Research On Multi-Label Classification Algorithm Based On Co-Occurrence Relation

The Classification problem has been a hot topic in data mining. Traditionally, the classification task is based on the assumption that each example only corresponds to a class label. With the growing development of Internet technology, the forms of data we can get are more and more complex, and the data example can not be definded with a only single label accurately. As a result, it generated a multi-label classification problem. In this problem, the goal of learning is to assign a number of appropriate labels to unknown instances.From above studies, we can see that multiple labels are correlated in multi-label learning. Exploiting these correlations effectively not only can improve the performance of classification, but also can learn from rare labels and scale up to large number of labels From the perspective of the lables correlations,We studies the multi-label algorithm by exploiting these correlations to acheive better performance. The main research contents are as follows:1. We first introduce the concepts of multi-label learning in detail, and then describe the current research results about label correlations, it can provide a theoretical support and reference for the following research in multi-label classification algorithm based on the label correlations.2.We summarize the Rakel algorithm in detail, and the analysis shows that the Rakel algorithm has a lack of correlation between the labels in the labels selection process, which influences the performance of the algorithm. In this paper, we propose an improved algorithm based on label correlation to improve the performance.The proposed method constructs co-occurrence Matrixthe between labels, which can find pairwise relationships of labels. Experiments demonstrates that it can improve the performance of the alrorithm effectively.3.In this paper we designs a financial product recommendation system based on PwRakel algorithm. The system first grabs the new financial product information for data processing from the Internet using the crawler algorithm, we predict the labels of the financial products using the proposed classification method. Finally, it can recommend appropriate financial products for diferent users according to their needs.