Ponnurangam Kumaraguru

Mehrad Moradshahi, Tianhao Shen, Kalika Bali et al. · stanford

Task-oriented dialogue research has mainly focused on a few popular languages like English and Chinese, due to the high dataset creation cost for a new language. To reduce the cost, we apply manual editing to automatically translated data. We create a new multilingual benchmark, X-RiSAWOZ, by translating the Chinese RiSAWOZ to 4 languages: English, French, Hindi, Korean; and a code-mixed English-Hindi language. X-RiSAWOZ has more than 18,000 human-verified dialogue utterances for each language, and unlike most multilingual prior work, is an end-to-end dataset for building fully-functioning agents. The many difficulties we encountered in creating X-RiSAWOZ led us to develop a toolset to accelerate the post-editing of a new language dataset after translation. This toolset improves machine translation with a hybrid entity alignment technique that combines neural with dictionary-based methods, along with many automated and semi-automated validation checks. We establish strong baselines for X-RiSAWOZ by training dialogue agents in the zero- and few-shot settings where limited gold data is available in the target language. Our results suggest that our translation and post-editing methodology and toolset can be used to create new high-quality multilingual dialogue agents cost-effectively. Our dataset, code, and toolkit are released open-source.

Shrey Gupta, Anmol Agarwal, Manas Gaur et al.

Conversational Agents (CAs) powered with deep language models (DLMs) have shown tremendous promise in the domain of mental health. Prominently, the CAs have been used to provide informational or therapeutic services to patients. However, the utility of CAs to assist in mental health triaging has not been explored in the existing work as it requires a controlled generation of follow-up questions (FQs), which are often initiated and guided by the mental health professionals (MHPs) in clinical settings. In the context of depression, our experiments show that DLMs coupled with process knowledge in a mental health questionnaire generate 12.54% and 9.37% better FQs based on similarity and longest common subsequence matches to questions in the PHQ-9 dataset respectively, when compared with DLMs without process knowledge support. Despite coupling with process knowledge, we find that DLMs are still prone to hallucination, i.e., generating redundant, irrelevant, and unsafe FQs. We demonstrate the challenge of using existing datasets to train a DLM for generating FQs that adhere to clinical process knowledge. To address this limitation, we prepared an extended PHQ-9 based dataset, PRIMATE, in collaboration with MHPs. PRIMATE contains annotations regarding whether a particular question in the PHQ-9 dataset has already been answered in the user's initial description of the mental health condition. We used PRIMATE to train a DLM in a supervised setting to identify which of the PHQ-9 questions can be answered directly from the user's post and which ones would require more information from the user. Using performance analysis based on MCC scores, we show that PRIMATE is appropriate for identifying questions in PHQ-9 that could guide generative DLMs towards controlled FQ generation suitable for aiding triaging. Dataset created as a part of this research: https://github.com/primate-mh/Primate2022

Arnav Kapoor, Mudit Dhawan, Anmol Goel et al.

Many populous countries including India are burdened with a considerable backlog of legal cases. Development of automated systems that could process legal documents and augment legal practitioners can mitigate this. However, there is a dearth of high-quality corpora that is needed to develop such data-driven systems. The problem gets even more pronounced in the case of low resource languages such as Hindi. In this resource paper, we introduce the Hindi Legal Documents Corpus (HLDC), a corpus of more than 900K legal documents in Hindi. Documents are cleaned and structured to enable the development of downstream applications. Further, as a use-case for the corpus, we introduce the task of bail prediction. We experiment with a battery of models and propose a Multi-Task Learning (MTL) based model for the same. MTL models use summarization as an auxiliary task along with bail prediction as the main task. Experiments with different models are indicative of the need for further research in this area. We release the corpus and model implementation code with this paper: https://github.com/Exploration-Lab/HLDC

Ishan Kavathekar, Anku Rani, Ashmit Chamoli et al.

The widespread adoption of Large Language Models (LLMs) and awareness around multilingual LLMs have raised concerns regarding the potential risks and repercussions linked to the misapplication of AI-generated text, necessitating increased vigilance. While these models are primarily trained for English, their extensive training on vast datasets covering almost the entire web, equips them with capabilities to perform well in numerous other languages. AI-Generated Text Detection (AGTD) has emerged as a topic that has already received immediate attention in research, with some initial methods having been proposed, soon followed by the emergence of techniques to bypass detection. In this paper, we report our investigation on AGTD for an indic language Hindi. Our major contributions are in four folds: i) examined 26 LLMs to evaluate their proficiency in generating Hindi text, ii) introducing the AI-generated news article in Hindi ($AG_{hi}$) dataset, iii) evaluated the effectiveness of five recently proposed AGTD techniques: ConDA, J-Guard, RADAR, RAIDAR and Intrinsic Dimension Estimation for detecting AI-generated Hindi text, iv) proposed Hindi AI Detectability Index ($ADI_{hi}$) which shows a spectrum to understand the evolving landscape of eloquence of AI-generated text in Hindi. The code and dataset is available at https://github.com/ishank31/Counter_Turing_Test

Mann Khatri, Mirza Yusuf, Yaman Kumar et al.

The burdensome impact of a skewed judges-to-cases ratio on the judicial system manifests in an overwhelming backlog of pending cases alongside an ongoing influx of new ones. To tackle this issue and expedite the judicial process, the proposition of an automated system capable of suggesting case outcomes based on factual evidence and precedent from past cases gains significance. This research paper centres on developing a graph neural network-based model to address the Legal Judgment Prediction (LJP) problem, recognizing the intrinsic graph structure of judicial cases and making it a binary node classification problem. We explored various embeddings as model features, while nodes such as time nodes and judicial acts were added and pruned to evaluate the model's performance. The study is done while considering the ethical dimension of fairness in these predictions, considering gender and name biases. A link prediction task is also conducted to assess the model's proficiency in anticipating connections between two specified nodes. By harnessing the capabilities of graph neural networks and incorporating fairness analyses, this research aims to contribute insights towards streamlining the adjudication process, enhancing judicial efficiency, and fostering a more equitable legal landscape, ultimately alleviating the strain imposed by mounting case backlogs. Our best-performing model with XLNet pre-trained embeddings as its features gives the macro F1 score of 75% for the LJP task. For link prediction, the same set of features is the best performing giving ROC of more than 80%

Sahil Girhepuje, Anmol Goel, Gokul S Krishnan et al.

