Event Detection and Knowledge Mining from Unlabelled Bengali News Articles
Iftakhar Ali Khandokar, Imtiaz Mamun, Tasmia Ishrat Alam Chadni, Zubair Ahmed Anas, Swakkhar Shatabda · 2020
Text mining is one of the most important tasks in Natural Language Processing (NLP). It is often very difficult to detect events of interest from vast unstructured or data that is not properly labeled. In this paper, we present a machine learning-based approach to discriminate particular event-related news from its text content. We focus here particularly on violence and related events. We have created a dataset from popular Bengali newspapers and used a keyword-based search method to select violence-related news articles. Based on the dataset, we first trained a supervised learning model. We have experimented with six classification algorithms where Logistic Regression outperformed each of the contemporary algorithms in both full and reduced feature set acquiring 75.7% and 83.4% accuracy respectively. Using the models, we then further analyzed violence- related news data to find out important insights from unlabeled data.