Papers by Rajiv Ratn Shah

26 papers
An Empirical Investigation of Bias in the Multimodal Analysis of Financial Earnings Calls (2021.naacl-main)

Copied to clipboard

Challenge: Existing research focuses on textual elements of financial disclosures but ignores the rich acoustic features in the executives’ speech.
Approach: They propose to use a multimodal approach that leverages the verbal and vocal cues of speakers in financial disclosures to predict volatility and risk.
Outcome: The proposed models outperform existing models in the financial realm but still underrepresent the diverse communities spanning demographics, gender, and native speech.
Better and Worse with Scale: How Contextual Entrainment Diverges with Model Size (2026.findings-acl)

Copied to clipboard

Challenge: Larger language models become better and worse at handling contextual information . et al. (2017) formalized contextual entrainment as a tendency to favor tokens in context .
Approach: They formalize the first scaling laws for contextual entrainment . they find large models are four times more resistant to counterfactual misinformation .
Outcome: The largest models are four times more resistant to counterfactual misinformation than the smallest, but twice as prone to copying arbitrary tokens.
Med-CoDE: Medical Critique based Disagreement Evaluation Framework (2025.naacl-srw)

Copied to clipboard

Challenge: Existing evaluation methods for large language models lack robustness and accuracy in medical contexts.
Approach: They propose an evaluation framework for medical LLMs that measures disagreement between model-generated responses and established medical ground truths.
Outcome: The proposed evaluation framework captures accuracy and reliability in medical settings.
Investigating Political Herd Mentality: A Community Sentiment Based Approach (P19-2)

Copied to clipboard

Challenge: polarities inherent in political speeches and debates pose an important problem today.
Approach: They propose to use community-based graphs to augment hand-crafted features based on topic modeling and emotion detection on debate transcripts.
Outcome: The proposed approach surpasses the benchmark results on the same dataset.
A Preliminary Exploration of GANs for Keyphrase Generation (2020.emnlp-main)

Copied to clipboard

Challenge: Existing studies on extractive keyphrases have shown promising results, but the results suggest that there is room for improvement.
Approach: They propose a new keyphrase generation approach using Generative Adversarial Networks (GANs) their model produces a sequence of keyphrases and a discriminator distinguishes between human-curated and machine-generated keyphrase.
Outcome: The proposed model outperforms the state-of-the-art generative models on benchmark datasets and is comparable to the best performing extractive models.
Multilingual Coreference Resolution in Low-resource South Asian Languages (2024.lrec-main)

Copied to clipboard

Challenge: Existing coreference resolution models for South Asian languages are limited . a a sanity check for the prediction of translations is required to ensure accuracy of the model, authors say .
Approach: They evaluate an end-to-end coreference resolution model on a Hindi golden set . they use translation and word-alignment tools to translate a translated dataset into 31 languages .
Outcome: The proposed model scored 64 and 68 on a Hindi golden set.
An Annotated Dataset of Discourse Modes in Hindi Stories (2020.lrec-1)

Copied to clipboard

Challenge: Using a new corpus of sentences from Hindi short stories, we analyze the annotations for five different discourse modes argumentative, narrative, descriptive, dialogic and informative.
Approach: They propose to annotate sentences from Hindi short stories for five different discourse modes argumentative, narrative, descriptive, dialogic and informative.
Outcome: The proposed corpus has a high inter-annotator agreement (0.87 k-alpha) and is able to capture the nuances of the embedded discourse structures.
A Video Is Worth 4096 Tokens: Verbalize Videos To Understand Them In Zero Shot (2023.emnlp-main)

Copied to clipboard

Challenge: Existing annotated training datasets hinder development of supervised learning models for multimedia content . lack of annotating benchmarks hinders development of models with satisfactory performance . a recent study shows that large language models have zero-shot performance in multimedia understanding .
Approach: They propose to verbalize long videos to generate their descriptions in natural language . they then perform video-understanding tasks on the generated story as opposed to the original video .
Outcome: The proposed method achieves better results than baselines for video understanding.
GupShup: Summarizing Open-Domain Code-Switched Conversations (2021.emnlp-main)