Recent advances and applications of language technology and artificial intelligence have enabled much success across multiple domains like law, medical and mental health. AI-based Language Models, like Judgement Prediction, have recently been proposed for the legal sector. However, these models are strife with encoded social biases picked up from the training data. While bias and fairness have been studied across NLP, most studies primarily locate themselves within a Western context. In this work, we present an initial investigation of fairness from the Indian perspective in the legal domain. We highlight the propagation of learnt algorithmic biases in the bail prediction task for models trained on Hindi legal documents. We evaluate the fairness gap using demographic parity and show that a decision tree model trained for the bail prediction task has an overall fairness disparity of 0.237 between input features associated with Hindus and Muslims. Additionally, we highlight the need for further research and studies in the avenues of fairness/bias in applying AI in the legal sector with a specific focus on the Indian context.

Prashant Kodali, Tanmay Sachan, Akshay Goindani et al.

Code-Mixing is a phenomenon of mixing two or more languages in a speech event and is prevalent in multilingual societies. Given the low-resource nature of Code-Mixing, machine generation of code-mixed text is a prevalent approach for data augmentation. However, evaluating the quality of such machine generated code-mixed text is an open problem. In our submission to HinglishEval, a shared-task collocated with INLG2022, we attempt to build models factors that impact the quality of synthetically generated code-mix text by predicting ratings for code-mix quality.

Ishwar B Balappanawar, Ashmit Chamoli, Ruwan Wickramarachchi et al.

Distracted driving is a leading cause of road accidents globally. Identification of distracted driving involves reliably detecting and classifying various forms of driver distraction (e.g., texting, eating, or using in-car devices) from in-vehicle camera feeds to enhance road safety. This task is challenging due to the need for robust models that can generalize to a diverse set of driver behaviors without requiring extensive annotated datasets. In this paper, we propose KiD3, a novel method for distracted driver detection (DDD) by infusing auxiliary knowledge about semantic relations between entities in a scene and the structural configuration of the driver's pose. Specifically, we construct a unified framework that integrates the scene graphs, and driver pose information with the visual cues in video frames to create a holistic representation of the driver's actions.Our results indicate that KiD3 achieves a 13.64% accuracy improvement over the vision-only baseline by incorporating such auxiliary knowledge with visual information.

Arvindh Arun, Aakash Aanegola, Amul Agrawal et al.

Unsupervised Representation Learning on graphs is gaining traction due to the increasing abundance of unlabelled network data and the compactness, richness, and usefulness of the representations generated. In this context, the need to consider fairness and bias constraints while generating the representations has been well-motivated and studied to some extent in prior works. One major limitation of most of the prior works in this setting is that they do not aim to address the bias generated due to connectivity patterns in the graphs, such as varied node centrality, which leads to a disproportionate performance across nodes. In our work, we aim to address this issue of mitigating bias due to inherent graph structure in an unsupervised setting. To this end, we propose CAFIN, a centrality-aware fairness-inducing framework that leverages the structural information of graphs to tune the representations generated by existing frameworks. We deploy it on GraphSAGE (a popular framework in this domain) and showcase its efficacy on two downstream tasks - Node Classification and Link Prediction. Empirically, CAFIN consistently reduces the performance disparity across popular datasets (varying from 18 to 80% reduction in performance disparity) from various domains while incurring only a minimal cost of fairness.

Devansh Gupta, Aditya Saini, Drishti Bhasin et al.

Retrieving facial images from attributes plays a vital role in various systems such as face recognition and suspect identification. Compared to other image retrieval tasks, facial image retrieval is more challenging due to the high subjectivity involved in describing a person's facial features. Existing methods do so by comparing specific characteristics from the user's mental image against the suggested images via high-level supervision such as using natural language. In contrast, we propose a method that uses a relatively simpler form of binary supervision by utilizing the user's feedback to label images as either similar or dissimilar to the target image. Such supervision enables us to exploit the contrastive learning paradigm for encapsulating each user's personalized notion of similarity. For this, we propose a novel loss function optimized online via user feedback. We validate the efficacy of our proposed approach using a carefully designed testbed to simulate user feedback and a large-scale user study. Our experiments demonstrate that our method iteratively improves personalization, leading to faster convergence and enhanced recommendation relevance, thereby, improving user satisfaction. Our proposed framework is also equipped with a user-friendly web interface with a real-time experience for facial image retrieval.

Ritwik Mishra, Sreeram Vennam, Rajiv Ratn Shah et al.

Most existing Question Answering Datasets (QuADs) primarily focus on factoid-based short-context Question Answering (QA) in high-resource languages. However, the scope of such datasets for low-resource languages remains limited, with only a few works centered on factoid-based QuADs and none on non-factoid QuADs. Therefore, this work presents MuNfQuAD, a multilingual QuAD with non-factoid questions. It utilizes interrogative sub-headings from BBC news articles as questions and the corresponding paragraphs as silver answers. The dataset comprises over 578K QA pairs across 38 languages, encompassing several low-resource languages, and stands as the largest multilingual QA dataset to date. Based on the manual annotations of 790 QA-pairs from MuNfQuAD (golden set), we observe that 98\% of questions can be answered using their corresponding silver answer. Our fine-tuned Answer Paragraph Selection (APS) model outperforms the baselines. The APS model attained an accuracy of 80\% and 72\%, as well as a macro F1 of 72\% and 66\%, on the MuNfQuAD testset and the golden set, respectively. Furthermore, the APS model effectively generalizes a certain language within the golden set, even after being fine-tuned on silver labels. We also observe that the fine-tuned APS model is beneficial for reducing the context of a question. These findings suggest that this resource would be a valuable contribution to the QA research community.

