This paper presents the findings of the ICON-2021 shared task on Multilingual Gender Biased and Communal Language Identification, which aims to identify aggression, gender bias, and communal bias in data presented in four languages: Meitei, Bangla, Hindi and English. The participants were presented the option of approaching the task as three separate classification tasks or a multi-label classification task or a structured classification task. If approached as three separate classification tasks, the task includes three sub-tasks: aggression identification (sub-task A), gender bias identification (sub-task B), and communal bias identification (sub-task C). For this task, the participating teams were provided with a total dataset of approximately 12,000, with 3,000 comments across each of the four languages, sourced from popular social media sites such as YouTube, Twitter, Facebook and Telegram and the the three labels presented as a single tuple. For the test systems, approximately 1,000 comments were provided in each language for every sub-task. We attracted a total of 54 registrations in the task, out of which 11 teams submitted their test runs. The best system obtained an overall instance-F1 of 0.371 in the multilingual test set (it was simply a combined test set of the instances in each individual language). In the individual sub-tasks, the best micro f1 scores are 0.539, 0.767 and 0.834 respectively for each of the sub-task A, B and C. The best overall, averaged micro f1 is 0.713. The results show that while systems have managed to perform reasonably well in individual sub-tasks, especially gender bias and communal bias tasks, it is substantially more difficult to do a 3-class classification of aggression level and even more difficult to build a system that correctly classifies everything right. It is only in slightly over 1/3 of the instances that most of the systems predicted the correct class across the board, despite the fact that there was a significant overlap across the three sub-tasks.

The ComMA@ICON 2021 Shared Task involved identifying the level of aggression and identifying gender bias and communal bias from texts in various languages from the domain of social media. In this paper, we present the description and analyses of systems we implemented towards these tasks. We built systems utilizing Transformer-based models, experimented by individually and jointly modelling these tasks, and investigated the performance of a feature engineering method in conjunction with a joint modelling approach. We demonstrate that the joint modelling approaches outperform the individual modelling approach in most cases.

Due to the exponential increasing reach of social media, it is essential to focus on its negative aspects as it can potentially divide society and incite people into violence. In this paper, we present our system description of work on the shared task ComMA@ICON, where we have to classify how aggressive the sentence is and if the sentence is gender-biased or communal biased. These three could be the primary reasons to cause significant problems in society. Our approach utilizes different pretrained models with Attention and mean pooling methods. We were able to get Rank 1 with 0.253 Instance F1 score on Bengali, Rank 2 with 0.323 Instance F1 score on multilingual set, Rank 4 with 0.129 Instance F1 score on meitei and Rank 5 with 0.336 Instance F1 score on Hindi. The source code and the pretrained models of this work can be found here.

Aggressive and hate-filled messages are prevalent on the internet more than ever. These messages are being targeted against a person or an event online and making the internet a more hostile environment. Since this issue is widespread across many users and is not only limited to one language, there is a need for automated models with multilingual capabilities to detect such hostile messages on the online platform. In this paper, the performance of our classifiers is described in the Shared Task on Multilingual Gender Biased and Communal Language Identification at ICON 2021. Our team “Beware Haters” took part in Hindi, Bengali, Meitei, and Multilingual tasks. Our team used various models like Random Forest, Logistic Regression, Bidirectional Long Short Term Memory, and an ensemble model. Model interpretation tool LIME was used before integrating the models. The instance F1 score of our best performing models for Hindi, Bengali, Meitei, and Multilingual tasks are 0.289, 0.292, 0.322, and 0.294 respectively.

This paper presents our system description on participation in ICON-2021 Shared Task sub-task 1 on multilingual gender-biased and communal language identification as team name: DELab@IIITSM. We have participated in two language-specific Meitei, Hindi, and one multi-lingualMeitei, Hindi, and Bangla with English code-mixed languages identification task. Our method includes well design pre-processing phase based on the dataset, the frequency-based feature extraction technique TF-IDF which creates the feature vector for each instance using(Decision Tree). We obtained weights are 0.629, 0.625, and 0.632 as the overall micro F1 score for the Hindi, Meitei, and multilingual datasets.

