Best feature performance in codeswitched hate speech texts

arXiv (Cornell University), 2021

The pervasiveness of the Internet and social media have enabled the rapid and anonymous spread of Hate Speech content on microblogging platforms such as Twitter. Current EU and US legislation against hateful language, in conjunction with the large amount of data produced in these platforms has led to automatic tools being a necessary component of the Hate Speech detection task and pipeline. In this study, we examine the performance of several, diverse text representation techniques paired with multiple classification algorithms, on the automatic Hate Speech detection and abusive language discrimination task. We perform an experimental evaluation on binary and multiclass datasets, paired with significance testing. Our results show that simple hate-keyword frequency features (BoW) work best, followed by pretrained word embeddings (GLoVe) as well as N-gram graphs (NGGs): a graph-based representation which proved to produce efficient, very lowdimensional but rich features for this task. A combination of these representations paired with Logistic Regression or 3-layer neural network classifiers achieved the best detection performance, in terms of micro and macro F-measure.

Lecture Notes in Computer Science, 2021

The pervasiveness of the Internet and social media have enabled the rapid and anonymous spread of Hate Speech content on microblogging platforms such as Twitter. Current EU and US legislation against hateful language, in conjunction with the large amount of data produced in these platforms has led to automatic tools being a necessary component of the Hate Speech detection task and pipeline. In this study, we examine the performance of several, diverse text representation techniques paired with multiple classification algorithms, on the automatic Hate Speech detection and abusive language discrimination task. We perform an experimental evaluation on binary and multiclass datasets, paired with significance testing. Our results show that simple hate-keyword frequency features (BoW) work best, followed by pretrained word embeddings (GLoVe) as well as N-gram graphs (NGGs): a graph-based representation which proved to produce efficient, very lowdimensional but rich features for this task. A combination of these representations paired with Logistic Regression or 3-layer neural network classifiers achieved the best detection performance, in terms of micro and macro F-measure.

MIJST, 2024

Currently social media and online platforms have become a major source of cyberbullying and hate speech. It is currently affecting people and communities in harmful ways. Hate speech on social media is rising in Bangladesh and it is creating a need for effective tools to prevent and detect these incidents. This study introduces a deep learning model to mitigate this issue of identifying hate speech in text using three types of word embedding methods: Word2Vec, FastText, and BERT. The text data was labeled to mark hate speech and nonhate speech content. After that, these texts are preprocessed by removing punctuation and symbols to help improve model accuracy. Five deep learning models Bi-GRU-LSTM-CNN, Bi-LSTM, CNN, LSTM, and XGBoost were trained to classify the text as hate speech or non-hate speech. The study found that the LSTM model accomplished the highest accuracy at 95.66% with the Word2Vec embedding method, while CNN reached 87.70% with FastText embeddings. Word2Vec is effective for capturing word meanings in general text classification. FastText works well with rare words and languages that have complex word forms. These findings help advance effective hate speech detection techniques. It could promote more respectful and inclusive interactions on social media. This proposed deep-learning model can help stop cyberbullying and hate speech on social media.

International Journal for Research in Applied Science & Engineering Technology (IJRASET), 2022

With the increased popularity of social media sites like Twitter and Instagram over the years, it has become easier for users of the sites to remain anonymous while taking part in hate speech against various peoples and communities. As a result, in an effort to curb such hate speech online, detection of the same has gained a lot more attention of late. Since curbing the growing amount of hate speech online by manual methods is not feasible, detection and control via Natural Language Processing and Deep Learning methods has gained popularity. In this paper, we evaluate the performance of a sequential model with the Universal Sentence Encoder against the RoBERTa method on different datasets for hate speech detection. The result of this study has shown a greater performance overall from using a Sequential model with a multilingual USE layer.

International Journal of Information Technology and Computer Science

Presidential campaign periods are a major trigger event for hate speech on social media in almost every country. A systematic review of previous studies indicates inadequate publicly available annotated datasets and hardly any evidence of theoretical underpinning for the annotation schemes used for hate speech identification. This situation stifles the development of empirically useful data for research, especially in supervised machine learning. This paper describes the methodology that was used to develop a multidimensional hate speech framework based on the duplex theory of hate [1] components that include distance, passion, commitment to hate, and hate as a story. Subsequently, an annotation scheme based on the framework was used to annotate a random sample of ~51k tweets from ~400k tweets that were collected during the August and October 2017 presidential campaign period in Kenya. This resulted in a goldstandard codeswitched dataset that could be used for comparative and empiri...