Sharath Srivatsa, Tushar Mohan, Kumari Neha et al.

Online news and social media have been the de facto mediums to disseminate information globally from the beginning of the last decade. However, bias in content and purpose of intentions are not regulated, and managing bias is the responsibility of content consumers. In this regard, understanding the stances and biases of news sources towards specific entities becomes important. To address this problem, we use pretrained language models, which have been shown to bring about good results with no task-specific training or few-shot training. In this work, we approach the problem of characterizing Named Entities and Tweets as an open-ended text classification and open-ended fact probing problem.We evaluate the zero-shot language model capabilities of Generative Pretrained Transformer 2 (GPT-2) to characterize Entities and Tweets subjectively with human psychology-inspired and logical conditional prefixes and contexts. First, we fine-tune the GPT-2 model on a sufficiently large news corpus and evaluate subjective characterization of popular entities in the corpus by priming with prefixes. Second, we fine-tune GPT-2 with a Tweets corpus from a few popular hashtags and evaluate characterizing tweets by priming the language model with prefixes, questions, and contextual synopsis prompts. Entity characterization results were positive across measures and human evaluation.

Varshita Kolipaka, Akshit Sinha, Debangan Mishra et al.

Graph Neural Networks (GNNs) are increasingly being used for a variety of ML applications on graph data. Because graph data does not follow the independently and identically distributed (i.i.d.) assumption, adversarial manipulations or incorrect data can propagate to other data points through message passing, which deteriorates the model's performance. To allow model developers to remove the adverse effects of manipulated entities from a trained GNN, we study the recently formulated problem of Corrective Unlearning. We find that current graph unlearning methods fail to unlearn the effect of manipulations even when the whole manipulated set is known. We introduce a new graph unlearning method, Cognac, which can unlearn the effect of the manipulation set even when only 5% of it is identified. It recovers most of the performance of a strong oracle with fully corrected training data, even beating retraining from scratch without the deletion set while being 8x more efficient. We hope our work assists GNN developers in mitigating harmful effects caused by issues in real-world data, post-training. Our code is publicly available at https://github.com/cognac-gnn-unlearning/corrective-unlearning-for-gnns

Raghav Awasthi, Keerat Kaur Guliani, Saif Ahmad Khan et al.

A COVID-19 vaccine is our best bet for mitigating the ongoing onslaught of the pandemic. However, vaccine is also expected to be a limited resource. An optimal allocation strategy, especially in countries with access inequities and temporal separation of hot-spots, might be an effective way of halting the disease spread. We approach this problem by proposing a novel pipeline VacSIM that dovetails Deep Reinforcement Learning models into a Contextual Bandits approach for optimizing the distribution of COVID-19 vaccine. Whereas the Reinforcement Learning models suggest better actions and rewards, Contextual Bandits allow online modifications that may need to be implemented on a day-to-day basis in the real world scenario. We evaluate this framework against a naive allocation approach of distributing vaccine proportional to the incidence of COVID-19 cases in five different States across India (Assam, Delhi, Jharkhand, Maharashtra and Nagaland) and demonstrate up to 9039 potential infections prevented and a significant increase in the efficacy of limiting the spread over a period of 45 days through the VacSIM approach. Our models and the platform are extensible to all states of India and potentially across the globe. We also propose novel evaluation strategies including standard compartmental model-based projections and a causality-preserving evaluation of our model. Since all models carry assumptions that may need to be tested in various contexts, we open source our model VacSIM and contribute a new reinforcement learning environment compatible with OpenAI gym to make it extensible for real-world applications across the globe. (http://vacsim.tavlab.iiitd.edu.in:8000/).

Shashwat Goel, Joschka Struber, Ilze Amanda Auzina et al.

As Language Model (LM) capabilities advance, evaluating and supervising them at scale is getting harder for humans. There is hope that other language models can automate both these tasks, which we refer to as ''AI Oversight''. We study how model similarity affects both aspects of AI oversight by proposing Chance Adjusted Probabilistic Agreement (CAPA): a metric for LM similarity based on overlap in model mistakes. Using CAPA, we first show that LLM-as-a-judge scores favor models similar to the judge, generalizing recent self-preference results. Then, we study training on LM annotations, and find complementary knowledge between the weak supervisor and strong student model plays a crucial role in gains from ''weak-to-strong generalization''. As model capabilities increase, it becomes harder to find their mistakes, and we might defer more to AI oversight. However, we observe a concerning trend -- model mistakes are becoming more similar with increasing capabilities, pointing to risks from correlated failures. Our work underscores the importance of reporting and correcting for model similarity, especially in the emerging paradigm of AI oversight.

Rahul Garg, Trilok Padhi, Hemang Jain et al.

Toxicity identification in online multimodal environments remains a challenging task due to the complexity of contextual connections across modalities (e.g., textual and visual). In this paper, we propose a novel framework that integrates Knowledge Distillation (KD) from Large Visual Language Models (LVLMs) and knowledge infusion to enhance the performance of toxicity detection in hateful memes. Our approach extracts sub-knowledge graphs from ConceptNet, a large-scale commonsense Knowledge Graph (KG) to be infused within a compact VLM framework. The relational context between toxic phrases in captions and memes, as well as visual concepts in memes enhance the model's reasoning capabilities. Experimental results from our study on two hate speech benchmark datasets demonstrate superior performance over the state-of-the-art baselines across AU-ROC, F1, and Recall with improvements of 1.1%, 7%, and 35%, respectively. Given the contextual complexity of the toxicity detection task, our approach showcases the significance of learning from both explicit (i.e. KG) as well as implicit (i.e. LVLMs) contextual cues incorporated through a hybrid neurosymbolic approach. This is crucial for real-world applications where accurate and scalable recognition of toxic content is critical for creating safer online environments.

