Research On Automatic Extraction Of Construction Contract Risk Terms Based On Natural Language Processing Technology

With the substantial growth of the scale of construction projects and the increase of engineering complexity,higher requirements are put forward for the writing and review of construction contracts in order to ensure the contractual status and rights between all parties involved.Most international construction projects require the contract management team to review all possible risks in the contract during the bidding period.However,in the face of such a huge construction contract text,it is very difficult to examine the risks in a short time.Therefore,there is an urgent need for an automatic audit tool to assist people to complete their work.However,most of the current automated audit tools can only audit the most basic format problems,and cannot audit semantic risks;And most of the systems are oriented to the general field,and there are few proprietary automatic review systems for the design of architecture field.In view of this,based on natural language processing technology,this paper carries out the research on automatic extraction of construction contract risk terms.The work content is divided as follows:(1)Aiming at the problems of high abstraction and complex semantic information of construction contract text,Albert-bilstm-attention-CRF model is proposed for entity recognition of contract text.The model uses Albert to represent vocabulary,bilstm and attention to extract text features,and CRF decoding module to predict entity tags.Through experimental verification,the comprehensive F1 value of the model in the self built construction contract data set reaches 87.2%;Compared with the existing model,the experimental effect is the best,and compared with the Bert pre training model,it has the advantages of less parameters and less training cost.(2)Aiming at the problem of multi relationship classification of construction contract text,a Bert+Multi-head relationship extraction model combined with entity location is proposed.The model integrates the entity location information into the Bert pre training model to extract text features,and then uses the multi head multi head classifier to complete the classification of multiple relationships.Compared with the model without entity location information and using linear relationship classifier,the model F1 proposed in this paper has the best effect on the self built construction contract data set,and the introduction of multi head classifier reduces the overhead of memory space.(3)Facing the demand of automatic risk review of construction contract,this paper constructs an automatic risk clause extraction system combined with natural language processing technology.Firstly,semantic retrieval and data mining technology are used to mine risk sources and associate relevant legal provisions,then the risk level of construction contract is defined and the risk clause matching rule base is formulated.Finally,semantic similarity technology is used to extract risk clauses.After practical testing,the system has high recognition efficiency and can meet the needs of engineering application.In this paper,natural language processing technology is used to meet the task requirements of automatic extraction of construction contract risk terms.Through the processes of text information extraction,risk source identification and risk level definition,risk rule base and matching mechanism,an automatic extraction system of construction contract risk terms is finally completed.Thus,it supports the intelligent risk management of construction enterprises,promotes the intellectualization of construction field,and has great practical application value.