Task Recommendation Approach Based On Reliability In Crowdsourcing Platforms

With the increasing popularity of crowdsourcing services,it has become more and more frequent that cooperating to complete specific tasks by publishing crowdsourced tasks.One of the main functions of crowdsourcing platforms is to recommend suitable tasks to workers.To ensure that assignments work well,the algorithm need to take into account the credibility of crowdsourced workers before recommending tasks for them.However,it is difficult to guarantee the source of workers’ credit.In other words,the credit evaluation of workers is inaccurate and the problem of brushing comments.Therefore,in order to improve the credibility of worker credibility,this paper conducts feature extraction and identification for this special fake reviewer,and studies a crowdsourcing task recommendation strategy based on worker credibility.Therefore,this paper proposes a solution from two steps:crowdturfing detection and crowdsourcing task recommendation.For the task of crowdturfing detection,to extract valuable feature information from numerous review data,firstly,this paper defines a series of feature indicators and classifies them.Further,in this paper,the attention mechanism is used to replace the original mean strategy,and the reviewer features are aggregated from the two dimensions of review space and social space,and the final reviewer vector representation is obtained after fusion.Finally,considering the disturbance of the sample imbalance to the detection results,the original loss function is improved,which makes the matrix or vector sensitive to the cost.In the crowdsourcing task recommendation problem,this article first build a user reputation and preference model,and builds user reputation based on high-quality reviews or ratings of workers.Second,this paper maps the crowdsourcing task recommendation scene to a bipartite graph of workers and tasks,and solves the task recommendation problem by predicting links in the bipartite graph.Finally,this paper proposes a graph autoencoder model that incorporates both worker content features and heterogeneous graph structure features.Personalized task recommendation for trusted workers is implemented.In the experimental part,it is verified that the proposed method in this paper can effectively detect the fakers and their fake comments in the comment collection which has a good detection accuracy.At the same time,experiments show that the graph model can achieve better recommendation performance when considering worker reputation.