Priyanshul Govil, Hemang Jain, Vamshi Krishna Bonagiri et al.

Large Language Models (LLMs) often inherit biases from the web data they are trained on, which contains stereotypes and prejudices. Current methods for evaluating and mitigating these biases rely on bias-benchmark datasets. These benchmarks measure bias by observing an LLM's behavior on biased statements. However, these statements lack contextual considerations of the situations they try to present. To address this, we introduce a contextual reliability framework, which evaluates model robustness to biased statements by considering the various contexts in which they may appear. We develop the Context-Oriented Bias Indicator and Assessment Score (COBIAS) to measure a biased statement's reliability in detecting bias, based on the variance in model behavior across different contexts. To evaluate the metric, we augmented 2,291 stereotyped statements from two existing benchmark datasets by adding contextual information. We show that COBIAS aligns with human judgment on the contextual reliability of biased statements (Spearman's $ρ= 0.65, p = 3.4 * 10^{-60}$) and can be used to create reliable benchmarks, which would assist bias mitigation works.

Sreeram Vennam, David Valente, David Herel et al.

Thinking Tokens (TT) have been proposed as an unsupervised method to facilitate reasoning in language models. However, despite their conceptual appeal, our findings show that TTs marginally improves performance and consistently underperforms compared to Chain-of-Thought (CoT) reasoning across multiple benchmarks. We hypothesize that this underperformance stems from the reliance on a single embedding for TTs, which results in inconsistent learning signals and introduces noisy gradients. This paper provides a comprehensive empirical analysis to validate this hypothesis and discusses the implications for future research on unsupervised reasoning in LLMs.

Vivek Hruday Kavuri, Vysishtya Karanam, Venkata Jahnavi Venkamsetty et al.

Vision Language Models achieve impressive multi-modal performance but often inherit gender biases from their training data. This bias might be coming from both the vision and text modalities. In this work, we dissect the contributions of vision and text backbones to these biases by applying targeted debiasing using Counterfactual Data Augmentation and Task Vector methods. Inspired by data-efficient approaches in hate-speech classification, we introduce a novel metric, Degree of Stereotypicality and a corresponding debiasing method, Data Augmentation Using Degree of Stereotypicality - DAUDoS, to reduce bias with minimal computational cost. We curate a gender annotated dataset and evaluate all methods on VisoGender benchmark to quantify improvements and identify dominant source of bias. Our results show that CDA reduces the gender gap by 6% and DAUDoS by 3% but using only one-third of the data. Both methods also improve the model's ability to correctly identify gender in images by 3%, with DAUDoS achieving this improvement using only almost one-third of training data. From our experiment's, we observed that CLIP's vision encoder is more biased whereas PaliGemma2's text encoder is more biased. By identifying whether bias stems more from vision or text encoders, our work enables more targeted and effective bias mitigation strategies in future multi-modal systems.

Sreeram Vennam, Anish Joishy, Ponnurangam Kumaraguru

We present a method to compress the final linear layer of language models, reducing memory usage by up to 3.4x without significant performance loss. By grouping tokens based on Byte Pair Encoding (BPE) merges, we prevent materialization of the memory-intensive logits tensor. Evaluations on the TinyStories dataset show that our method performs on par with GPT-Neo and GPT2 while significantly improving throughput by up to 3x, making it suitable for low-compute environments.

Prashant Kodali, Vaishnavi Shivkumar, Swarang Joshi et al.

We study model merging as a practical alternative to conventional adaptation strategies for code-mixed NLP. Starting from a multilingual base model, we: (i) perform continued pre-training (CPT) on unlabeled code-mixed text to obtain an adapted checkpoint, (ii) merge checkpoint with the base model, and (iii) fine-tune (FT) on the downstream task data. We evaluate our approach for sentence classification (sentiment and hate speech) task in English-Hindi (En-Hi) and English-Spanish (En-Es) using XLM-R and Llama-3.2-1B models. Our results show that merged models consistently outperform full fine-tuning and CPT->FT. We observe gains of 2--5 points in F1 over full fine-tuning and ~1-2 points over CPT->FT, indicating that unlabeled data is leveraged more effectively via merging than via CPT alone. Zero-/few-shot prompting with larger LLMs (e.g., Llama-3.3-70B) lags behind fine-tuned and merged checkpoints, underscoring limits of in-context learning for code-mixed inputs. We further test cross-pair transfer by training on En-Hi and evaluating on En-Ta and En-Ml: merged checkpoints transfer more strongly than monolingual-English baselines (e.g., TV/TIES variants reaching 0.65-0.68 F1 vs 0.61-0.63 for full fine-tuning), suggesting that code-mixed knowledge is a more reliable substrate for low-resource pairs. We conclude with adaptation recipes matched to common data regimes (labeled only; labeled+unlabeled; transfer-only) and discuss limitations and scaling considerations for broader tasks and larger models.

Ishwar B Balappanawar, Vamshi Krishna Bonagiri, Anish R Joishy et al.

