Document Based Clustering For Detecting Events in Microblogging Websites

Social media has a great in?uence in our daily lives. People share their opinions, stories, news, and broadcast events using social media. This results in great amounts of information in social media. It is cumbersome to identify and organize the interesting events with this massive volumes of data, typically browsing, searching, monitoring events becomes more and more challenging. A lot of work has been done in the area of topic detection and tracking (TDT). Most of these methods are based on single-modality (e.g., text, images) information or multi-modality information. In the single-modality analysis, many existing methods adopt visual information (e.g., images and videos) or textual information (e.g., names, time references, locations, title, tags, and description) in isolation to model event data for event detection and tracking. This problem can be resolved by a novel multi-model social event tracking and an evolutionary framework not only effectively capturing the events, but also generates the summary of these events over time. We proposed a novel method works with mmETM, which can effectively model the social documents, which includes the long text along with the images. It learns the similarities between the textual and visual modalities to separate the visual and non-visual representative topics. To incorporate our method to social tracking, we adopted an incremental learning technique represented as mmETM, which gives informative textual and visual topics of event in social media with respect to the time. To validate our work, we used a sample data set and conducted various experiments on it. Both subjective and quantitative assessments show that the proposed mmETM technique performs positively against a few best state-of-the art techniques