Natural Language Research Articles

A framework for mitigating malicious RLHF feedback in LLM training using consensus based reward

Large Language models (LLMs) have demonstrated impressive capabilities in natural language processing and understanding. LLMs are being rapidly adopted in major industry sectors including mobile computing, healthcare, finance, government, and education driven by technology giants such as NVIDIA, OpenAI, Microsoft, Apple, Meta, Google, Broadcom, AMD, and IBM. However, due to the emerging nature of this technology, many security/privacy challenges remain unresolved that we must tackle before rolling out LLMs to critical applications (e.g. Healthcare, Legal). In this article, we focus on the Reinforcement Learning via Human Feedback (RLHF) process that is widely used for training LLMs giving them the human-like feel most applications value. The RLHF process involves employing human experts to generate feedback based on an LLM’s query-response pairs and using this feedback to then retrain (fine-tune) the model. However, RLHF can also expose the LLM to malicious feedback generated by one or more individuals in the process leading to degraded performance of the LLM and harmful responses. Most state-of-the-art (SOTA) solutions to this problem involve utilizing a KL-Divergence-based brute-force update-rejection approach that can render the whole RLHF process completely useless (model quality is not improved) in the presence of malicious entities in the process. We propose the COnsensus-Based RewArd framework (COBRA), a consensus-based technique that can effectively negate the malicious noise generated by a certain segment of the RLHF human-expert pool, leading to improved LLM training performance in a mixed-trust scenario. We have evaluated COBRA for two separate LLM use cases, Sentiment Analysis and Conversational Task. We have experimented with a wide range of LLM models (e.g. GPT-2 XL - 1.5B parameters). COBRA outperformed the standard unprotected reward generation scheme by for the generative conversational task and by for the sentiment analysis task. We have also quantitatively compared COBRA with Coste et al. and observed state-of-the-art performance, particularly when a lower number of reward models are used ( increased reward accuracy at ).

Proteomic Learning of Gamma-Aminobutyric Acid (GABA) Receptor-Mediated Anesthesia.

Anesthetics are crucial in surgical procedures and therapeutic interventions, but they come with side effects and varying levels of effectiveness, calling for novel anesthetic agents that offer more precise and controllable effects. Targeting Gamma-aminobutyric acid (GABA) receptors, the primary inhibitory receptors in the central nervous system, could enhance their inhibitory action, potentially reducing side effects while improving the potency of anesthetics. In this study, we introduce a proteomic learning of GABA receptor-mediated anesthesia based on 24 GABA receptor subtypes by considering over 4000 proteins in protein-protein interaction (PPI) networks and over 1.5 millions known binding compounds. We develop a corresponding drug-target interaction network to identify potential lead compounds for novel anesthetic design. To ensure robust proteomic learning predictions, we curated a data set comprising 136 targets from a pool of 980 targets within the PPI networks. We employed three machine learning algorithms, integrating advanced natural language processing (NLP) models such as pretrained transformers and autoencoder embeddings. Through a comprehensive screening process, we evaluated the side effects and repurposing potential of over 180,000 drug candidates targeting the GABRA5 receptor. Additionally, we assessed the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of these candidates to identify those with near-optimal characteristics. This approach also involved optimizing the structures of existing anesthetics. Our work presents an innovative strategy for the development of new anesthetic drugs, optimization of anesthetic use, and a deeper understanding of potential anesthesia-related side effects.

Large Language Model-Driven 3D Hyper-Realistic Interactive Intelligent Digital Human System

Digital technologies are undergoing comprehensive integration across diverse domains and processes of the human economy, politics, culture, society, and ecological civilization. This integration brings forth novel concepts, formats, and models. In the context of the accelerated convergence between the digital and physical worlds, a discreet yet momentous transformation is being steered by artificial intelligence generated content (AIGC). This transformative force quietly reshapes and potentially disrupts the established patterns of digital content production and consumption. Consequently, it holds the potential to significantly enhance the digital lives of individuals and stands as an indispensable impetus for the comprehensive transition towards a new era of digital civilization in the future. This paper presents our award-winning project, a large language model (LLM)-powered 3D hyper-realistic interactive digital human system that employs automatic speech recognition (ASR), natural language processing (NLP), and emotional text-to-speech (TTS) technologies. Our system is designed with a modular concept and client–server (C/S) distributed architecture that emphasizes the separation of components for scalable development and efficient progress. The paper also discusses the use of computer graphics (CG) and artificial intelligence (AI) in creating photorealistic 3D environments for meta humans, and explores potential applications for this technology.