Large Language Models (LLMs) demonstrate impressive reasoning capabilities in familiar contexts, but struggle when the context conflicts with their parametric knowledge. To investigate this phenomenon, we introduce CounterLogic, a dataset containing 1,800 examples across 9 logical schemas, explicitly designed to evaluate logical reasoning through counterfactual (hypothetical knowledge-conflicting) scenarios. Our systematic evaluation of 11 LLMs across 6 different datasets reveals a consistent performance degradation, with accuracies dropping by 27% on average when reasoning through counterfactual information. We propose Self-Segregate, a prompting method enabling metacognitive awareness (explicitly identifying knowledge conflicts) before reasoning. Our method dramatically narrows the average performance gaps from 27% to just 11%, while significantly increasing the overall accuracy (+7.5%). We discuss the implications of these findings and draw parallels to human cognitive processes, particularly on how humans disambiguate conflicting information during reasoning tasks. Our findings offer practical insights for understanding and enhancing LLMs reasoning capabilities in real-world applications, especially where models must logically reason independently of their factual knowledge.

Prashant Kodali, Anmol Goel, Likhith Asapu et al.

Current computational approaches for analysing or generating code-mixed sentences do not explicitly model ``naturalness'' or ``acceptability'' of code-mixed sentences, but rely on training corpora to reflect distribution of acceptable code-mixed sentences. Modelling human judgement for the acceptability of code-mixed text can help in distinguishing natural code-mixed text and enable quality-controlled generation of code-mixed text. To this end, we construct Cline - a dataset containing human acceptability judgements for English-Hindi~(en-hi) code-mixed text. Cline is the largest of its kind with 16,642 sentences, consisting of samples sourced from two sources: synthetically generated code-mixed text and samples collected from online social media. Our analysis establishes that popular code-mixing metrics such as CMI, Number of Switch Points, Burstines, which are used to filter/curate/compare code-mixed corpora have low correlation with human acceptability judgements, underlining the necessity of our dataset. Experiments using Cline demonstrate that simple Multilayer Perceptron (MLP) models when trained solely using code-mixing metrics as features are outperformed by fine-tuned pre-trained Multilingual Large Language Models (MLLMs). Specifically, among Encoder models XLM-Roberta and Bernice outperform IndicBERT across different configurations. Among Encoder-Decoder models, mBART performs better than mT5, however Encoder-Decoder models are not able to outperform Encoder-only models. Decoder-only models perform the best when compared to all other MLLMS, with Llama 3.2 - 3B models outperforming similarly sized Qwen, Phi models. Comparison with zero and fewshot capabilitites of ChatGPT show that MLLMs fine-tuned on larger data outperform ChatGPT, providing scope for improvement in code-mixed tasks. Zero-shot transfer from En-Hi to En-Te acceptability judgments are better than random baselines.

Vamshi Krishna Bonagiri, Sreeram Vennam, Priyanshul Govil et al.

Despite recent advancements showcasing the impressive capabilities of Large Language Models (LLMs) in conversational systems, we show that even state-of-the-art LLMs are morally inconsistent in their generations, questioning their reliability (and trustworthiness in general). Prior works in LLM evaluation focus on developing ground-truth data to measure accuracy on specific tasks. However, for moral scenarios that often lack universally agreed-upon answers, consistency in model responses becomes crucial for their reliability. To address this issue, we propose an information-theoretic measure called Semantic Graph Entropy (SaGE), grounded in the concept of "Rules of Thumb" (RoTs) to measure a model's moral consistency. RoTs are abstract principles learned by a model and can help explain their decision-making strategies effectively. To this extent, we construct the Moral Consistency Corpus (MCC), containing 50K moral questions, responses to them by LLMs, and the RoTs that these models followed. Furthermore, to illustrate the generalizability of SaGE, we use it to investigate LLM consistency on two popular datasets -- TruthfulQA and HellaSwag. Our results reveal that task-accuracy and consistency are independent problems, and there is a dire need to investigate these issues further.

Vamshi Krishna Bonagiri, Sreeram Vennam, Manas Gaur et al.

A Large Language Model (LLM) is considered consistent if semantically equivalent prompts produce semantically equivalent responses. Despite recent advancements showcasing the impressive capabilities of LLMs in conversational systems, we show that even state-of-the-art LLMs are highly inconsistent in their generations, questioning their reliability. Prior research has tried to measure this with task-specific accuracy. However, this approach is unsuitable for moral scenarios, such as the trolley problem, with no "correct" answer. To address this issue, we propose a novel information-theoretic measure called Semantic Graph Entropy (SGE) to measure the consistency of an LLM in moral scenarios. We leverage "Rules of Thumb" (RoTs) to explain a model's decision-making strategies and further enhance our metric. Compared to existing consistency metrics, SGE correlates better with human judgments across five LLMs. In the future, we aim to investigate the root causes of LLM inconsistencies and propose improvements.

Mann Khatri, Pritish Wadhwa, Gitansh Satija et al.

In legal document writing, one of the key elements is properly citing the case laws and other sources to substantiate claims and arguments. Understanding the legal domain and identifying appropriate citation context or cite-worthy sentences are challenging tasks that demand expensive manual annotation. The presence of jargon, language semantics, and high domain specificity makes legal language complex, making any associated legal task hard for automation. The current work focuses on the problem of citation-worthiness identification. It is designed as the initial step in today's citation recommendation systems to lighten the burden of extracting an adequate set of citation contexts. To accomplish this, we introduce a labeled dataset of 178M sentences for citation-worthiness detection in the legal domain from the Caselaw Access Project (CAP). The performance of various deep learning models was examined on this novel dataset. The domain-specific pre-trained model tends to outperform other models, with an 88% F1-score for the citation-worthiness detection task.

Mudit Dhawan, Shakshi Sharma, Aditya Kadam et al.

