Research And Implementation Of Multi-label Classification In Judicial Field

With the development of information technology,objective data accumu-lated by all walks of life,in order to cope with the challenges brought about by information explosion,urgently need some information technology means,analyze these data to help people solve some practical problems in real life,liberate Human productivity.In the judicial field,judges usually read the case description and decide on the final charge according to the relevant laws.This task is very time consuming and requires additional professional knowledge.Through technical means,the case description is taken as input,and the rele-vant law strip solves the problem of judgment prediction for output,which can effectively save labor cost and make the judicial judgment more accurate and effective.This topic transforms the judicial decision prediction problem into a multi-label classification problem.There are two main problems in multi-label clas-sification.On the one hand,the number of different labels appears to be very different,which is called label imbalance problem.The majority label is in In the process of learning,a small number of tags are often involved in the error calculation.This leads to a small number of tag classifications in the prediction process,which tends to predict the number of tags with a large proportion,resulting in inaccurate prediction results.Therefore,this problem needs to be solved.important question.On the other hand,the existing multi-label classifi-cation model often learns the label association information,which is called the label association problem.They often model the co-occurrence relationship of the label itself,ignoring the semantic information of the label itself.Based on the above problems,this paper proposes two models for decision prediction.Firstly,we propose a unifed Dynamic Pairwise Attention Model(DPAM for short).Specifcally,DPAM adopts the multi-task learning paradigm to learn the multi-label classifer and the threshold predictor jointly,and thus DPAM can improve the generalization performance by leveraging the information learned in both of the two tasks.In addition,a pairwise attention model based on article defnitions is incorporated into the classifcation model to help alleviate the label imbalance problem.Experimental results on two real-world datasets show that our proposed approach signifcantly outperforms state-of-the-art multi-label classifcation methods.Secondly,we propose a Recurrent Attention Network(RAN for short).RAN utilizes a LSTM to obtain both evidence and article representations,then a recurrent process is designed to model the iterative interactions be?tween evidences and articles to make a correct match.Experimental results on real-world datasets demonstrate that our proposed model achieves significant improvements over all of the baseline methods.Finally,a prototype system is designed and implemented.The prototype system is based on the basic needs of judges.Based on the above two models proposed in this topic,four functions are designed and implemented,namely,extracting elements of judgment documents,semantic matching of judgment documents,case retrieval and judgment prediction.The application of the model in actual scenarios is experimented,which effectively proves the practical application value of the model.