judgment and corresponding criminal law from the maximum similarity text in court texts based in this model. While the highest similarity of the case as the best recommendation case to the user. Experimental results show that the method based on topic model proposed in this paper is superior to the traditional text similarity measure in F-metrics.