In this paper we detailed a multinomial classification-based methodology that combines different algorithms (SVM and MLP) with document representations (Tf Idf vectorization and Doc2vec embedding) and: (i) can distinguish between crime-related news and not-crime related news and; (ii) allows the assignment of each crime-related news to its corresponding crime type. With a F1-score of 84% achieved by the MLP with Doc2vec approach, it can be concluded that it is possible to answer the question of how the crimes are committed (what types of crime are perpetrated) and, in this way, offer a thermometer to citizens about criminal activity in a given territory, as reported by news articles.
Alatrista-Salas, H., Morzán-Samamé, J., & Nunez-del-Prado, M. (2020). Crime alert! crime typification in news based on text mining. In Lecture Notes in Networks and Systems (Vol. 69, pp. 725–741). Springer. https://doi.org/10.1007/978-3-030-12388-8_50