This paper presents the overview of the shared task on emotional analysis in Tamil. The result of the shared task is presented at the workshop. This paper presents the dataset used in the shared task, task description, and the methodology used by the participants and the evaluation results of the submission. This task is organized as two Tasks. Task A is carried with 11 emotions annotated data for social media comments in Tamil and Task B is organized with 31 fine-grained emotion annotated data for social media comments in Tamil. For conducting experiments, training and development datasets were provided to the participants and results are evaluated for the unseen data. Totally we have received around 24 submissions from 13 teams. For evaluating the models, Precision, Recall, micro average metrics are used.

/pdf/findings-of-the-shared-task-on-emotion-analysis-in-tamil-2vbzofvm.pdf

Findings of the Shared Task on Emotion Analysis in Tamil

Due to the rapid development of technology, social media has become more and more common in human daily life. Social media is a platform for people to express their feelings, feedback, and opinions. To understand the sentiment context of the text, sentiment analysis plays the role to determine whether the sentiment of the text is positive, negative, neutral or any other personal feeling. Sentiment analysis is prominent from the perspective of business or politics where it highly impacts the strategic decision making. The challenges of sentiment analysis are attributable to the lexical diversity, imbalanced dataset and long-distance dependencies of the texts. In view of this, a data augmentation technique with GloVe word embedding is leveraged to synthesize more lexically diverse samples by similar word vector replacements. The data augmentation also focuses on the oversampling of the minority classes to mitigate the imbalanced dataset problems. Apart from that, the existing sentiment analysis mostly leverages sequence models to encode the long-distance dependencies. Nevertheless, the sequence models require a longer execution time as the processing is done sequentially. On the other hand, the Transformer models require less computation time with parallelized processing. To that end, this paper proposes a hybrid deep learning method that combines the strengths of sequence model and Transformer model while suppressing the limitations of sequence model. Specifically, the proposed model integrates Robustly optimized BERT approach and Long Short-Term Memory for sentiment analysis. The Robustly optimized BERT approach maps the words into a compact meaningful word embedding space while the Long Short-Term Memory model captures the long-distance contextual semantics effectively. The experimental results demonstrate that the proposed hybrid model outshines the state-of-the-art methods by achieving F1-scores of 93&#x0025;, 91&#x0025;, and 90&#x0025; on IMDb dataset, Twitter US Airline Sentiment dataset, and Sentiment140 dataset, respectively. 

/pdf/roberta-lstm-a-hybrid-model-for-sentiment-analysis-with-1nn0f331.pdf

RoBERTa-LSTM: A Hybrid Model for Sentiment Analysis With Transformer and Recurrent Neural Network

Due to the rapid development of technology, social media has become more and more common in human daily life. Social media is a platform for people to express their feelings, feedback, and opinions. To understand the sentiment context of the text, sentiment analysis plays the role to determine whether the sentiment of the text is positive, negative, neutral or any other personal feeling. Sentiment analysis is prominent from the perspective of business or politics where it highly impacts the strategic decision making. The challenges of sentiment analysis are attributable to the lexical diversity, imbalanced dataset and long-distance dependencies of the texts. In view of this, a data augmentation technique with GloVe word embedding is leveraged to synthesize more lexically diverse samples by similar word vector replacements. The data augmentation also focuses on the oversampling of the minority classes to mitigate the imbalanced dataset problems. Apart from that, the existing sentiment analysis mostly leverages sequence models to encode the long-distance dependencies. Nevertheless, the sequence models require a longer execution time as the processing is done sequentially. On the other hand, the Transformer models require less computation time with parallelized processing. To that end, this paper proposes a hybrid deep learning method that combines the strengths of sequence model and Transformer model while suppressing the limitations of sequence model. Specifically, the proposed model integrates Robustly optimized BERT approach and Long Short-Term Memory for sentiment analysis. The Robustly optimized BERT approach maps the words into a compact meaningful word embedding space while the Long Short-Term Memory model captures the long-distance contextual semantics effectively. The experimental results demonstrate that the proposed hybrid model outshines the state-of-the-art methods by achieving F1-scores of 93%, 91%, and 90% on IMDb dataset, Twitter US Airline Sentiment dataset, and Sentiment140 dataset, respectively.

https://ieeexplore.ieee.org/ielx7/6287639/9668973/09716923.pdf

• An overview of Homophobia and Transphobia detection on social media. • Fine-grained and Coarse-grained evaluation of Homophobia and Transphobia. • A novel data augmentation approach to detect in multilingual code mixed setting. • An extensive evaluation of the models on the benchmark dataset. • Statistical tests on predictions to enhance the significance of evaluation. Homophobia or Transphobia can be defined as the hatred, discomfort, or dislike of lesbian, gay, transgender or bisexual people. Studies have shown that these individuals were more likely to develop mental health issues, likely due to being subjected to more forms of abuse on social media. Hence there is an ardent need to develop automated abusive speech detection systems to tackle the abusive content on social media. There has been an elevation in hate speech or abuse and this paper focuses on the LGBTQIA+ community. Due to the shortage of resources in the said study area, we hypothesize that data augmentation via Pseudolabeling by transliterating the code-mixed text to the parent language will improve the models’ performances on the newly constructed dataset. We put our hypothesis into testing, and studied the performances of several multilingual language models for our cause. 

