Recommending Temporally Relevant News Content Using Implicit Feedback Data: A Tag-Based Approach

Abstract

News has in this day and age, transformed primarily into a digital format with leading newspapers and news agencies having a significant online presence. The speed at which news reaches the reader notwithstanding, the proliferation of blogs and microblogs to deliver specialized content has become the order of the day. Although a large amount of news content is being generated each day, only a subset of content is relevant to a user’s interests. Out of this subset of relevant content, the content that the user is likely to engage with is considerably smaller. With the generation and proliferation of such a large amount of news content, it becomes hard for a user to sift through it all and find the news articles that they are likely interested in. Even highly engaged users tend to dis-engage with a website when the content they are served is unappealing to them. This deluge of information has given rise to a challenge known as Information Overload. Previous research has shown that variations in amount of information presented to the user, has an impact on their decision making process and in turn has an effect on decision quality. [1]. Previous works like [2],[3],[4] and [5] analyze the effects of decision making by consumers in the face of information overload during the purchase of different products like CD players, peanut butter, dinner and other small and large purchases. In the digital news media domain, recommendation systems have been used to ensure delivery of content to the user in tune with their tastes. However, these systems face an unprecedented challenge - the transient nature of popular news and users changing interests. Recommendation systems have been popularized mostly by e-commerce websites like Amazon.com, Netflix, Yelp and other such retailers. Recommendations in the e-commerce context however has in most cases, explicit feedback data of customer ratings for a particular product or comments about the product by customers, pre or post purchase. This explicit feedback data provides the recommendation engine with some notion of user likes and dislikes. The abscence of such explicit feedback by users compounds the recommendation problem in the digital news media domain. Most recommendation systems for recommending digital news content rely on inferring user engagement through clicks; which is not necessarily an accurate measure as it gives us no explicit information about the degree to which a user is interested in a news article. Although there is a gross lack of explicit information about user interests, we find that some other factors strongly influence user engagement in the news media domain. These additional factors and their combined effects on recommendation quality and user-engagement are the subject of this thesis. A study about the behavior of temporal and tag-based models for news article recommendation is carried out. Our experiments indicate that incorporating temporal and user-tag information improves recommendation quality and increases user engagement. We argue through experimental evaluation that the improved performance is due to recommendation of more personalized news content by the tag-based recommendation algorithms as compared to other models that do not explicitly incorporate user-tag information.