Codewords Detection in Microblogs Focusing on Differences in Word Use Between Two Corpora

Abstract

In recent years, the number of drug trafficking using microblogs has been increasing, which has become a social problem. While cyber patrols have been conducted to crack down on such crimes, those who post crime-inducing messages use terms that camouflage their criminal intentions so-called “codewords” to avoid keywords such as “enjo kosai,” “marijuana,” and “methamphetamine” that may be monitored and attract police attention. These codewords change once they become popular, so it is always necessary to keep track of the latest codewords. Therefore, we propose a new method for detecting the latest codewords. In this paper, we offer a new way of detecting code words from the differences in the words used in posts to detect codewords used in a crime. Specifically, we propose a new method in which we divide words into two corpora, depending on whether a post containing a word has a criminal intention and detect codewords from the differences between similar words of the same word between two corpora. To confirm the effectiveness of the proposed method, we conducted an experiment to detect codewords. The experimental results showed that the proposed method was able to detect codewords with an accuracy of 0.56 persentages points higher than that of the baseline method. The experiment shows that the proposed method can reduce the burden of continuously monitoring code words by rapidly and automatically detecting new codewords that change with time; thus, it provides the possibility of showing clues for crimes.