How we use machine learning to enhance our user's experience
ML has been used on Headlines Today for a long time to provide a better, more personalised reading experience. We previously asked users what topics they preferred during the onboarding process. We now allow users to subscribe to topics. As a result, the results for topic queries must be far more accurate. Personalisation allows for mistakes, but if a technology article appears in the sports category, this erodes user trust in the system. Our classification models are now far more accurate, providing a reliable experience to our users.
We also have started to extract keywords and named entities from articles. We hope to add an experience in which users can subscribe to dynamic and static topics. The news is constantly changing. A user may wish to see more news about Donald Trump or Brexit. In 50 years, however, these topics may no longer exist. We are creating a system which will allow a user to subscribe to a dynamic topic like Donald Trump. Once there is no longer any news on this particular topic, the system will automatically unsubscribe the user from the topic. If the topic picks up again, the system will resubscribe to the topic. This will soon be added to a live build of Headlines Today
We are also using ML for utility purposes. For example, to prevent users from downloading too many articles (which contain photos and text), we developed a model which predicts how many articles a user will read, based on the frequency of news update. This, of course, assumes that a major event results in a higher frequency of news update.
We are looking to provide our users with an even more intelligent experience. Read more on Blog/tag/ml to see how we are doing that.