Predicting the locations of unrest using social media
Résumé
The public often relies on social media to discuss and organize activities such as rallies and demonstrations. Monitoring and analyzing open-source social media platforms can provide insights into the locations and scales of rallies and demonstrations, and help ensure that they are peaceful and orderly.This chapter describes a dictionary-based, semi-supervised learning methodology for obtaining location information from Chinese web forums. The methodology trains a named entity recognition model using a small amount of labeled data and employs n-grams and association rule mining to validate the results. The validated data becomes the new training dataset; this step is performed iteratively to train the named entity recognition model. Experimental results demonstrate that the iteratively-trained model has much better performance than other models described in the research literature.
Domaines
Informatique [cs]Origine | Fichiers produits par l'(les) auteur(s) |
---|