Multi-Modal Feature Collaborative Fusion for Court Verdicts Recommendation

In order to improve the efficiency of judicial decisions, the judges need to search and preview the existing court verdicts which are similar to unsolved issues. To satisfy their needs, this paper proposes a court verdicts recommendation method based on multi-modal feature collaborative fusion, the aim of which is to recommend similar court verdicts with feature representation fused from various modalities. The execution phase of this proposed method is divided into four stages, i.e., data preprocessing, feature extraction, multi-modal feature collaborative fusion, and court verdicts recommendation. Current methods seldom consider the complete context semantics investigation and suffer from the inefficient fusion strategies. In this paper, we first extract features based on comprehensive analysis from lexical frequency, word-wise and document-wise feature learning. Additionally, a generic collaborative analysis network is proposed to fuse features based on the local detail enhancement and global semantic preservation. Various experiment results demonstrate that this method performs better than some other basic methods in precision, recall, and F1 value for court verdicts recommendation, which shows that the method proposed is highly effective for court verdicts recommendation.