Task Recommendation Method for Online Labor Market Based on Contrastive Learning

The emergence of the online labor market has been gradual, but workers face challenges in finding tasks that align with their interests among the vast number of available tasks. Consequently, the study of task recommendation algorithms becomes crucial for the advancement of the online labor market. In this paper, we propose a method which utilizing contrastive learning in task recommendation of online labor market. We use constrastive learning to pre-train better embedding features to represent workers and tasks. These features are employed to initialize the embedding layer of popular recommendation system models, thereby exploring the model’s effectiveness further. We find that this method can improve the performance of existing recommendation models. Experimental results using an online labor market dataset indicate that our approach, which incorporates features learned through contrastive learning into PNN, WideDeep, or DeepFM recommendation models, leads to improvements in five evaluation metrics. These results demonstrate the benefits of our method in recommending tasks within the online labor market. The primary contribution of this paper is the utilization of pre-trained embeddings obtained through contrastive learning to initialize the embedding layer within popular recommendation system models.