Extracting Implicit Twitter Replies to News Topics

With the increase in the number of users on social networking services (SNSs) in the recent times, such services have been widely adopted to gather information. In general, users on SNSs typically use keyword searches to find information on news topics. However, multiple representative keywords are often related to any given topic. Therefore, comprehensively searching for opinions and reactions to news using only keyword search is typically difficult. In this study, we focus on a function implemented on Twitter, as a representative popular SNS, which allows users to reply to others' posts. We propose a method to discover tweets that express opinions and reactions to news topics by using posted news articles and replies as training data for a machine learning model. The model learns the relationship between news articles and replies, which do not necessarily include typical keywords. The proposed model can extract tweets that implicitly reply to news topics without directly replying to any specific post. This work contributes to the literature on social media in terms of understanding news topics and their associated discourse.