KMI-Coling at SemEval-2019 Task 6: Exploring N-grams for Offensive Language detection

Abstract

In this paper, we present the system description of offensive language detection tool which is developed by the KMI−Coling Group under the OffensEval Shared task. The OffensEval Shared Task was conducted in SemEval 2019 workshop. To develop the system, we have explored n-grams up to 8-gram and trained three different systems namely A, B and C system for three different sub tasks within the OffensEval task which achieves the accuracy of 79.76%, 87.91% and 44.37% respectively. The task was completed using the data set provided to us by OffensEval organisers, which was the part of OLID data set. It consists of 13,240 tweets extracted from twitter and were annotated at three levels using crowd sourcing.

Automatic Detection of Offensive Language for Urdu and Roman Urdu

“…It also endorsed the findings of [2]. Regressionbased classifier SimpleLogistic and SVM linear outperform the other models on uni-gram as concluded in [16] and achieved 94.2% F-measure value. For combined word ngrams, after uni-gram, uni+bi+tri-gram shows better performance than the other four n-grams.…”

“…Several studies show that n-gram approaches at the character and word level are very effective to detect offensive language than Bag of Word (BoW) [2], [15], [23]. [16] explored word n-grams to detect offensive language from tweets. [9], [19] employed word n-grams to detect offensive language from YouTube comments [20], [22], [24] also used word n-grams to detect offensive language from the comments collected from blogs and emails.…”

“…There are several feature extraction and classification techniques. The conventional features for offensive language detection are based on a blacklist [13], lexicon-based [14], pattern matching [15] and n-grams methods [16]. In past several studies used n-gram method for feature selection.…”

“…N-gram features are based on a sequence of characters or words in the text. Several studies reported that n-gram models outperform the other models [9], [11], [13], [16]. N-gram approach has several applications like spelling correction, next word prediction and text translation.…”

Threatening Language Detection and Target Identification in Urdu Tweets

“…Some examples of such widely known terms used to threaten are "Blood, kill, murder, death, and stab". To detect threatening language, the computational linguistics community has been focusing on online platforms like YouTube, Twitter, [25] English BoW, char n-grams SVM, LR, CNN YouTube [26] English BoW, word n-grams (2, 3, 5) SVM, NB YouTube [15] English BoW, GloVe, fastText 1D-CNN, LSTM, and BiLSTM Twitter [27] English unigram SVM, CNN, BiLSTM Twitter [28] English word n-grams (1−8) SVM Twitter [6] English Latent Dirichlet Allocation (LDA) LR Online Comments [1] English word n-grams, char n-grams NB, SVM Twitter [29] English word n-gram (3−8), char n-grams (1−3) CNN, RNN, RF, NB, SVM Twitter [30] English word n-grams SVM (linear, polynomial, radial) Twitter, Articles [31] English abusive and non-abusive word list k-means Twitter, Blogs [32] English, Portuguese hateword2vec, hatedoc2vec, unigrams NB, SVM YouTube [22] Arabic word n-gram SVM Twitter [33] Spanish word n-grams, char n-grams LR Twitter [34] Indonesian Latent Dirichlet Allocation (LDA) -Twitter, Facebook, Reddit [35] Danish, English char n-grams LR, BiLSTM Twitter [17] German Wikipedia embedding CNN Twitter [19] Italian BERT tokens AlBERTo Blogs [36] Japanese word n-grams (1-5) SVM Facebook [37] Bengla word n-grams (1-3) MNB, SVM, CNN, LSTM Twitter, Instagram [37] Turkish -MNB, SVM, DT (C4.5), KNN Facebook, Instagram, and blogs [2, 22, 24-28, 31, 36, 38-41].…”