Had Enough of Experts? Quantitative Knowledge Retrieval From Large Language Models

ABSTRACTLarge language models (LLMs) have been extensively studied for their ability to generate convincing natural language sequences; however, their utility for quantitative information retrieval is less well understood. Here, we explore the feasibility of LLMs as a mechanism for quantitative knowledge retrieval to aid two data analysis tasks: elicitation of prior distributions for Bayesian models and imputation of missing data. We introduce a framework that leverages LLMs to enhance Bayesian workflows by eliciting expert‐like prior knowledge and imputing missing data. Tested on diverse datasets, this approach can improve predictive accuracy and reduce data requirements, offering significant potential in healthcare, environmental science and engineering applications. We discuss the implications and challenges of treating LLMs as ‘experts’.

Retraction Note: Efficient diabetes mellitus prediction with grid based random forest classifier in association with natural language processing

Retraction Note: Efficient diabetes mellitus prediction with grid based random forest classifier in association with natural language processing

AI-Driven Assistive Technology for Speech Disabilities: Transforming Communication with Natural Language Processing Techniques

AI-Driven Assistive Technology for Speech Disabilities: Transforming Communication with Natural Language Processing Techniques

Abstractive text summarization from Amazon food review dataset using modified attention based Bi-LSTM autoencoder model

ABSTRACT Automatic Text Summarization (ATS) is extensively performed to obtain insights from long texts or documents through Natural Language Processing (NLP). Even though various DL and ML-based algorithms have been used to perform text summarization, there is a lack of extensive studies on abstractive text summarization. The attention mechanisms used in traditional text summarization methods generate a fixed length vector irrespective of input sequence length. Hence, this study enhances the summarization efficiency through the implementation of a modified attention mechanism to generate dynamic weights for the input sequence. The proposed Modified Attention-based Bi-directional LSTM with Auto Encoder (Matt BiLSTM-AE) model is implemented for abstractive text summarization. The Amazon food review dataset is loaded as an input dataset, which is pre-processed to remove stop words, punctuation, and special characters. The optimized weight adjustment is obtained through the dynamic attention mechanism in the proposed model. The performance of the proposed MAtt-Bi-LSTM-AE text summarization is evaluated through ROUGE, accuracy, and loss metrics. From performance analysis, the accuracy rate was found to be 74.6%, while the loss rate was found to be 1.1751.

Intelligent De-Identification of Medical Discharge Summaries Using Hybrid NLP Techniques

Medical discharge summaries are vital documents in healthcare, often containing Personally Identifiable Information (PII), raising concerns regarding privacy and regulatory compliance. This paper proposes a cutting-edge approach that utilizes intelligent data de-identification to address this challenge. This paper employs Natural Language Processing (NLP) techniques such as Named Entity Recognition (NER), a hybrid approach that integrates Machine Learning (ML) models, Regular Expressions (REGEX)-based recognizers, and extensive lists of names and addresses. The proposed method focuses on achieving a delicate balance between extracting valuable insights from data and safeguarding sensitive information. The evaluation against benchmarks demonstrates significant improvements in de-identification performance, particularly in discharge summaries. We present findings from our system’s evaluation of synthesized discharge summaries, the OntoNotes dataset, and the CoNLL-2003 dataset, demonstrating its effectiveness in anonymizing diverse medical text sources.

Public sentiment, stock and energy prices during the Russia-Ukraine war: global evidence

ABSTRACT This study uses public emotions shown through social media to identify public sentiment toward the current Russia-Ukraine war. Utilizing the development of natural language processing algorithms, this study tests the correlation between the public psychological factor and the fluctuation in financial markets and energy prices during the ongoing Russia-Ukraine war. This study emphasizes the public’s initial response to this event (from 1/2022 to 5/2022). It aims to evaluate the public sentiment on sudden shock instead of taking the incident comprehensively. The study results ascertain the public sentiment index contained from social media as a market indicator. During shock events such as the Russia-Ukraine war, public sentiment intensifies energy and financial asset price fluctuation, indicating that public psychology tends to be influenced by negative news and causes them to act accordingly, resulting in a sell-off in financial and energy markets.