How can we detect Homophobia and Transphobia? Experiments in a multilingual code-mixed setting for social media governance

On social networking sites, online hate speech has become more prevalent due to the quick expansion of mobile computing and Web technology. Previous research has found that being exposed to Internet hate speech has substantial offline implications for historically disadvantaged communities. Therefore, there is a lot of interest in research on automated hate-based comment and post detection. Hate speech can have an influence on any population group, but some are more vulnerable than others. From this background, detecting and reporting such hate related comments and posts can help to avoid the harmful effects of hate speech. There are some studies available on this context and it was found that machine learning algorithms are more efficient in detecting abusive texts in social media. In this research, we applied selected seven machine learning algorithms such as Support Vector Machine (SVM), Naïve Bayes (NB), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boost (GB) and K Nearest Neighbor (KNN) to detect hate speech and compare the performances of those algorithms to develop an ensemble model. Researchers collected and combined Tamil – English code-mixed hate speech tweets dataset which was created in HASOC. This dataset's tweets are divided into two groups: not offensive and offensive. This dataset includes 35,442 tweets. In this research, NB has obtained highest F1 scores in detecting offensive and not offensive tweets with highest weighted average. But SVM has obtained highest accuracy in detecting Tamil – English hate speech texts with 80% in 10-fold cross-validation. Based on the stand-alone performances, researchers developed two ensemble classifiers including max-voting and averaging ensemble. Averaging ensemble classification obtained 90.67% in accuracy. The research study's findings are significant because these results can be applied as a model for Tamil – English code-mixed hate speech to evaluate future research works using various algorithms for identifying hate contents more accurately and professionally.

Development of an Efficient Method to Detect Mixed Social Media Data with Tamil-English Code Using Machine Learning Techniques

Sentiment Analysis (SA) is the process of extracting the information from the given text in which the text consists of various emotions such as happy, sad, proud, fear and so on about various events, people, system, and facts. Since the social media is burgeoned with huge amount of data, it is necessary to derive some useful information from the available data. This extracted useful information can be used in product purchase, politics, sports, and more. To be precise, it helps in decision making. For implementing the sentiment analysis, a lot of research work has been carried out which ranges from lexicon based approach to machine learning based approach. In recent days deep learning and neural networks plays a major role in sentiment analysis and it is considered as a state-of-the art method for analysing various languages. Tamil is one of the Indian languages which still requires state-of-art model for sentiment analysis. The language specific features, grammar structure, agglutinative nature of Tamil language poses more challenges. In this paper, we proposed a combined character based Deep Bidirectional long short-term memory neural networks (DBLS TM) to analyse the Tamil tweets.

Analyzing Sentiment in Tamil Tweets using Deep Neural Network

Neural network-based error handler in natural language processing

Social media is a great source of communication. People use various social media platforms, such as Twitter, Facebook, and Instagram, for sharing their ideas, opinions, and feelings. Users of different age groups, cultures, education backgrounds manipulate these powerful mediums of communication. Even though it gives all the benefits of knowledge sharing among the users, it has a dark side too. Despite setting restrictions from the corresponding sites, many users use abusive language to blemish the status and image of someone. So it is highly the need of the hour for the government or the particular social media platform to sift out those unwanted hate texts before diffusing them. Finding the hate text is one of the emerging research topics in Natural Language Processing where the model predicts the given text as hate text or not. This automated hate text detection becomes tedious when we consider the Indian languages due to a lack of data. Moreover, Indian people are multilingual and use code-mixed patterns to express their thoughts. The unavailability of the annotated Tamil-English dataset and the lack of a standard model make this task more challenging. In our paper, to handle such code-mixed data, a dataset is created with 10000 Tamil-English code-mixed texts collected from Twitter. These are annotated as hate text/non-hate text. In this paper, we use a synonym-based Bi-LSTM model for classifying hate non-hate text in tweets. 

Deep Learning-based Hate Speech Detection in Code-mixed Tamil Text

A common foundation for Natural Language Processing (NLP) applications is Named-Entity-Recognition (NER). Organizing data into pre-defined labels is a challenging task, and due to the Tamil language's unique characteristics and complexity, it becomes even more difficult. With the Multilingual Universal Sentence Encoder, we constructed a GRU model using transfer learning and deep neural networks. The enormous amount of unstructured data transmitted every day necessitates the development of effective information retrieval and extraction methods. The act of categorizing the data into pre-defined categories, known as named entity recognition, is a tough task. This work provides a novel Deep Learning technique for Standard Arabic Named Entity Recognition that outperforms previous studies. The main purpose of creating a new model is to provide more fine-grained results for Natural Language Processing applications. We compared RNN, LSTM, GRU, Bi-LSTM, and GRU using deep neural networks. (Abstract)

Named Entity Recognition for Tamil text Using Deep Learning

Depression perception be a complex task on social media Complex properties associated with mental illness. There was a recent development in this area of research; social media platforms have established themselves as growing popularity. A basic part of people`s daily life. Social media platforms and their users share goals Relationships where the user's personal life is reflected on these platforms at several levels. Aside from the complexity associated with detecting mental illness through social media platforms, it is inherently difficult to obtain an enough annotated training data, so a supervised deep learning approach such as deep neural networks the implementation has not yet been widely implemented. We tried to find them for these reasons. The most effective deep learning model of the architectures selected in the previous architecture Achievements of supervised learning methods. The selected model will be used for recognition online users showing depression. Due to the limited amount of unstructured text data Extracted from social media text. Recently, Deep learning has been effectively used to a variety of application challenges, including stock market forecasting, traffic flow and accident risk forecasting, and mental disease diagnosis. Furthermore, deep learning has been used to predict sadness on social media and has outperformed classical machine learning method.

S. Anbukkarasi

Papers

Analyzing Sentiment in Tamil Tweets using Deep Neural Network

Neural network-based error handler in natural language processing

Deep Learning-based Hate Speech Detection in Code-mixed Tamil Text

Named Entity Recognition for Tamil text Using Deep Learning

Deep learning based depression detection from social media text