Topic Oriented Feature Location Based On RNNLM

Software feature location is the precondition to smoothly carry out software evolution activities,the evaluation scheme of the results of software feature location determines the range of software evolution,a good feature location method can improve the efficiency of software evolution.According to different analysis methods,the feature location can be divided into static feature location,dynamic feature location,text based feature location and integrated feature location.At present,the integrated feature location has the best performance,and the quality of integrated feature location depends on the quality of the integrated method.The traditional method mainly uses the word bag model to locate features.However,the word bag model regards key words in the corpus as an independent and identically distributed,difficult to retain word order information,which can not reflect semantic information and can not solve the problem of data sparseness.Because the code is a kind of special text symbols,but also has a rich context semantic relations.So the code can also be regarded as the expression of a kind of special language symbols,with RNN prominent in natural language performance.This paper proposes a topic-oriented software feature location method using RNNLM(Recurrent Neural Networks Language Model)in order to obtain better context semantic relations with RNNLM.First,the original system is divided at the class level to obtain the initial corpus.Followed by preprocessing,the preprocessing of this paper is not the same as the traditional pretreatment of the initial corpus for word segmentation,word and stem reduction and other operations.After the corpus is obtained,one-hot is used to digitize the corpus as input to the RNNLM,and the vector is generated at the same time.Finally,the mapping relationship between the class and the class lable is established.The same word vectorization is performed on the feature description,and then the similarity of the feature description with each class is retrieved as the output of the experimental result.Experimental results show that the proposed method has better precision.The evaluation method of the research results of software feature location affects the spread of software evolution,a good evaluation method can greatly improve the efficiency of software evolution.Combined with the method of this paper,ripple effect analysis and current evaluation methods,in this paper,a new method for the evaluation of the results of software feature localization is proposed,the data analysis shows that the evaluation method drastically reduces the scope of software evolution.