arXiv subjects

Sunil Saumya

Publications and source records attributed to Sunil Saumya.

7 recordsLinked to original sources

arXiv2026

Extending the Frontiers of QNLP Beyond English: Grammar-Sensitive Pipeline for Hindi Sentiment Classification Using Compositional Quantum Models

Advancements in Natural Language Processing (NLP), whether on classical or quantum platforms, have predominantly focused on English due to its widespread use and abundant linguistic resources. Although English remains the most studied language in computational linguistics, Hindi, the third most spoken language worldwide after Mandarin, has received comparatively limited attention. Spoken primarily in India, Hindi differs significantly from English in its script, syntactic structure, and linguistic characteristics. Hindi uses the Devanagari script, exhibits rich morphological inflection, and follows a subject-object-verb (SOV) word order, unlike English's subject-verb-object (SVO) structure. Motivated by Hindi's linguistic complexity and its underrepresentation in Quantum Natural Language Processing (QNLP), we propose a grammar-aware QNLP pipeline for Hindi sentiment classification with a focus on sentential negation. We use a manually annotated Hindi sentiment dataset labeled as positive, negative, or neutral, and encode sentences using pregroup grammar types. Sentences are processed with Lambeq to generate quantum circuits using a novel negation-aware compositional grammar. Hybrid Quantum Neural Networks (HQNNs) are trained for both binary and ternary sentiment classification. Our results demonstrate effective sentiment classification and highlight the potential of compositional QNLP for morphologically rich languages.

quant-ph↗

arXiv2026

HateMirage: An Explainable Multi-Dimensional Dataset for Decoding Faux Hate and Subtle Online Abuse

Subtle and indirect hate speech remains an underexplored challenge in online safety research, particularly when harmful intent is embedded within misleading or manipulative narratives. Existing hate speech datasets primarily capture overt toxicity, underrepresenting the nuanced ways misinformation can incite or normalize hate. To address this gap, we present HateMirage, a novel dataset of Faux Hate comments designed to advance reasoning and explainability research on hate emerging from fake or distorted narratives. The dataset was constructed by identifying widely debunked misinformation claims from fact-checking sources and tracing related YouTube discussions, resulting in 4,530 user comments. Each comment is annotated along three interpretable dimensions: Target (who is affected), Intent (the underlying motivation or goal behind the comment), and Implication (its potential social impact). Unlike prior explainability datasets such as HateXplain and HARE, which offer token-level or single-dimensional reasoning, HateMirage introduces a multi-dimensional explanation framework that captures the interplay between misinformation, harm, and social consequence. We benchmark multiple open-source language models on HateMirage using ROUGE-L F1 and Sentence-BERT similarity to assess explanation coherence. Results suggest that explanation quality may depend more on pretraining diversity and reasoning-oriented data rather than on model scale alone. By coupling misinformation reasoning with harm attribution, HateMirage establishes a new benchmark for interpretable hate detection and responsible AI research.

cs.CL↗

arXiv2025

Deceptive Humor: A Synthetic Multilingual Benchmark Dataset for Bridging Fabricated Claims with Humorous Content

In the evolving landscape of online discourse, misinformation increasingly adopts humorous tones to evade detection and gain traction. This work introduces Deceptive Humor as a novel research direction, emphasizing how false narratives, when coated in humor, can become more difficult to detect and more likely to spread. To support research in this space, we present the Deceptive Humor Dataset (DHD) a collection of humor-infused comments derived from fabricated claims using the ChatGPT-4o model. Each entry is labeled with a Satire Level (from 1 for subtle satire to 3 for overt satire) and categorized into five humor types: Dark Humor, Irony, Social Commentary, Wordplay, and Absurdity. The dataset spans English, Telugu, Hindi, Kannada, Tamil, and their code-mixed forms, making it a valuable resource for multilingual analysis. DHD offers a structured foundation for understanding how humor can serve as a vehicle for the propagation of misinformation, subtly enhancing its reach and impact. Strong baselines are established to encourage further research and model development in this emerging area.

