Research on Mining Method Based on Multi-task Iterative Learning

In real life, people's ability to obtain information is increasing, and they often need to deal with data from different source domains. There are differences in the representation, distribution, scale and density of these data. How to learn these multi-source data has become a hot research topic at present. This research point proposes a traffic structure mining method based on multi-task iterative learning. Taking the four key factors affecting traffic travel as characteristic attributes, this paper analyzes the law of their effects on travel behavior. In the learning process, the learner selects unlabeled samples with high confidence to mark each other, which can be used to update the model. After analyzing the shape of the weight matrix of road sections representing extra cost in the superposition model, we find that the model in this paper can effectively restore the peak traffic hours, which is in line with the actual scene.