| The content of a contract involves the interests of the parties to the contract,so it is often necessary to carefully review the contract before signing it to check for omissions.With the development of the Internet and the popularization of electronic contracts,the prospects and needs of contract risk review applications have gradually become prominent.However,manual review of contracts is time wasting,less efficient,and costly.It is therefore worthwhile to assist the process of manual review to predict contract risks and improve review efficiency consequently reducing cost.This article combines deep learning algorithms to conduct in-depth analysis and research on contract risk prediction technology.First of all,this article uses BERTwwm-ext(Bidirectional Encoder Representations from Transformers Whole Word Masking extends)to process the whole word masking of Chinese word segmentation to resolve the problem of contract entity recognition.BERT-wwm-ext supports finetuning based on advantages of the pre-training model.So,a contract entity recognition algorithm based on BERT-wwm-ext is proposed;secondly,this article proposes a relationship extraction algorithm based on BERT+multi-head combined with entity location information,which addresses the problem relating to the existence of the same sentence in the contract relationship extraction task as well as the problem of a large number of unknown relational entity pairs.Then,the unstructured contract text content is converted into a structured information,combined with the constructed contract rule library which uses the contract rule similarity calculation method based on word2 vec combined cosine similarity with inherent designed support for the detection of missing clauses and clause content.Finally,combined with the above methods,this paper designs and implements a contract risk prediction system,and uses a house lease contract to verify the feasibility and effectiveness of the proposed algorithm and system. |
PDF Full Text Request