Classification of Illegal Drug Sales Posts using Clustering-Based Topic Modeling.

Abstract

Drugs illegally traded online are causing social problems around the world wide. One of the ways to solve this problem is to automatically delete sales posts quickly even if they are uploaded. We propose new data on illegal drug sales posts in Korean collected directly from Twitter. There are about 100K collected data, and labels were added directly to each data. Supervised learning-based models generally show high performance, but label information is essential. It is difficult to add labels to all texts in situations where a large amount of text occurs. In this work, we propose a topic modeling-based classification model that can perform higher with even a small number of labels. As a result of the experiment, higher classification performance is shown when Topic modeling is used as a small number of data.

Tasks

Reproductions