Abstract:Every day, a huge number of information is posted on social media platforms online. In a sense, the social media platform serves as a hybrid sensor system, where people communicate information through the system just like in a sensor world (we call it Social Sensor): they observe the events, and they report it. However, the information from the social sensors (like Facebook, Twitter, Instagram) typically is in the form of multimedia (text, image, video, etc.), thus coping with such information and mining useful knowledge from it will be an increasingly difficult task. In this paper, we first crawl social video data (text and video) from social sensor Twitter for a specific social event. Then the textual, acoustic, and visual features are extracted from these data. At last, we classify the social videos into subjective and objective videos by merging these different modality affective information. Preliminary experiments show that our proposed method is able to accurately classify the social sensor data's subjectivity.

Sentiment Analysis for Social Sensor.