Copied to clipboard

Challenge: Abstractive summarization is the process of generating a condensed version of a given conversation while preserving the most salient aspects.
Approach: They propose to use a dataset to analyze code-switched conversations in Hindi and English to summarize them.
Outcome: The proposed dataset contains over 6,800 code-switched conversations and their corresponding human-annotated summaries in English (En) and Hi-En.
A Time-Aware Transformer Based Model for Suicide Ideation Detection on Social Media (2020.emnlp-main)

Copied to clipboard

Challenge: Suicide ideation is often linked to a history of mental depression.
Approach: They propose a time-aware transformer based model for preliminary screening of suicidal risk on social media that augments linguistic models with historical context.
Outcome: The proposed model outperforms competing models and shows that it is time-aware and contextually useful for suicide risk assessment.
Multitask Learning for Emotionally Analyzing Sexual Abuse Disclosures (2021.naacl-main)

Copied to clipboard

Challenge: Prior work on identifying narratives related to sexual abuse disclosures did not consider this as an independent task.
Approach: They propose to identify narratives related to sexual abuse disclosures as a joint modeling task that leverages their emotional attributes through multitask learning.
Outcome: The proposed model leverages emotional attributes of textual conversations to identify narratives related to sexual abuse disclosures in homogeneous and heterogeneously settings.
#YouToo? Detection of Personal Recollections of Sexual Harassment on Social Media (P19-1)

Copied to clipboard

Challenge: a recent study has found that the disclosure of sexual abuse has positive psychological im- pacts.
Approach: They propose to aggregate personal experiences of sexual harassment from Twitter posts to facilitate a better understanding of social media constructs and bring about social change.
Outcome: The proposed model is compared with state-of-the-art models and is based on a three part Twitter-Specific Social Media Language Model.
VolTAGE: Volatility Forecasting via Text Audio Fusion with Graph Convolution Networks for Earnings Calls (2020.emnlp-main)

Copied to clipboard

Challenge: Existing approaches to stock volatility forecasting ignore correlations between stocks.
Approach: They propose to combine vocal cues with verbal and financial cue data to create a multimodal stock volatility prediction model that accounts for stock interdependence via graph convolutions.
Outcome: The proposed model outperforms existing methods showing that it can predict volatility using multimodal learning.
Suicide Ideation Detection via Social and Temporal User Representations using Hyperbolic Learning (2021.naacl-main)

Copied to clipboard

Challenge: Recent studies indicate that individuals exhibiting suicidal ideation increasingly turn to social media rather than mental health practitioners.
Approach: They propose a framework leveraging a user’s emotional history and social information from a users neighborhood in a network to contextualize the interpretation of the latest tweet of a Twitter user.
Outcome: The proposed framework outperforms state-of-the-art methods on this task, showing the benefits of both socially and personally contextualized representations.
Quantitative Day Trading from Natural Language using Reinforcement Learning (2021.naacl-main)

Copied to clipboard

Challenge: Existing approaches to stock prediction are not optimized to make profitable investment decisions.
Approach: They propose a deep reinforcement learning approach that makes time-aware decisions to trade stocks while optimizing profit using textual data.
Outcome: The proposed method outperforms state-of-the-art in terms of risk-adjusted returns on two benchmarks: Tweets (English) and financial news (Chinese)
Augmenting NLP models using Latent Feature Interpolations (2020.coling-main)

Copied to clipboard

Challenge: Existing data augmentation methods with a large number of parameters are prone to over-fitting and often fail to capture the underlying input distribution.
Approach: They propose a data augmentation technique that uses embeddings and hidden layer representations to construct virtual examples.
Outcome: The proposed method outperforms existing methods in terms of accuracy and robustness to weight pruning.
Two-Step Classification using Recasted Data for Low Resource Settings (2020.aacl-main)

Copied to clipboard

Challenge: Existing studies on NLP models focus on high resource languages like English, but there are only two datasets for Hindi.
Approach: They propose a novel two-step classification method which uses textual-entailment predictions for classification task.
Outcome: The proposed method improves classification performance by using a joint-objective for classification and textual entailment.
Multimodal Multi-Speaker Merger & Acquisition Financial Modeling: A New Task, Dataset, and Neural Baselines (2021.acl-long)

