Accurately extracting content from Web news is a key technology for quality improvement in Web news analysis and applications.Due to the lack of publication standards, differences in publishing formats, and a highly heterogeneous big data carrier of the Web itself, Web news extraction has become an open research problem.Extensive case studies by this research indicate that there is potential relevance between Web content layouts and their tag paths.Inspired by this observation, this paper designs a series of tag path extraction features to distinguish the Web content and noise from different perspectives.Based on the similarity analysis of these features, the paper proposes a features fusion strategy with group feature selection, and provides a Web news extraction method via feature fusion, CEPF.CEPF is a fast, universal, no-training and online Web news extraction algorithm.It can extract Web news pages across multi-resources, multi-styles, and multi-languages.Experimental results with public data sets such as CleanEval show that the CEPF method achieves better performance than the state-of-the-art CETR method.