Analyzing Law Decision Texts for Personalized Recommendation

The legal domain grows every day since the legislations are updated or new legislations are defined daily. Thus, it can be a challenge, especially for non-lawyers, to find appropriate legislations due to a legal problem. Since most of the people can not nd relevant legal information, they rely on legal platforms. This thesis aims to predict a personalized recommendation for a given user issue according the law. Therefore, over 600,000 law decision texts are analyzed by the word embedding algorithms Doc2Vec, GloVe and FastText. These algorithms allow to sum up all similar law decisions together, so that a pool of corresponding decisions can be recommended to a given textual issue. Detailed evaluations of the legal recommendations prove that it is a difficult issue to create a personalized legal recommender system. Thus, further researches are necessary as the analyzed word embedding algorithms alone are not suitable for a personalized legal recommender system.