The efficient and accurate retrieval of similar cases has always been an important task in the field of intelligent justice, while in this field, long legal documents and complex case information make it difficult for AI to quickly and accurately retrieve similar cases. This paper proposes a method of case retrieval based on key information of the case and summary information as an important basis for judging the similarity of cases, respectively constructs the extraction and generation model to extract the content of the two, and designs the position score type and logarithmic gain type result set reordering scheme, aiming to integrate multi-dimensional information ranking results to get the optimal reordering set. Through experimental results, it is proved that case key information and case summary information can fully represent the judicial characteristics of documents, and also solve the problem of inputting long documents. The reordering scheme also combines multiple dimensions of information such as case situation, summary, word frequency, etc., further optimizes the retrieval effect, and provides an innovative and practical solution to the problems in the classification case retrieval.