Intelligent Search Of Software Crowdsourcing Service Using User Reviews

Software crowdsourcing has attracted a lot of attention in industry and academia due to its open use of the collective mass wisdom to solve problems.In recent years,with the continuous development of software crowdsourcing platforms,the number of users has also shown a rapid growth trend.How to help users choose service providers and services they need,that is,building an excellent crowdsourcing service search system has become one of the key issues that platforms urgently need to solve.In order to solve the problems that current service search systems fail to make full use of existing platform data especially user reviews and fail to accurately rank services,this paper proposes a software crowdsourcing intelligent search model using user reviews.In one hand,by streamlining user reviews,the key information that can be used for service modeling is mined;in another hand,the pre-filtered service list is precisely ranked by the integrated learning-to-rank model method.To make up for the interpretability of search results to users,this paper proposes a template-based recommendation reason generation method.The main contributions of this paper are as follows:1)Propose a review streamlining model based on deep learning and transfer learning.This neural network method,that selects the target area first and then labels the text in the area,can effectively remove the redundant text in the reviews and extract the core content from the reviews.In order to transfer the model to service field with a relatively small amount of data,this paper proposes a model transfer learning method to build a review streamlining model suitable for the crowdsourcing service field through fine-tuning of parameters.The experimental results show that the proposed review streamlining model has improved by more than 4% on the crowdsourcing service review data set compared with previous models in evaluation indicators such as ROUGE-L.2)Propose a review-based service modeling method and an integrated learning-to-rank model.Each clause in the streamlined review text is classified according to its aspect and emotional color.The results are incorporated into the construction of service feature vectors.Aiming at the shortcomings of the single ranking model,this paper integrates multiple advanced model,which combines the advantages of the point-wise model and the pair-wise model.The experimental results show that the review-based integrated ranking model improves the performance by more than 3% compared with the single ranking model.3)Propose a template-based recommendation reason generation method.Aiming at the lack of interpretability of the service search system,recommendation reasons are generated through template matching for the top search results according to user's search request text,service information and service reviews.This provide users with the right to know,and increase the credibility of the platform.The recommendation reason generation method achieved high expert scores in terms of grammar and accuracy,4.78 points and 4.51 points respectively.