In modern online publishing, user comments are an integral part of any media platform. Between the high volume of generated comments and the need for moderation of inappropriate content, human approval becomes a serious bottleneck with negative consequences for both operating cost and user experience.
To alleviate this problem we present a text classiﬁcation model for automatic approval of user comments on text articles. With multiple textual input from both the comment in question and the host article, the model uses a neural network with multiple encoders. Diﬀerent choices for encoder networks and combination methods for encoder outputs are investigated. The system is evaluated on news articles from a leading Vietnamese online media provider, and is currently on a test run with said newspaper.
See more HERE.
Author: Vu Dang
FPT Technology Research Institute, FPT University, Hanoi, Vietnam