Aggregating User-Centric and Post-Centric Sentiments from Social Media for Topical Stance Prediction

Abstract Conventional opinion polls were usually conducted via questionnaires or phone interviews, which are time-consuming and error-prone. With the advances in social networking platforms, it’s easier for us to automatically collect and aggregate the overall topical stance for a specific topic. In this paper, we propose to predict topical stances by aggregating user-centric and post-centric sentiments from social media. Firstly, related posts of a given topic are collected from social media and clustered by word embeddings, where major keywords are extracted as the expanded concepts. Then, machine learning methods are used to train sentiment lexicon with word embeddings. Finally, the sentiment scores from user-centric and post-centric views are aggregated as the total stance on the topic. In the experiments on data from online forums, the proposed approach can obtain the best performance with a mean absolute error (MAE) of 0.52% for stance prediction of 2016 Taiwan Presidential Election. This shows the effectiveness of our proposed approach in topical stance aggregation and prediction. Further investigation is needed to evaluate the performance of the proposed method in larger scales. Keywords: Topical stance detection, Sentiment analysis, Word embeddings, Document clustering 1 Introduction People usually express their opinions in social occasions with friends and to the public. To know what the general public think about a specific topic, it usually takes much human efforts in designing questionnaires, collecting feedbacks and analyzing them. It’s time-consuming and error prone. Depending on the participation of people, there could be not too many effective responses. With the advances of social networking platforms, it’s very easy to post articles and reply with comments. For example, Twitter, Facebook, and Instagram are among the most popular social networking sites with different functions. This facilitates users to make online discussions in an immediate way. Given huge amount of social opinions, it would be useful if we can automatically collect and aggregate the general stances from them. There are some challenges to the problem. Firstly, given the very diverse contents in social media, it would be difficult to obtain the most relevant contents from huge amount of data. Secondly, people might express their opinions in different ways. It would be difficult to extract what they really think about specific topics from very short texts in social media. Content in a short text is usually limited in scope. Without explaining the ideas and referencing related documents, we might only obtain fragmented terms or named entities just from the sole content of a single post. It might even contain emotional feelings or noises that cannot help us clarify the main idea. On the other hand, users have different types of activities in addition to posting. For example, most social networking platforms provide mechanisms for making friends, following people or topics that you are interested in, and expressing agreement or disagreement, replying, or commenting on others’ posts. These social relations, both explicit and implicit, provide useful clues for understanding what people really think, in addition to what they explicitly mention in post contents. This makes it …

Content truncated to 500 words