Research Of Detection Algorithm And System Implementation For The Basis Of Contract Oriented House-renting

With the popularity of artificial intelligence,human can be saved from a lot of repetitive work.The combination of artificial intelligence and legal field can realize the automatic analysis and reading of legal documents.Using artificial intelligence technology for content understanding and simple audit of contract,compared with the previous manual audit,can greatly reduce the audit burden and improve the efficiency of contract audit.At present,in the rental market,many tenants will still be cheated by "black intermediary" to sign a "non-standard" risked contract.When signing the contract,some terms of the contract are unclear and ambiguous,and some risk points,which are easy to ignore and in which infringement disputes can happen frequently,are forgotten,which makes it difficult to protect the subsequent rights,and thus leads to the deterioration of the rental market environment and the increase of supervision difficulty.It is necessary for all parties to the contract(the lessor,the lessee and the intermediary)to audit the contract.At present,the contract review before signing is mainly conducted by professional lawyers.The audit work includes a large number of low-difficulty and repetitive reviews of basic terms,as well as examining whether the descriptions of the terms are standard and legal,and what risks and loopholes exist in the contract.In this paper,we detect the risk loopholes of the contract elements in the field of housing rental,which not only considers the completeness of the contract elements,but also includes the semantic audit of the elements.We propose the detection algorithm,detection framework and detection process,and design and implement a simple prototype detection system.This paper mainly studies from four aspects: the construction of text domain resources,text feature extraction,the construction of matching detection model for semantic similarity recognition of text pairs,and the design and implementation of detection system.The contribution and work in these aspects are as follows:1.Collect contract text data and build domain resources.From the perspective of domain legal experts,this paper constructs domain-related resources:(1)summarize the common professional terms related to the contract,to build a domain dictionary,and improve the segmentation effect;(2)Based on many contract templates and domain experts' opinions,extract and summarize a variety of standard contract elements for each audit point,and construct Essentials Library for the subsequent matching detection of housing rental contract.In order to improve the generalization ability of the followup model,this paper proposes the construction ideas and methods of domain related resources.(3)Based on the idea of mutual matching,this paper constructs the training corpus of manually labeled sentence pairs in the field;and uses the experience of predecessors for reference,uses EDA(easy data augmentation)text data enhancement to expand the corpus appropriately,so as to improve the generalization ability of subsequent models;in this process,this paper puts forward the construction ideas and methods of domain-related resources.2.Semantic feature extraction of the matched sentences.For each sentence,use the pre-trained Bert(Bidirectional Encoder Representations from Transformers)model to extract the sentence's text vector features,which can better combine the context information when learning the features.In addition,the literal matching method including the improved entropy method and Jaccard coefficient calculation,the method based on semantic role annotation and the method of syntax tree information extraction are used to extract the semantic similarity features of the text.As the supplement and enhancement of the vector features,these feature projects extract the key words and semantic information from multiple dimensions which can be as the inputs of the subsequent similar matching detection model.3.Training the matching detection model for the similarity recognition of mutual matching sentence pairs using the corpus combined with domain.In this paper,the full connection network and convolutional neural network are used as the basic model to build the recognition model,and different feature splicing methods and corresponding network structures are tried to build.The LCQMC semantic computing corpus and the domain training corpus constructed in this paper are combined as the comprehensive corpus of training model to carry out experiments and model selection.4.Integrate the detection model and apply it.The paper proposes a detection algorithm of contract condition matching,and designs and implements a convenient detection system to achieve that input the house lease contract and the system outputs the risk detection results of the contract elements.