IEEE Access

Automatic hate speech identification in unstructured Twitter is significantly more difficult to analyze, posing a significant challenge. Existing models heavily depend on feature engineering, which increases the time complexity of detecting hate speech. This work aims to classify and detect hate speech using a linguistic pattern-based approach as pre-trained transformer language models. As a result, a novel Pattern-based Deep Hate Speech (PDHS) detection model was proposed to detect the presence of hate speech using a cross-attention encoder with a dual-level attention mechanism. Instead of concatenating the features, our model computes dot product attention for better representation by reducing the irrelevant features. The first level of Attention is extracting aspect terms using predefined parts-of-speech tagging. The second level of Attention is extracting the sentiment polarity to form a pattern. Our proposed model trains the extracted patterns with term frequency, parts-of-speech tag, and Sentiment Scores. The experimental results on Twitter Dataset can learn effective features to enhance the performance with minimum training time and attained 88%F1Score.

Bulletin of Electrical Engineering and Informatics

Social media has become a vital part of most modern human personal life. Twitter is one of the social media that was formed from the development of communication technology. A lot of social media gives users the freedom to express themselves. This facility is misused by users, so hate speech is spread. Designing a system to detect hate speech intelligently is needed. This study uses the hybrid deep learning (HDL) and solo deep learning (SDL) approach with the convolutional neural networks (CNN) and bidirectional gated recurrent unit (Bi-GRU) algorithm. There are 4 models built, namely CNN, Bi-GRU, CNN+Bi-GRU, and Bi-GRU+CNN. Term frequency-inverse document frequency (TF-IDF) is used for feature extraction, which is to get linguistic features to be analyzed and studied. FastText is used to perform feature expansion to minimize mismatched vocabulary. Four scenarios are run. CNN with an accuracy of 87.63%, Bi-GRU produces an accuracy of 87.46%, CNN+Bi-GRU provides an accuracy of 87.47% and Bi-GRU+CNN provides an accuracy of 87.34%. The ability of this approach to understand the context is qualified. HDL outperforms SDL in terms of n-gram type, where HDL can understand sentences broken down by hybrid n-gram types, namely Unigram-Bigram-Trigram which is a complex n-gram hybrid.

SN Computer Science

This paper presents online hate speech as a societal and computational challenge. Offensive content detection in social media is considered as a multilingual, multi-level, multi-class classification problem for three Indo-European languages. This research problem is offered to the community through the HASOC shared task. HASOC intends to stimulate research and development in hate speech recognition across different languages. Three datasets (in English, German, and Hindi) were developed from Twitter and Facebook, and made available. This paper describes the creation of the multilingual datasets and the annotation method. We will present the numerous approaches based on traditional classifiers, deep neural models, and transfer learning models, along with features used for the classification. Results show that the best classifier for the binary classification might not perform best in the multi-class classification, and the performance of the same classifier varies across the languages. Overall, transfer learning models such as BERT, and deep neural models based on LSTMs and CNNs perform similar but better than traditional classifiers such as SVM. We will conclude the discussion with a list of issues that needs to be addressed for future datasets.

Proceedings of the First Workshop on Abusive Language Online

The paper introduces a deep learningbased Twitter hate-speech text classification system. The classifier assigns each tweet to one of four predefined categories: racism, sexism, both (racism and sexism) and non-hate-speech. Four Convolutional Neural Network models were trained on resp. character 4-grams, word vectors based on semantic information built using word2vec, randomly generated word vectors, and word vectors combined with character n-grams. The feature set was down-sized in the networks by maxpooling, and a softmax function used to classify tweets. Tested by 10-fold crossvalidation, the model based on word2vec embeddings performed best, with higher precision than recall, and a 78.3% F-score.

2019

Hate speeches are defined as utterances, writings, actions, or performances that are intended to incite violence or prejudice against a person on the basis of the characteristics of a particular group that he or she is representing, such as race, ethnicity. In this study, we built a hate speech classification model using word representation with continous bag of words (CBOW) and fastText algorithm. This algorithms was chosen, because it is able to achieve a good performance, specially in the case of rare words by making use of character level information. Based on this result, we can see that there is no single, universal variations that outperform other. But in general, models that use pre-trained vectors from Wiki outperform models that do not use pre-trained vectors.