Thai Insult Detection System Based on Linguistic Features Analysis

Verbal insults often appear in online communities during textual communication between users. Current automatic prevention algorithms which employ regular expression techniques for word filtering tend to result in high false-positive errors. This paper presents an alternative method for detecting insults in Thai textual conversations based on the analysis of linguistic features. The performance of the presented algorithms was compared with the regular expression based algorithms, in terms of precision and recall scores. The results of the experiments showed that the inaccuracies in the employed third-party natural language processing procedures affected the performance of the proposed insult detection method. Once the problematic NLP procedures were improved, the proposed method outperforms regular expression based algorithms, showing lower false-positive error rate.

Keywords:

Regular Expression , Word Filter , Natural Language Processing , Online Communities , Linguistic Features