Research On Cross-domain Recommendation Algorithms For Crowdsourced Testing Task Allocation

Crowdsourced testing,as a new model of Internet collaborative problem solving,has broken the barriers of high threshold,high cost and low universality of traditional software testing.Cross-domain task recommendation refers to the process of personalized task recommendation for crowdsourced testing workers by taking advantage of the characteristics of diversified fields and shared information among fields in the crowdsourced testing scenarios.Compared with the traditional task bidding,cross-domain task recommendation can provide workers with tasks that are more in line with their expectations and abilities.This paper mainly studies the cross-domain task recommendation problem of crowdsourced testing,that is,given a crowdsourced testing worker,personalized task recommendation is made for him based on data from different fields.The existing cross-domain recommendation methods are divided into feature modeling,knowledge transferring and cross-domain recommendation.In the process of feature modeling,most of the methods do not consider the auxiliary feature information of users or tasks,which leads to the problem of insufficient feature modeling.In addition,the methods used in knowledge transferring often ignore the differences between the transferred objects,which leads to negative transfer phenomenon.Moreover,there is a complementary relationship between domains,and the existing methods mostly supplement the target domain according to the source domain knowledge,but ignore the supplement of the target domain to the source domain knowledge.In order to solve the above problems,this paper studies the application of cross-domain recommendation algorithms in actual scenes based on the characteristics of crowdsourced testing scenes.The work of this paper includes:1)This paper aimed at the problem that existing cross-domain recommendation algorithms is not fully excavate the unique attribute of wokers in the crowdsourced testing scenario,proposed a multi-channel information latent factor model based crossdomain recommendation algorithm.which incorporate the attribute of wokers into the latent factor model.The experimental results show that the cross-domain recommendation algorithm based on the multi-channel information latent factor model can capture workers’ characteristics better and bring better results for recommendation.2)Existing cross-domain recommendation algorithms do not take the differences among workers into account,which leads to the introduction of unnecessary or irrelevant information in the target domain in the process of knowledge transferring.To solve this problem,this paper introduces the self-attention mechanism to make the cross-domain recommendation system more targeted on inter-domain information in the process of knowledge transferring.Subsequent experiments show that the our model can effectively alleviate the negative transfer phenomenon and improve the quality of task recommendation after introducing the self-attention mechanism.3)In crowdsourced testing scenarios,information is shared between different task domains,and each domain can complement with others.In the process of knowledge transferring,the existing cross-domain recommendation algorithms mostly transfer knowledge from source domain to target domain,without considering the knowledge supplement of target domain to source domain.Inspired by multi-task learning,this paper proposes a dual cross-domain recommendation algorithm by introducing a domain discriminant,so that the source domain and the target domain can assist each other,so as to improve the task recommendation quality of each domain simultaneously.The final experimental results show that the dual cross-domain recommendation model with the domain discriminator can alleviate the single domain information sparsity problem and effectively improve the performance of the cross-domain recommendation algorithm.