Software Requirements Verification Based On Natural Language Processing

Digital technology begins to be widely used in nuclear power plants.As an important part of it,the verification and validation of safety-grade software in nuclear power has received widespread attention,especially for nuclear digital control systems.The process of software verification and validation is the verification and validation of the relationship between software requirements and software design.It runs through the entire life cycle of the software products.Therefore,seeking a simpler and more accurate verification and validation method of software requirement is of great significance for the development of nuclear power system and the utilization of nuclear energy.Structural integrity verification and description rationality verification of software features in software requirements documents are important parts of software verification and validation.This paper mainly studies the application of natural language processing in software requirement verification,the technology of part-of-speech tagging and text similarity calculation is mainly applied.The main work and innovations are as follows:1.In the training of Hidden Markov's part-of-speech tagging model,the training corpus is continuously expanded by semi-supervisedlearning.This method does not require a large number of labeledtraining corpora,and the iterative training of auxiliary corpus canhelp to automatically obtain the corpus with higher accuracy,savinglabor and time cost.2.Bidirectional Viterbi annotation is used in the process of labelingthe corpus with annotation model.The traditional hidden Markovlabeling model only considers the one-way transfer of part of speechbut misses the relationship between the current and the subsequentpart of speech.Bidirectional annotation method gets over theshortcomings of the traditional hidden Markov model,and improvesthe reliability of part-of-speech annotation.3.The method of calculating the similarity of words is improved bythe depth information of sememe.This method describes therelationship between words more accurate.