Social media in present times has a significant and growing influence. Fake news being spread on these platforms have a disruptive and damaging impact on our lives. Furthermore, as multimedia content improves the visibility of posts more than text data, it has been observed that often multimedia is being used for creating fake content. A plethora of previous multimodal-based work has tried to address the problem of modeling heterogeneous modalities in identifying fake content. However, these works have the following limitations: (1) inefficient encoding of inter-modal relations by utilizing a simple concatenation operator on the modalities at a later stage in a model, which might result in information loss; (2) training very deep neural networks with a disproportionate number of parameters on small but complex real-life multimodal datasets result in higher chances of overfitting. To address these limitations, we propose GAME-ON, a Graph Neural Network based end-to-end trainable framework that allows granular interactions within and across different modalities to learn more robust data representations for multimodal fake news detection. We use two publicly available fake news datasets, Twitter and Weibo, for evaluations. Our model outperforms on Twitter by an average of 11% and keeps competitive performance on Weibo, within a 2.6% margin, while using 65% fewer parameters than the best comparable state-of-the-art baseline.

Prashant Kodali, Akshala Bhatnagar, Naman Ahuja et al.

Hashtag segmentation is the task of breaking a hashtag into its constituent tokens. Hashtags often encode the essence of user-generated posts, along with information like topic and sentiment, which are useful in downstream tasks. Hashtags prioritize brevity and are written in unique ways -- transliterating and mixing languages, spelling variations, creative named entities. Benchmark datasets used for the hashtag segmentation task -- STAN, BOUN -- are small in size and extracted from a single set of tweets. However, datasets should reflect the variations in writing styles of hashtags and also account for domain and language specificity, failing which the results will misrepresent model performance. We argue that model performance should be assessed on a wider variety of hashtags, and datasets should be carefully curated. To this end, we propose HashSet, a dataset comprising of: a) 1.9k manually annotated dataset; b) 3.3M loosely supervised dataset. HashSet dataset is sampled from a different set of tweets when compared to existing datasets and provides an alternate distribution of hashtags to build and validate hashtag segmentation models. We show that the performance of SOTA models for Hashtag Segmentation drops substantially on proposed dataset, indicating that the proposed dataset provides an alternate set of hashtags to train and assess models.

Ashwin Singh, Mallika Subramanian, Anmol Agarwal et al.

In the last two decades, ICTs have played a pivotal role in empowering rural populations in India by making knowledge more accessible. Digital Green (DG) is one such ICT that employs a participatory approach with smallholder farmers to produce instructional videos that encompass content specific to them. With help of human mediators, they disseminate these videos using projectors to improve the adoption of agricultural practices. DG's web-based data tracker stores attendance and adoption logs of millions of farmers, videos screened and their demographic information. We leverage this data for a period of ten years between 2010-2020 across five states in India and use it to conduct a holistic evaluation of the ICT. First, we find disparities in adoption rates of farmers, following which we use statistical tests to identify different factors that lead to these disparities and gender-based inequalities. Second, to provide assistance to farmers facing challenges, we model the adoption of practices from a video as a prediction problem and experiment with different model architectures. Our classifier achieves accuracies ranging from 79% to 90% across the five states, demonstrating its potential for assisting future ethnographic investigations. Third, we use SHAP values in conjunction with our model for explaining the impact of various network, content and demographic features on adoption. Our research finds that farmers greatly benefit from past adopters of a video from their group and village. We also discover that videos with a low content-specificity benefit some farmers more than others. Next, we highlight the implications of our findings by translating them into recommendations for community building, revisiting participatory approach and mitigating inequalities. We conclude with a discussion on how our work can assist future investigations into the lived experiences of farmers.

Sezal Chug, Priya Kaushal, Ponnurangam Kumaraguru et al.

Data is expanding at an unimaginable rate, and with this development comes the responsibility of the quality of data. Data Quality refers to the relevance of the information present and helps in various operations like decision making and planning in a particular organization. Mostly data quality is measured on an ad-hoc basis, and hence none of the developed concepts provide any practical application. The current empirical study was undertaken to formulate a concrete automated data quality platform to assess the quality of incoming dataset and generate a quality label, score and comprehensive report. We utilize various datasets from healthdata.gov, opendata.nhs and Demographics and Health Surveys (DHS) Program to observe the variations in the quality score and formulate a label using Principal Component Analysis(PCA). The results of the current empirical study revealed a metric that encompasses nine quality ingredients, namely provenance, dataset characteristics, uniformity, metadata coupling, percentage of missing cells and duplicate rows, skewness of data, the ratio of inconsistencies of categorical columns, and correlation between these attributes. The study also provides an illustrative case study and validation of the metric following Mutation Testing approaches. This research study provides an automated platform which takes an incoming dataset and metadata to provide the DQ score, report and label. The results of this study would be useful to data scientists as the value of this quality label would instill confidence before deploying the data for his/her respective practical application.

Mohit Chandra, Manvith Reddy, Shradha Sehgal et al.

The COVID-19 pandemic has disrupted people's lives driving them to act in fear, anxiety, and anger, leading to worldwide racist events in the physical world and online social networks. Though there are works focusing on Sinophobia during the COVID-19 pandemic, less attention has been given to the recent surge in Islamophobia. A large number of positive cases arising out of the religious Tablighi Jamaat gathering has driven people towards forming anti-Muslim communities around hashtags like #coronajihad, #tablighijamaatvirus on Twitter. In addition to the online spaces, the rise in Islamophobia has also resulted in increased hate crimes in the real world. Hence, an investigation is required to create interventions. To the best of our knowledge, we present the first large-scale quantitative study linking Islamophobia with COVID-19. In this paper, we present CoronaBias dataset which focuses on anti-Muslim hate spanning four months, with over 410,990 tweets from 244,229 unique users. We use this dataset to perform longitudinal analysis. We find the relation between the trend on Twitter with the offline events that happened over time, measure the qualitative changes in the context associated with the Muslim community, and perform macro and micro topic analysis to find prevalent topics. We also explore the nature of the content, focusing on the toxicity of the URLs shared within the tweets present in the CoronaBias dataset. Apart from the content-based analysis, we focus on user analysis, revealing that the portrayal of religion as a symbol of patriotism played a crucial role in deciding how the Muslim community was perceived during the pandemic. Through these experiments, we reveal the existence of anti-Muslim rhetoric around COVID-19 in the Indian sub-continent.