BERT applications in natural language processing: a review

BERT (Bidirectional Encoder Representations from Transformers) has revolutionized Natural Language Processing (NLP) by significantly enhancing the capabilities of language models. This review study examines the complex nature of BERT, including its structure, utilization in different NLP tasks, and the further development of its design via modifications. The study thoroughly analyses the methodological aspects, conducting a comprehensive analysis of the planning process, the implemented procedures, and the criteria used to decide which data to include or exclude in the evaluation framework. In addition, the study thoroughly examines the influence of BERT on several NLP tasks, such as Sentence Boundary Detection, Tokenization, Grammatical Error Detection and Correction, Dependency Parsing, Named Entity Recognition, Part of Speech Tagging, Question Answering Systems, Machine Translation, Sentiment analysis, fake review detection and Cross-lingual transfer learning. The review study adds to the current literature by integrating ideas from multiple sources, explicitly emphasizing the problems and prospects in BERT-based models. The objective is to comprehensively comprehend BERT and its implementations, targeting both experienced researchers and novices in the domain of NLP. Consequently, the present study is expected to inspire more research endeavors, promote innovative adaptations of BERT, and deepen comprehension of its extensive capabilities in various NLP applications. The results presented in this research are anticipated to influence the advancement of future language models and add to the ongoing discourse on enhancing technology for understanding natural language.

Improving EFL speaking performance among undergraduate students with an AI-powered mobile app in after-class assignments: an empirical investigation

English speaking represents one of the most challenging competencies for EFL learners, mainly due to the limited opportunities for authentic practice, especially within monolingual contexts like China. However, the ubiquitous personal mobile devices (smartphones) and the advent of AI-powered mobile apps equipped with automatic speech recognition, natural language processing, and text-to-speech present novel solutions to overcome these hurdles. This study investigates the effects of an AI-powered mobile application (Liulishuo) on Chinese undergraduate EFL students’ speaking performance, using features such as automatic feedback, process-oriented monitoring, and tailored instructions. Employing a quasi-experimental design, this 10-week study was conducted at a Chinese university with two groups of participants. The control group (n = 31) engaged with WeChat for after-class assignments, whereas the experimental group (n = 32) supplemented WeChat use with Liulishuo, diversifying their educational tools. Results indicated that participants in the experimental group significantly outperformed those in the control group regarding overall speaking performance. More specifically, notable improvements were observed in pronunciation and fluency, while vocabulary and grammar enhancements were not statistically significant. These findings underscore the capacity of AI-powered mobile apps to enhance EFL speaking performance, signaling novel pathways for the evolution of language education.

A Review on Predictive Analytics for Early Disease Detection in Neonatal Healthcare using Artificial Intelligence

The early detection of diseases plays a crucial role in improving patient outcomes, reducing healthcare costs, and enabling timely interventions. In recent years, the integration of Artificial Intelligence (AI) and Predictive Analytics (PA) has emerged as a transformative approach in healthcare, offering significant advancements in detecting diseases at their earliest stages. This paper provides a comprehensive review of the application of AI-driven predictive analytics in early disease detection, focusing on various AI techniques such as machine learning (ML), deep learning (DL), natural language processing (NLP), and neural networks. These techniques have shown exceptional promise in identifying patterns and correlations within medical data—including electronic health records (EHRs), medical imaging, genetic data, and wearable devices—that can signal the onset of diseases before they become clinically evident. The paper discusses the effectiveness of AI-based predictive models in detecting a wide range of diseases, including cancer, cardiovascular diseases, diabetes, neurological disorders, neonatal conditions, and infectious diseases. Special attention is given to AI applications in neonatal healthcare, where early detection of conditions such as neonatal sepsis, respiratory distress syndrome, and congenital anomalies can significantly improve survival rates and long-term health outcomes. By leveraging large datasets and advanced algorithms, AI systems can provide accurate predictions, risk assessments, and personalized treatment plans, leading to improved early diagnosis and targeted interventions. However, the integration of AI in disease detection also presents challenges such as data privacy concerns, model interpretability, ethical issues, and the need for robust regulatory frameworks. Furthermore, the paper highlights key advancements in AI technologies that have contributed to the success of predictive analytics in healthcare, along with real-world applications, case studies, and examples of AI models that have been implemented in clinical settings. The limitations and potential solutions to these challenges are also examined, with an emphasis on the importance of high-quality, representative datasets and continuous collaboration between AI researchers, clinicians, and regulatory bodies. This review aims to provide a thorough understanding of the current landscape of AI-powered predictive analytics for early disease detection and to highlight future directions in the field. As AI technologies continue to evolve, their role in enhancing early disease detection, particularly in neonatal care, improving patient outcomes, and enabling preventive healthcare will become increasingly significant, ultimately leading to a more efficient, effective, and equitable healthcare system.

