Research On The Method Of Software Feature Mining And Recommendation Based On Community Discovery

Currently,a large number of software feature descriptions in natural language have accumulated in the software application market.Extracting the common features of software products from these natural language descriptions and recommending them to new software developers have gained a lot of attention in recent years.However,how to extract features that can clearly express the meaning of a certain function when a single sentence in the software product description may involve multiple functional features and how to find users' favorite features remain the issues.To address these issues,this paper proposes an overlapping community detection based feature mining method and a user oriented feature recommendation method.Our main works are as follows.(1)This paper proposes a method for extracting software features by discovering clusters of sentences in the description of software products.The method is based on the LMF overlapping community detection algorithm.Firstly,through computing the similarity between sentences in software descriptions,a similarity network between sentences is constructed.Then,the sentence community is detected from the similarity network in order to extract the features implied in the descriptions.Each sentence community represents a feature,which contains all potential sentences that describe the feature.There may be overlapping sentences between sentence communities.These sentences are the ones describing more than one feature.In order to better understand the features implied in the sentence communities,this paper selects the community with the smallest entropy from all the sentence communities iteratively,and selects the most representative sentence which is not selected by other communities as the feature descriptor.(2)This paper proposes to mine the association rules between features from the perspective of software users,aiming at identifying and recommending users' favorite features for designers.Therefore,through assuming that the downloads of the software products can reflect the users' preference for the software products,and each product download represents the users' support for the product's features,this paper uses the association rule mining algorithm to mine the associations between the features in the users' perspective.(3)This paper crawls the software product descriptions from the Softpedia.com as experimental data.In the feature extraction experiment,it compares with the traditional algorithm IDC,and evaluates it by accuracy and time efficiency.The software feature extraction method of the algorithm has better performance.In the feature recommendation experiment,from the perspective of software users and software developers,through the evaluation of accuracy,recall,F1-Measure,the software feature recommendation method for software users proposed in this paper has better performance.