基于用户浏览日志的上下文相关新闻推荐

个性化新闻推荐的难点在于用户在浏览新闻时目的性不强，容易受各种环境因素的影响，导致其浏览行为难以预测。以往的研究仅仅强调推荐内容相关的或者和用户长期兴趣相匹配的新闻，忽视了环境因素的影响。为此，需要研究上下文相关的新闻推荐算法。具体做法是从用户的浏览日志中提取上下文相关特征，然后训练一个Logistic回归模型来预测用户接下来最可能阅读的新闻。真实数据上的实验结果表明，上下文相关新闻推荐方法效果明显好于传统方法，也验证了上下文信息对用户浏览行为的重要影响。

Context-awareness recommendation based on user browsing log

The challenge of personalized news recommendation lies in that users browse news without clear purposes. Thus they are susceptible to various environmental factors, making their browsing behavior hard to predict. Previous studies focus on recommending news based on the content similarity or users’ long-term interests, which ignores the impact of environmental factors. Therefore, there is a need to study context-aware news recommendation algorithm. Specifically, contextual features are extracted from the user’s browser log, and then are used to train a Logistic regression classifier to predict the most possible news users are going to read. Experimental results on real-world data set suggest that the context-aware method outperforms substantially traditional methods, and verify the important impact of contextual information on users’ browsing behavior.