cs.CL↗

arXiv2023

Enabling Quantum Natural Language Processing for Hindi Language

Quantum Natural Language Processing (QNLP) is taking huge leaps in solving the shortcomings of classical Natural Language Processing (NLP) techniques and moving towards a more "Explainable" NLP system. The current literature around QNLP focuses primarily on implementing QNLP techniques in sentences in the English language. In this paper, we propose to enable the QNLP approach to HINDI, which is the third most spoken language in South Asia. We present the process of building the parameterized quantum circuits required to undertake QNLP on Hindi sentences. We use the pregroup representation of Hindi and the DisCoCat framework to draw sentence diagrams. Later, we translate these diagrams to Parameterised Quantum Circuits based on Instantaneous Quantum Polynomial (IQP) style ansatz. Using these parameterized quantum circuits allows one to train grammar and topic-aware sentence classifiers for the Hindi Language.

cs.CL↗

arXiv2022

Hope Speech Detection on Social Media Platforms

Since personal computers became widely available in the consumer market, the amount of harmful content on the internet has significantly expanded. In simple terms, harmful content is anything online which causes a person distress or harm. It may include hate speech, violent content, threats, non-hope speech, etc. The online content must be positive, uplifting and supportive. Over the past few years, many studies have focused on solving this problem through hate speech detection, but very few focused on identifying hope speech. This paper discusses various machine learning approaches to identify a sentence as Hope Speech, Non-Hope Speech, or a Neutral sentence. The dataset used in the study contains English YouTube comments and is released as a part of the shared task "EACL-2021: Hope Speech Detection for Equality, Diversity, and Inclusion". Initially, the dataset obtained from the shared task had three classes: Hope Speech, non-Hope speech, and not in English; however, upon deeper inspection, we discovered that dataset relabeling is required. A group of undergraduates was hired to help perform the entire dataset's relabeling task. We experimented with conventional machine learning models (such as Naïve Bayes, logistic regression and support vector machine) and pre-trained models (such as BERT) on relabeled data. According to the experimental results, the relabeled data has achieved a better accuracy for Hope speech identification than the original data set.

cs.CL↗

arXiv2022

IIITDWD-ShankarB@ Dravidian-CodeMixi-HASOC2021: mBERT based model for identification of offensive content in south Indian languages

In recent years, there has been a lot of focus on offensive content. The amount of offensive content generated by social media is increasing at an alarming rate. This created a greater need to address this issue than ever before. To address these issues, the organizers of "Dravidian-Code Mixed HASOC-2020" have created two challenges. Task 1 involves identifying offensive content in Malayalam data, whereas Task 2 includes Malayalam and Tamil Code Mixed Sentences. Our team participated in Task 2. In our suggested model, we experiment with multilingual BERT to extract features, and three different classifiers are used on extracted features. Our model received a weighted F1 score of 0.70 for Malayalam data and was ranked fifth; we also received a weighted F1 score of 0.573 for Tamil Code Mixed data and were ranked eleventh.

cs.CL↗

arXiv2019

Ranking Online Consumer Reviews

The product reviews are posted online in the hundreds and even in the thousands for some popular products. Handling such a large volume of continuously generated online content is a challenging task for buyers, sellers, and even researchers. The purpose of this study is to rank the overwhelming number of reviews using their predicted helpfulness score. The helpfulness score is predicted using features extracted from review text data, product description data and customer question-answer data of a product using random-forest classifier and gradient boosting regressor. The system is made to classify the reviews into low or high quality by random-forest classifier. The helpfulness score of the high-quality reviews is only predicted using gradient boosting regressor. The helpfulness score of the low-quality reviews is not calculated because they are never going to be in the top k reviews. They are just added at the end of the review list to the review-listing website. The proposed system provides fair review placement on review listing pages and making all high-quality reviews visible to customers on the top. The experimental results on data from two popular Indian e-commerce websites validate our claim, as 3-4 new high-quality reviews are placed in the top ten reviews along with 5-6 old reviews based on review helpfulness. Our findings indicate that inclusion of features from product description data and customer question-answer data improves the prediction accuracy of the helpfulness score.

cs.IR↗