Mohit Chandra, Dheeraj Pailla, Himanshu Bhatia et al.

The exponential rise of online social media has enabled the creation, distribution, and consumption of information at an unprecedented rate. However, it has also led to the burgeoning of various forms of online abuse. Increasing cases of online antisemitism have become one of the major concerns because of its socio-political consequences. Unlike other major forms of online abuse like racism, sexism, etc., online antisemitism has not been studied much from a machine learning perspective. To the best of our knowledge, we present the first work in the direction of automated multimodal detection of online antisemitism. The task poses multiple challenges that include extracting signals across multiple modalities, contextual references, and handling multiple aspects of antisemitism. Unfortunately, there does not exist any publicly available benchmark corpus for this critical task. Hence, we collect and label two datasets with 3,102 and 3,509 social media posts from Twitter and Gab respectively. Further, we present a multimodal deep learning system that detects the presence of antisemitic content and its specific antisemitism category using text and images from posts. We perform an extensive set of experiments on the two datasets to evaluate the efficacy of the proposed system. Finally, we also present a qualitative analysis of our study.

Yaman Kumar, Swati Aggarwal, Debanjan Mahata et al.

In the era of MOOCs, online exams are taken by millions of candidates, where scoring short answers is an integral part. It becomes intractable to evaluate them by human graders. Thus, a generic automated system capable of grading these responses should be designed and deployed. In this paper, we present a fast, scalable, and accurate approach towards automated Short Answer Scoring (SAS). We propose and explain the design and development of a system for SAS, namely AutoSAS. Given a question along with its graded samples, AutoSAS can learn to grade that prompt successfully. This paper further lays down the features such as lexical diversity, Word2Vec, prompt, and content overlap that plays a pivotal role in building our proposed model. We also present a methodology for indicating the factors responsible for scoring an answer. The trained model is evaluated on an extensively used public dataset, namely Automated Student Assessment Prize Short Answer Scoring (ASAP-SAS). AutoSAS shows state-of-the-art performance and achieves better results by over 8% in some of the question prompts as measured by Quadratic Weighted Kappa (QWK), showing performance comparable to humans.

Mohit Chandra, Ashwin Pathak, Eesha Dutta et al.

While extensive popularity of online social media platforms has made information dissemination faster, it has also resulted in widespread online abuse of different types like hate speech, offensive language, sexist and racist opinions, etc. Detection and curtailment of such abusive content is critical for avoiding its psychological impact on victim communities, and thereby preventing hate crimes. Previous works have focused on classifying user posts into various forms of abusive behavior. But there has hardly been any focus on estimating the severity of abuse and the target. In this paper, we present a first of the kind dataset with 7601 posts from Gab which looks at online abuse from the perspective of presence of abuse, severity and target of abusive behavior. We also propose a system to address these tasks, obtaining an accuracy of ~80% for abuse presence, ~82% for abuse target prediction, and ~65% for abuse severity prediction.

Saurabh Gupta, Siddhant Bhambri, Karan Dhingra et al.

Smart homes require every device inside them to be connected with each other at all times, which leads to a lot of power wastage on a daily basis. As the devices inside a smart home increase, it becomes difficult for the user to control or operate every individual device optimally. Therefore, users generally rely on power management systems for such optimization but often are not satisfied with the results. In this paper, we present a novel multi-objective reinforcement learning framework with two-fold objectives of minimizing power consumption and maximizing user satisfaction. The framework explores the trade-off between the two objectives and converges to a better power management policy when both objectives are considered while finding an optimal policy. We experiment on real-world smart home data, and show that the multi-objective approaches: i) establish trade-off between the two objectives, ii) achieve better combined user satisfaction and power consumption than single-objective approaches. We also show that the devices that are used regularly and have several fluctuations in device modes at regular intervals should be targeted for optimization, and the experiments on data from other smart homes fetch similar results, hence ensuring transfer-ability of the proposed framework.

Saurabh Gupta, Arun Balaji Buduru, Ponnurangam Kumaraguru

Generative Adversarial Network (GAN) and its variants have shown promising results in generating synthetic data. However, the issues with GANs are: (i) the learning happens around the training samples and the model often ends up remembering them, consequently, compromising the privacy of individual samples - this becomes a major concern when GANs are applied to training data including personally identifiable information, (ii) the randomness in generated data - there is no control over the specificity of generated samples. To address these issues, we propose imdpGAN - an information maximizing differentially private Generative Adversarial Network. It is an end-to-end framework that simultaneously achieves privacy protection and learns latent representations. With experiments on MNIST dataset, we show that imdpGAN preserves the privacy of the individual data point, and learns latent codes to control the specificity of the generated samples. We perform binary classification on digit pairs to show the utility versus privacy trade-off. The classification accuracy decreases as we increase privacy levels in the framework. We also experimentally show that the training process of imdpGAN is stable but experience a 10-fold time increase as compared with other GAN frameworks. Finally, we extend imdpGAN framework to CelebA dataset to show how the privacy and learned representations can be used to control the specificity of the output.

Gurpriya Kaur Bhatia, Shubham Gupta, Alpana Dubey et al.

Crowdsourcing is a process wherein an individual or an organisation utilizes the talent pool present over the Internet to accomplish their task. The existing crowdsourcing platforms and their reputation computation are centralised and hence prone to various attacks or malicious manipulation of the data by the central entity. A few distributed crowdsourcing platforms have been proposed but they lack a robust reputation mechanism. So we propose a decentralised crowdsourcing platform having an immutable reputation mechanism to tackle these problems. It is built on top of Ethereum network and does not require the user to trust a third party for a non malicious experience. It also utilizes IOTAs consensus mechanism which reduces the cost for task evaluation significantly.

