One-class classification models for financial industry information recommendation
Jun Xu, Qingcai Chen, Xiaolong Wang, Zhongyu Wei · 2010
In finance domain, the acquisition of in-time and comprehensive intra-industry information is important for decision-makers and stock investors to maximize their investment profits. But there are following problems in the retrieval and recommendation of financial industry information. (1) Unlike the concrete conceptions, industry could not be perfectly delineated with keywords. (2) It's difficult to calculate the relevance between document and industry. (3)The massive search results confused the user as a result of the information overload. In this paper, this problem is treated as a classification of relevance. The one-class classification model is adopted to calculate the relevance between document and industry since the lack of well sampled non-relevant documents. Based on selected industry-specific description terms, three different one-class classifiers k-means, one-class SVM and language model algorithm are trained with only relevant (positive) documents to help making recommendation decisions. The experimental results show that the proposed methods perform well with high micro-average F1 and macro-average F1 both up to the 80%. We also perform experiments to verify the relationship between parameters and performance.