A Survey on Artificial Intelligence in Precision Medicine and Healthcare Analysis for Neonatal Surgery

Artificial Intelligence (AI) has emerged as a transformative force in precision medicine, healthcare analysis, and neonatal surgery, enabling personalized treatment, early disease detection, and optimized clinical decision-making. This survey explores the evolving role of AI in healthcare, focusing on its applications, challenges, and future prospects. AI-driven approaches, including machine learning (ML) and deep learning (DL), have demonstrated remarkable accuracy in medical imaging, genomics, drug discovery, neonatal diagnostics, and patient risk assessment. These technologies enhance diagnostic precision, facilitate predictive analytics, and support real-time monitoring of chronic diseases and neonatal conditions. Precision medicine, which tailors treatments based on an individual’s genetic, environmental, and lifestyle factors, benefits significantly from AI-powered analytics. The integration of AI with electronic health records (EHRs), wearable devices, and biomedical data accelerates early disease identification and personalized therapeutic strategies, including those crucial for neonatal care and surgery. AI models trained on vast healthcare datasets can predict disease progression, recommend targeted therapies, and improve patient outcomes. Furthermore, natural language processing (NLP) enhances clinical documentation, reducing administrative burdens and improving efficiency in healthcare systems. Despite its potential, AI in precision medicine and neonatal surgery faces challenges, including data privacy concerns, model interpretability, and regulatory compliance. Ethical considerations, such as bias in AI models and equitable access to AI-driven healthcare, must be addressed to ensure responsible implementation. Additionally, integrating AI with traditional clinical workflows requires collaboration between healthcare professionals, data scientists, and policymakers. This survey provides a comprehensive analysis of AI applications in precision medicine, healthcare analysis, and neonatal surgery, highlighting key advancements, challenges, and future research directions. As AI continues to evolve, its role in revolutionizing healthcare will expand, paving the way for more efficient, accurate, and patient-centric medical practices. The findings of this survey aim to guide researchers, clinicians, and policymakers in leveraging AI for the next generation of precision healthcare, particularly in neonatal surgical interventions.

GFSNet: Gaussian Fourier with sparse attention network for visual question answering

Visual question answering (VQA), a core task in multimodal learning, requires models to effectively integrate visual and natural language information to perform reasoning and semantic understanding in complex scenarios. However, self-attention mechanisms often struggle to capture multi-scale information and key region features within images comprehensively. Moreover, VQA involves multidimensional and deep reasoning about image content, particularly in scenarios involving spatial relationships and frequency-domain features. Existing methods face limitations in modeling multi-scale features and filtering irrelevant information effectively. This paper proposes an innovative Gaussian Fourier with Sparse Attention Network (GFSNet) to address these challenges. GFSNet leverages Fourier transforms to map image attention weights generated by the self-attention mechanism from the spatial domain to the frequency domain, enabling comprehensive modeling of multi-scale frequency information. This enhances the model’s adaptability to complex structures and its capacity for relational modeling. To further improve feature robustness, a Gaussian filter is introduced to suppress high-frequency noise in the frequency domain, preserving critical visual information. Additionally, a sparse attention mechanism dynamically selects optimized frequency-domain features, effectively reducing interference from redundant information while improving interpretability and computational efficiency. Without increasing parameter counts or computational complexity, GFSNet achieves efficient modeling of multi-scale visual information. Experimental results on benchmark VQA datasets (VQA v2, GQA, and CLEVR) demonstrate that GFSNet significantly enhances reasoning capabilities and cross-modal alignment performance, validating its superiority and effectiveness. The code is available at https://github.com/shenxiang-vqa/GFSNet.

Ensemble learning based on bi-directional gated recurrent unit and convolutional neural network with word embedding module for bioactive peptide prediction.

Ensemble learning based on bi-directional gated recurrent unit and convolutional neural network with word embedding module for bioactive peptide prediction.

Determinants of Household Debt: A Systematic Review of the Literature

Household debt plays a crucial role in influencing the performance of the economy and its vulnerability to shocks. This study synthesises studies on this topic. We used the systematic literature review technique to analyse 97 studies from the EBSCO, Google Scholars, Sabinet African Journals, SAGE, ScienceDirect, Scopus, Springer Link, Taylor and Francis, and Web of Science databases from 2004 to 2024. The results reveal that country-level and household-specific factors are important in explaining debt. We delineate supply-side and demand factors that drive debt. Country-level factors that explain indebtedness include housing prices, unemployment, economic growth, interest rates, inflation, and income levels. At the household level, factors such as education level, behaviour, social status, employment, house prices, income, family size, occupation, wealth, and marital status explain indebtedness. Policies impact loan types differently; therefore, due consideration should be taken and prescriptions should aim to address both supply and demand factors. Future studies should rely on AI-driven methods and techniques that utilise natural language processing capabilities such as sentiment analysis in order to handle large data sets and ensure objectivity.