This work aims to evaluate the ability that both probabilistic and state-of-the-art vector space modeling (VSM) methods provide to well known machine learning algorithms to identify social network documents to be classified as aggressive, gender biased or communally charged. To this end, an exploratory stage was performed first in order to find relevant settings to test, i.e. by using training and development samples, we trained multiple algorithms using multiple vector space modeling and probabilistic methods and discarded the less informative configurations. These systems were submitted to the competition of the ComMA@ICON’21 Workshop on Multilingual Gender Biased and Communal Language Identification.

The proliferation in Social Networking has increased offensive language, aggression, and hate-speech detection, which has drawn the focus of the NLP community. However, people’s difference in perception makes it difficult to distinguish between acceptable content and aggressive/hateful content, thus making it harder to create an automated system. In this paper, we propose multi-class classification techniques to identify aggressive and offensive language used online. Two main approaches have been developed for the classification of data into aggressive, gender-biased, and communally charged. The first approach is an ensemble-based model comprising of XG-Boost, LightGBM, and Naive Bayes applied on vectorized English data. The data used was obtained using an Indic Transliteration on the original data comprising of Meitei, Bangla, Hindi, and English language. The second approach is a BERT-based architecture used to detect misogyny and aggression. The proposed model employs IndicBERT Embeddings to define contextual understanding. The results of the models are validated on the ComMA v 0.2 dataset.

In today’s world, online activity and social media are facing an upsurge of cases of aggression, gender-biased comments and communal hate. In this shared task, we used a CNN-LSTM hybrid method to detect aggression, misogynistic and communally charged content in social media texts. First, we employ text cleaning and convert the text into word embeddings. Next we proceed to our CNN-LSTM based model to predict the nature of the text. Our model achieves 0.288, 0.279, 0.294 and 0.335 Overall Micro F1 Scores in multilingual, Meitei, Bengali and Hindi datasets, respectively, on the 3 prediction labels.

Social media analytics are widely being explored by researchers for various applications. Prominent among them are identifying and blocking abusive contents especially targeting individuals and communities, for various reasons. The increasing abusive contents and the increasing number of users on social media demands automated tools to detect and filter the abusive contents as it is highly impossible to handle this manually. To address the challenges of detecting abusive contents, this paper describes the approaches proposed by our team MUCIC for Multilingual Gender Biased and Communal Language Identification shared task (ComMA@ICON) at International Conference on Natural Language Processing (ICON) 2021. This shared task dataset consists of code-mixed multi-script texts in Meitei, Bangla, Hindi as well as in Multilingual (a combination of Meitei, Bangla, Hindi, and English). The shared task is modeled as a multi-label Text Classification (TC) task combining word and char n-grams with vectors obtained from Multilingual Sentence Encoder (MSE) to train the Machine Learning (ML) classifiers using Pre-aggregation and Post-aggregation of labels. These approaches obtained the highest performance in the shared task for Meitei, Bangla, and Multilingual texts with instance-F1 scores of 0.350, 0.412, and 0.380 respectively using Pre-aggregation of labels.

Due to the rapid rise of social networks and micro-blogging websites, communication between people from different religion, caste, creed, cultural and psychological backgrounds has become more direct leading to the increase in cyber conflicts between people. This in turn has given rise to more and more hate speech and usage of abusive words to the point that it has become a serious problem creating negative impacts on the society. As a result, it is imperative to identify and filter such content on social media to prevent its further spread and the damage it is going to cause. Further, filtering such huge data requires automated tools since doing it manually is labor intensive and error prone. Added to this is the complex code-mixed and multi-scripted nature of social media text. To address the challenges of abusive content detection on social media, in this paper, we, team MUM, propose Machine Learning (ML) and Deep Learning (DL) models submitted to Multilingual Gender Biased and Communal Language Identification (ComMA@ICON) shared task at International Conference on Natural Language Processing (ICON) 2021. Word uni-grams, char n-grams, and emoji vectors are combined as features to train a ML Elastic-net regression model and multi-lingual Bidirectional Encoder Representations from Transformers (mBERT) is fine-tuned for a DL model. Out of the two, fine-tuned mBERT model performed better with an instance-F1 score of 0.326, 0.390, 0.343, 0.359 for Meitei, Bangla, Hindi, Multilingual texts respectively.

This paper presents the system that has been submitted to the multilingual gender biased and communal language identification shared task by BFCAI team. The proposed model used Support Vector Machines (SVMs) as a classification algorithm. The features have been extracted using TF/IDF model with unigram and bigram. The proposed model is very simple and there are no external resources are needed to build the model.