Obtaining Answers from Social Media Data
Alon Y. Halevy · 2021
Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. Content posted on social media is a rich set of data from which we would like to answer a broad set of questions. As one important example, keeping users safe on social networks requires that we identify content that contains misinformation or hateful speech and remove it. Similarly, users may be interested in getting additional value from their network such as finding friends who recently travelled to a particular destination or finding out which movies their friends are discussing. The typical machine learning approach in which we develop a model for every question we want to ask works well for questions we ask constantly (e.g., find hate speech), but not for ad-hoc questions that occur infrequently. In this talk I will advocate for an approach that combines the benefits of machine learning and database-style query answering. I will illustrate this approach through the idea of Neural Databases, a new kind of database system that leverages the strength of pre-trained language models to answer database queries over text and other modalities.