The Design And Implementation Of Judgment Recommendation System Based On NLP

With the need of social information transparency the society pays more attention to the judicial results of law cases. The increasing social comments on these cases bring reference for the judgment. Recommending similar judgment documents gives reference to judicial decision to the judge. Therefore, we design a judgment recommendation system to help with case judgment decision.Judicial Opinions of China website is a major source of judgment documents. It supports key word search but not sentiment search. For example if one search for noxious bean sprouts he cannot get any related or similar records such as rootless bean sprouts. Therefore, a sentiment similarity analysis system is needed. To learn the core content of a judgment document it’s necessary to extract the content of case fact and evidence, main conflict of litigious parties and applicable clauses. With the case fact and evidence content possible to be unclear it’s not sufficient to decide the similarity of two cases with simple key words comparison. Instead, sentiment comparison is needed for such situations.With the use of Natural Language Processing (NLP) it’s possible to analyze sentiment of context with the techniques of phrase partition, polarity annotation and dependency analysis. For example, noxious bean sprouts and rootless bean sprouts are both ATT construct regarding bean sprouts so they can be classified to a bean sprouts category with classification.Compared with general NLP the system designed in our paper focuses on the area of judgment content processing. A corpus of judgment documents is built with manual intervention with domain knowledge. The machine learning model is trained with the built corpus and then applied for this recommendation system. Our model outperforms in word partition, polarity annotation and dependency analysis compared with other existing models and shows success in the recommendation system.