Semantics and Structure Based Recommendation of Similar Legal Cases

In order to realise the recommendation of similar legal cases, by integrating the latent semantics and structured data this paper proposes a method for calculating the similarity between criminal fact texts of two legal cases. In particular, due to the characteristics of different lengths of criminal fact texts and the high dimensional complexity of their features, we use TextRank algorithm to preprocess the criminal fact text of a legal case to realise the extraction of its key features. Finally, we conduct some experiments on 1,000 legal judgment documents of theft. More specifically, in term of the similarity between a recommended case and its real judgement, we benchmark our method with a state-of-art method and find ours significantly outperforms it.