Detecting dynamic topics in social network using citation based anomaly detection

Sharing of information with their friends on social networks has become an integral part in everyone´s life. The post made by the social network user may have text, image and video. The previous approaches, [1] link anomaly detection method and [2] text-based anomaly detection method may not discover the dynamic post immediately as it concentrates on either links or text respectively, but the posts contain not only text or only links it may have images, videos or the combinations of links, text and image. The proposed model is to identify the new dynamic topics by analyzing the content of the message by extracting features of text, using [6] WordNet tool, this tool takes the words as synsets and the synonym, homonyms are identified The user may share the post by forwarding the post to their friends that may create a forward link. The normal behavior of user such as frequent forwarding friends list and the number of people on the list is considered for training model, future forwarding behavior is predicted and the anomalous behavior is detected using the training data set, from that the anomaly score is calculated. Aggregate the anomaly score from hundred users for each post, with that aggregated score and analyze the aggregated result with Sequentially Discounting Normalized Maximum-Likelihood (SDNML) coding [3] and Moving Average Convergence Divergence (MACD) burst-detection method [5] and pinpoint which post is about the dynamic topic that is discussed in a social network. In this approach, the change-point score is compared with the dynamically optimized threshold, if the score exceeds the threshold value that particular post will be pinpointed.