| The traceability between artifacts such as code,requirement documents and test sets,make a great impact on software understanding,impact analysis,and software maintenance.However,establishing traceability between software artifacts is time consuming,tedious,and may involve unforeseen difficulties.Therefore,establish the traceability between software artifacts automatically has become one of the most representative and challenging work in the academic community.Information retrieval(IR)is now the most widely accepted and applied technique in the research of traceability recovery.In general,typical IR-based approaches compute the textual similarity between two software artifacts and generate candidate trace links list.Unfortunately,the vocabulary mismatch problem between requirement and code artifacts makes the accuracy limited.So it is difficult to apply IR approach in practice.To address this issue,researchers have successfully proposed enhancing strategies from different perspectives.The two types of enhancement strategies below are research hotspots in the current field.A growing body of work optimize the candidate list by combining IR techniques with code dependency analysis.However,these approaches are sensitive to the correctness of the candidate links and they offer no help or even make the results deteriorate with the incorrect links.Recent work focused on utilizing user feedback to increase the accuracy of IR-based approaches.However,the users need to verify most of candidate list.This is infeasible in practice to improve IR-based approaches.Based on the analysis of the above related work,we have formed the following important research ideas to generate a high performance candidate traceability list:(1)Based on closeness measure,we are able to build separate sets of code classes that closely relate each other based on their code dependencies.We suggest that each set(named as code region)implicitly represents at least one aspect of the system functionalities.(2)We try to avoid the pollution of candidate list by making the candidate links verified through user feedback prior to adjust the similarity of links with code dependency analysis.(3)We employ different optimization strategies for code elements which are inside and outside region.Eventually,all links are re-ranked according to the combined information of IR values,user feedback,and our closeness analysis.In summary,the contribution of this paper is summarized as follows:1.We proposed an IR-based approach combining user feedback with closeness analysis on code dependencies.On the one hand,we build candidate regions through setting closeness threshold.On the other hand,for a given requirement,we choose the class that has the highest IR value in each region as the representative class.Then we ask user to iteratively verify these representative classes for each region and adjust the similarity of relevant candidate link base on user feedback.2.We evaluated the above traceability recovery approach on four different case studies.We have validated the effectiveness and practicality of our approach with a high-quality data set which is widely used in the domain for traceability validation and three open source systems that are widely used in everyday practice.And we organized the traceability between requirement and code through analyzing behavioral information of the open source software on the issue-tracking tool.We obtained the code dependencies required by our method by running the test cases which are used to verify system functionality of software system self.3.In order to apply our method to daily practice,we also developed an assistant tool for traceability recovery between requirement and code.In addition,we integrate the above approach into it.The evaluation also showed that our approach statistically outperforms other baseline approaches through a small amount of user feedback.And we can use this approach in daily practice through the assistant tool for traceability recovery. |
PDF Full Text Request