Copied to clipboard

Challenge: Merger and acquisition (M&A) calls provide key insights into claims made by company executives about restructuring of financial firms.
Approach: They propose a baseline architecture that leverages multimodal multi-speaker input to forecast financial risk associated with M&A calls.
Outcome: The proposed model performs marginally better than existing models based on BERT inputs . the proposed model is expected to be validated by the end of the year .
PHASE: Learning Emotional Phase-aware Representations for Suicide Ideation Detection on Social Media (2021.eacl-main)

Copied to clipboard

Challenge: Recent studies indicate that individuals exhibiting suicidal ideation increasingly turn to social media rather than mental health practitioners.
Approach: They propose a time-and-phase-aware framework that adaptively learns features from a user’s historical emotional spectrum to contextualize suicidal intent.
Outcome: The proposed framework outperforms state-of-the-art methods while outperforming existing methods.
Unsupervised Paraphasia Classification in Aphasic Speech (2020.acl-srw)

Copied to clipboard

Challenge: Aphasia is a speech and language disorder which results from brain damage resulting in word retrieval deficit (anomia) . supervised learning methods cant be properly utilized as there is no aphasic speech data.
Approach: They propose an unsupervised method which can be implemented without the need for labeled paraphasia data.
Outcome: The proposed method outperforms supervised learning methods and transfer learning approaches for English without labeled paraphasia data.
FAST: Financial News and Tweet Based Time Aware Network for Stock Trading (2021.eacl-main)

Copied to clipboard

Challenge: Existing methods for stock movement prediction are limited and do not account for the fine-grain temporal irregularities in the release of large volumes of text.
Approach: They propose a hierarchical, learning to rank approach that uses textual data to make time-aware predictions for ranking stocks based on expected profit.
Outcome: The proposed method outperforms state-of-the-art methods by over 8% in terms of cumulative profit and risk-adjusted returns on two benchmarks: English tweets and Chinese financial news spanning two major stock indexes and four global markets.
Deep Attentive Learning for Stock Movement Prediction From Social Media Text and Company Correlations (2020.emnlp-main)

Copied to clipboard

Challenge: Existing models that predict stock movements are based on time series and technical analysis, but price signals alone fail to capture market surprises and impacts of sudden unexpected events.
Approach: They propose a model that integrates chaotic temporal signals from financial data and social media to create hierarchical temporal networks.
Outcome: The proposed model can be used to forecast stock movements on real-world S&P 500 index data and English tweets.
GPolS: A Contextual Graph-Based Language Model for Analyzing Parliamentary Debates and Political Cohesion (2020.coling-main)

Copied to clipboard

Challenge: Parliamentary debates are a valuable language resource for analyzing comprehensive options in a functional, free society.
Approach: They propose a neural model for political speech sentiment analysis exploiting semantic representations and relations between debate transcripts, motions, and political party members.
Outcome: The proposed model exploits semantic representations and relations between debate transcripts, motions, and political party members to predict political polarity and polarities.
Key2Vec: Automatic Ranked Keyphrase Extraction from Scientific Articles using Phrase Embeddings (N18-2)

Copied to clipboard

Challenge: Keyphrase extraction is a fundamental task in natural language processing that facilitates mapping of documents to a set of representative phrases.
Approach: They propose an unsupervised technique that leverages phrase embeddings for ranking keyphrases extracted from scientific articles using theme-weighted PageRank.
Outcome: The proposed method performs better on benchmark datasets than other methods and is of high quality.
IRIS: Interleaved Reinforcement with Incremental Staged Curriculum for Cross-Lingual Mathematical Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Curriculum learning fails to generate consistent step-by-step reasoning in multilingual and low-resource settings.
Approach: They propose a framework that combines supervised fine-tuning with reverse curriculum reinforcement learning to generate consistent step-by-step reasoning.
Outcome: The proposed framework outperforms single-axis benchmarks and multilingual test sets on math reasoning tasks and in high-resource languages.
ARHNet - Leveraging Community Interaction for Detection of Religious Hate Speech in Arabic (P19-2)

Copied to clipboard

Challenge: Existing methods to detect hate speech in Arabic rely on textual cues and social network graphs.
Approach: They propose to use Arabic word embeddings and social network graphs to profile hate speech in Arabic.
Outcome: The proposed model incorporates Arabic Word Embeddings and Social Network Graphs for the detection of religious hate speech in Arabic.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations