Online news platforms have attracted massive users to read digital news online. The demographic information of these users such as gender is critical for these platforms to provide personalized services such as news recommendation and targeted advertising. However, the gender information of many users in online news platforms is not available. Fortunately, male and female users usually have different pattern in reading online news. Thus, the news browsing data of users can provide useful clues for inferring their genders. In this paper, we propose a neural gender prediction approach based on the news browsing data of users. Usually a news article has different kinds of information such as title, body and categories. However, the characteristics of these components are very different, and they should be processed differently. Thus, we propose to learn unified user representations for gender prediction by incorporating different components of browsed news as different views of users. In each view, we use a hierarchical framework to first learn news representations and then learn user representations from news representations. In addition, since different words in news titles and bodies usually have different informativeness for learning news representations, we use attention mechanisms to select important words. Besides, since different news articles may also have different informativeness for gender prediction, we use news-level attentions to attend to important news articles for learning informative user representations. Extensive experiments on a real-world dataset validate the effectiveness of our approach.
CITATION STYLE
Wu, C., Wu, F., Qi, T., Huang, Y., & Xie, X. (2019). Neural Gender Prediction from News Browsing Data. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11856 LNAI, pp. 664–676). Springer. https://doi.org/10.1007/978-3-030-32381-3_53
Mendeley helps you to discover research relevant for your work.