Deep Artificial Intelligence for Fantasy Football Language Understanding

Fantasy sports allow fans to manage a team of their favorite athletes and compete with friends and fellow managers. The fantasy platform aligns the real-world statistical performance of athletes to fantasy scoring and has steadily risen in popularity to an estimated 9.1 million players per month spending a total of 7.7 billion-million minutes with 4.4 billion player card views on the ESPN Fantasy Football platform from 2018-2019. In parallel, the sports media community produces news stories, blogs, forum posts, tweets, videos, podcasts and opinion pieces that are both within and outside the context of fantasy sports. However, human fantasy football players cannot consume and summarize billions of bytes of natural language text and multimedia data to make a roster decision. Before our system, fantasy managers relied on expert projections and their analysis of, on average, 3.9 sources of information to make roster decisions. While these experts excel at evaluating players based on traditional statistics, they are omitting the majority of data that would inform their assessments. Our work discusses and shows the results of a novel (patent pending) machine learning pipeline to effectively manage an ESPN Fantasy Football team. The use of trained statistical entity detectors and document2vector models applied to over 50,000 news sources and 2.3 million articles, videos and podcasts each day enables the system to comprehend natural language with an analogy test accuracy of 100% and keyword test accuracy of 80%. Next, deep learning feedforward neural networks that are 98 layers deep provide player classifications such as if a player will be a bust, boom, play with a hidden injury or play meaningful touches with a cumulative 72% accuracy with a 12% real world distribution. Finally, a multiple regression ensemble accepts the deep learning output and ESPN projection data to provide a point projection for each of the top 500+ fantasy football players in 2018. The point projection maintained a Root Mean Squared Error of 6.78 points. Next, the best out of the 24 probability density functions that were fit to the current projection and historical scores was selected to visualize score spreads. Within the first 6 weeks of the product launch, the total number of users spent a cumulative time of over 4.6 years viewing our AI insights. As a result, each user spent over 90 seconds using the evidence from our novel algorithms. The training data for our models was provided by a 2015 and 2016 web archive from Webhose, ESPN statistics, and Rotowire injury reports. We used 2017 fantasy football data as a test set.