Few-Shot Charge Prediction with Multi-Grained Features and MutualInformation

Few-Shot Charge Prediction with Multi-Grained Features and Mutual Information Han Zhang1, Zhicheng Dou2,*,Yutao Zhu3,Jirong Wen4,5 School of Information / Renmin University of China1 Gaoling School of Artificial Intelligence / Renmin University of China2 Universit´e de Montr´eal Beijing Key Laboratory of Big Data Management and Analysis Methods Key Laboratory of Data Engineering and Knowledge Engineering, MOE zhanghanjl@ruc.edu.cn, dou@ruc.edu.cn Abstract Charge prediction aims to predict the final charge for a case according to its fact description and plays an important role in legal assistance systems. With deep learning based methods, prediction on high-frequency charges has achieved promising results but that on few-shot charges is still challenging. In this work, we propose a framework with multi-grained features and mutual information for few-shot charge prediction. Specifically, we extract coarse- and fine-grained features to enhance the model’s capability on representation, based on which the few-shot charges can be better distinguished. Furthermore, we propose a loss function based on mutual information. This loss function leverages the prior distribution of the charges to tune their weights, so the few- shot charges can contribute more on model optimization. Experimental results on several datasets demonstrate the effectiveness and robustness of our method. Besides, our method can work well on tiny datasets and has better efficiency in the training, which provides better applicability in real scenarios. 1 Introduction Charge prediction aims to determine the final charge (e.g., manslaughter, traffic offence, or theft) for a case by analyzing the textual fact description of the defendants’ behavior. As a subtask of legal judgment prediction (LJP), charge prediction plays an important role in legal assistance systems, thus has been widely applied in real scenarios. On the one hand, a charge prediction system can provide legal professionals, such as judges and lawyers, a quick and effective reference to improve their work efficiency. On the other hand, it can also provide non-legal professionals with some simple and useful legal advice. Automatic charge prediction has been studied for decades. Early studies focused on applying math- ematical or statistical methods, such as counting the specific attributes (e.g., crime time and place) of the cases (Kort, 1957; Keown, 1980). Later, researchers began to frame the charge prediction as a text classification problem and paid attention to designing manual features or extracting shallow features from fact descriptions to predict the charge (Liu et al., 2004; Liu and Hsieh, 2006; Katz et al., 2017). However, these features rely heavily on human expertise and are specific for different types of cases, which limits their application to a larger range of domains. Recently, deep learning based methods have also achieved promising results on charge prediction due to their superiority on automatic feature extraction and combination (Luo et al., 2017; Zhong et al., 2018; Yang et al., 2019; Xu et al., 2020). However, the charge prediction is still a non-trivial problem. One of the challenges is few-shot charges, which is also the focus of this paper. In practice, the numbers of cases in different charges usually follow a long-tailed distribution, which means their …

Content truncated to 500 words