Natural Language Processing for Social Event Classification

In this paper, a simple but effective method for social event detection based mainly on natural language processing is introduced. Meanwhile existing approaches use many typical text-classification methods and disregard the importance of language characteristics, the proposed method exploits such language characteristics from text items in social metadata (e.g. title, description and tag) to leverage social event detection. First and foremost, we analyze the specific characteristics of natural language in social media to choose the most suitable features. Second, we employ common natural language processing techniques along with machine learning methods to extract features and perform classification. As a result, we experienced the F1 score higher than the results of related works that used state-of-the-art methods. The proposed method proves the significance of understanding language characteristics in building social event classification programs. It also offers good clues to improve existing works on social event detection.