Understanding Loneliness Through Analysis of Twitter and Reddit Data: Comparative Study.

Loneliness is a global public health issue contributing to a variety of mental and physical health issues. It increases the risk of life-threatening conditions and contributes to theburden on the economy in terms of the number of productive days lost. Loneliness is a highly varied concept, which is associated with multiple factors. This study aimed to understand loneliness through a comparative analysis of loneliness data on Twitter and Reddit, which are popular social media platforms. These platforms differ in terms of their use, as Twitter allows only short posts, while Reddit allows long posts in a forum setting. We collected global data on loneliness in October 2022. Twitter posts containing the words "lonely," "loneliness," "alone," "solitude," and "isolation" were collected. Reddit posts were extracted in March 2023. Using natural language processing techniques (valence aware dictionary for sentiment reasoning [VADER] tool from the natural language toolkit [NLTK]), the study identified and extracted relevant keywords and phrases related to loneliness from user-generated content on both platforms. The study used both sentiment analysis and the number of occurrences of a topic. Quantitative analysis was performed to determine the number of occurrences of a topic in tweets and posts, and overall meaningful topics were reported under a category. The extracted data were subjected to comparative analysis to identify common themes and trends related to loneliness across Twitter and Reddit. A total of 100,000 collected tweets and 10,000 unique Reddit posts, including comments, were analyzed. The results of the study revealed the relationships of various social, political, and personal-emotional themes with the expression of loneliness on social media. Both platforms showed similar patterns in terms of themes and categories of discussion in conjunction with loneliness-related content. Both Reddit and Twitter addressed loneliness, but they differed in terms of focus. Reddit discussions were predominantly centered on personal-emotional themes, with a higher occurrence of these topics. Twitter, while still emphasizing personal-emotional themes, included a broader range of categories. Both platforms aligned with psychological linguistic features related to the self-expression of mental health issues. The key difference was in the range of topics, with Twitter having a wider variety of topics and Reddit having more focus on personal-emotional aspects. Reddit posts provide detailed insights into data about the expression of loneliness, although at the cost of the diversity of themes and categories, which can be inferred from the data. These insights can guide future research using social media data to understand loneliness. The findings provide the basis for further comparative investigation of the expression of loneliness on different social media platforms and online platforms.

Enhancing model-based development with formalized requirements: integrating temporal logic and SysML v2 for comprehensive state and transition modeling

Modern vehicles are complex cyber-physical systems with numerous customer functions and operating states, requiring careful management during development. Ensuring that all potential states and transitions are rigorously defined at the requirement stage is critical for vehicle safety and reliability. Due to stringent regulatory requirements and competitive market pressures, it is essential to articulate these requirements unambiguously, free of redundancy, and correctly.However, even a formally correct set of requirements can be underspecified, leading to unintended operating states that compromise vehicle functionality. To ensure requirement completeness, various approaches have been proposed to generate operating states and transitions allowed by a given set of requirements, ensuring that only intended behaviors are permitted. Existing methodologies often rely on natural language requirements, prone to ambiguities and errors. Furthermore, several methods formalize requirements using temporal logic languages, which can be checked algorithmically.Because of the complexity of modern vehicles, Model Based Systems Engineering (MBSE) has become state of the art in vehicle requirements engineering, as MBSE enables rigorous traceability of requirements and early model-based validation of requirement fulfillment.Currently a gap remains in integrating the comprehensive generation of operating states and transitions with the formalization in temporal logic languages, specifically for the use in MBSE development environments.This paper introduces a multi-step approach that formalizes requirements using Linear Temporal Logic (LTL) and generates operating states and transitions, which are then modeled in a SysML v2 state machine diagram, enhancing both the development process and the system’s reliability.

How to leverage large language models for automatic ICD coding.

How to leverage large language models for automatic ICD coding.

Intelligent cognitive health monitoring for seniors using natural language processing-integrated GRU attention and Q-reinforcement learning

Intelligent cognitive health monitoring for seniors using natural language processing-integrated GRU attention and Q-reinforcement learning