Rohan Pandey, Vaibhav Gautam, Ridam Pal et al.

Background: The COVID-19 pandemic has uncovered the potential of digital misinformation in shaping the health of nations. The deluge of unverified information that spreads faster than the epidemic itself is an unprecedented phenomenon that has put millions of lives in danger. Mitigating this Infodemic requires strong health messaging systems that are engaging, vernacular, scalable, effective and continuously learn the new patterns of misinformation. Objective: We created WashKaro, a multi-pronged intervention for mitigating misinformation through conversational AI, machine translation and natural language processing. WashKaro provides the right information matched against WHO guidelines through AI, and delivers it in the right format in local languages. Methods: We theorize (i) an NLP based AI engine that could continuously incorporate user feedback to improve relevance of information, (ii) bite sized audio in the local language to improve penetrance in a country with skewed gender literacy ratios, and (iii) conversational but interactive AI engagement with users towards an increased health awareness in the community. Results: A total of 5026 people who downloaded the app during the study window, among those 1545 were active users. Our study shows that 3.4 times more females engaged with the App in Hindi as compared to males, the relevance of AI-filtered news content doubled within 45 days of continuous machine learning, and the prudence of integrated AI chatbot Satya increased thus proving the usefulness of an mHealth platform to mitigate health misinformation. Conclusion: We conclude that a multi-pronged machine learning application delivering vernacular bite-sized audios and conversational AI is an effective approach to mitigate health misinformation.

Saurabh Gupta, Asmit Kumar Singh, Arun Balaji Buduru et al.

Hashtags in online social media have become a way for users to build communities around topics, promote opinions, and categorize messages. In the political context, hashtags on Twitter are used by users to campaign for their parties, spread news, or to get followers and get a general idea by following a discussion built around a hashtag. In the past, researchers have studied certain types and specific properties of hashtags by utilizing a lot of data collected around hashtags. In this paper, we perform a large-scale empirical analysis of elections using only the hashtags shared on Twitter during the 2019 Lok Sabha elections in India. We study the trends and events unfolded on the ground, the latent topics to uncover representative hashtags and semantic similarity to relate hashtags with the election outcomes. We collect over 24 million hashtags to perform extensive experiments. First, we find the trending hashtags to cross-reference them with the tweets in our dataset to list down notable events. Second, we use Latent Dirichlet Allocation to find topic patterns in the dataset. In the end, we use skip-gram word embedding model to find semantically similar hashtags. We propose popularity and an influence metric to predict election outcomes using just the hashtags. Empirical results show that influence is a good measure to predict the election outcome.

Raghav Kapoor, Yaman Kumar, Kshitij Rajput et al.

In multilingual societies like the Indian subcontinent, use of code-switched languages is much popular and convenient for the users. In this paper, we study offense and abuse detection in the code-switched pair of Hindi and English (i.e. Hinglish), the pair that is the most spoken. The task is made difficult due to non-fixed grammar, vocabulary, semantics and spellings of Hinglish language. We apply transfer learning and make a LSTM based model for hate speech classification. This model surpasses the performance shown by the current best models to establish itself as the state-of-the-art in the unexplored domain of Hinglish offensive text classification.We also release our model and the embeddings trained for research purposes

Anupama Aggarwal, Bimal Viswanath, Saravana Kumar et al.

Several studies have been conducted on understanding third-party user tracking on the web. However, web trackers can only track users on sites where they are embedded by the publisher, thus obtaining a fragmented view of a user's online footprint. In this work, we investigate a different form of user tracking, where browser extensions are repurposed to capture the complete online activities of a user and communicate the collected sensitive information to a third-party domain. We conduct an empirical study of spying browser extensions on the Chrome Web Store. First, we present an in-depth analysis of the spying behavior of these extensions. We observe that these extensions steal a variety of sensitive user information, such as the complete browsing history (e.g., the sequence of web traversals), online social network (OSN) access tokens, IP address, and user geolocation. Second, we investigate the potential for automatically detecting spying extensions by applying machine learning schemes. We show that using a Recurrent Neural Network (RNN), the sequences of browser API calls can be a robust feature, outperforming hand-crafted features (used in prior work on malicious extensions) to detect spying extensions. Our RNN based detection scheme achieves a high precision (90.02%) and recall (93.31%) in detecting spying extensions.

Niharika Sachdeva, Ponnurangam Kumaraguru

Police agencies across the globe are increasingly using Online Social Media (OSM) to acquire intelligence and connect with citizens. Developed nations have well thought of strategies to use OSM for policing. However, developing nations like India are exploring and evolving OSM as a policing solution. India, in recent years, experienced many events where rumors and fake content on OSM instigated communal violence. In contrast to traditional media (e.g. television and print media) used by Indian police departments, OSM offers velocity, variety, veracity and large volume of information. These introduce new challenges for police like platforms selection, secure usage strategy, developing trust, handling offensive comments, and security / privacy implication of information shared through OSM. Success of police initiatives on OSM to maintain law and order depends both on their understanding of OSM and citizen's acceptance / participation on these platforms. This study provides multidimensional understanding of behavior, perceptions, interactions, and expectation regarding policing through OSM. First, we examined recent updates from four different police pages- Delhi, Bangalore, Uttar Pradesh and Chennai to comprehend various dimensions of police interaction with citizens on OSM. Second, we conducted 20 interviews with IPS officers (Indian Police Service) and 17 interviews with citizens to understand decision rationales and expectation gaps between two stakeholders (police and citizens); this was followed up with 445 policemen surveys and 204 citizen surveys. We also present differences between police expectations of Indian and police departments in developed countries.