Event detection model for Facebook news posts

Abstract

Event detection has wide application especially in the area of news streams analyzing where there is a need to monitor what events are emerging and affecting people’s lives. This is crucial for public administrations and policy makers to learn from their previous mistakes to make better decisions in the future. Different researchers have introduced several event detection models for Facebook news posts in. However, majority of these models have not provided adequate information about the discovered news events such as location, people and activity. In addition, existing models have ignored the problem of high dimensional feature space which affects the overall detection performance of the models. This research presents a conceptual event detection model for mining events from large volume of short text Facebook news posts and summarize their valuable information. This is crucial for public administrations and policy makers to learn from their previous mistakes to make better decisions in the future. The proposed model includes pre-processing, feature selection, event detection and summarization phases. The pre-processing phase involves several steps to convert unstructured text news posts into structure data. Feature selection phase to select the optimal feature subset. Meanwhile, event detection phase uses these features to construct undirected weighted graph and apply dynamic graph technique to identify the clusters from the graph and then annotate each cluster to its corresponding event. At the end of this paper, several unresolved problems in the construction of event detection model from Facebook news posts are reported to be used as future work for the current study.