Crowdsourcing Workers Recommendation Algorithm Based on Improved Deep Matrix Factorization

In the current research of recommending the workers to the demanders, most scholars only model and analyze task information and workers preferences while ignoring demander preferences. To solve this problem, this paper proposed an improved deep matrix factorization algorithm considering both workers preference and demanders preference, and then applied it to the workers recommendation of the crowdsourcing platform. First, this algorithm constructs user preferences based on tags, using ratings of tags as the initial weights. Second, this algorithm uses Word2vec to get tag vectors, multiply it with the weight, and then average all tag vectors to get the user vector. Third, this algorithm uses the cosine similarity to calculate the user similarity matrix. Last, the user similarity matrix is incorporated into the deep matrix factorization algorithm as auxiliary information. The experimental dataset comes from the Zhubajie.com. The results show that this algorithm is better than the baseline models in both MAE and RMSE.