Articles

Right in the context of digital marketing, this paper presents a comprehensive integrated system that combines the latest artificial intelligence advancement – large language models and diffusion models – to generate marketing email subjects and content that result in higher engagement. The system uses finetuned large language models for compelling email subject generation and finetuned Stable Diffusion model for visually appealing and convincing email content images creation. For the latter, both knowledge graphs and vector embeddings have been incorporated to improve contextual relevance. Experimental results demonstrated significant improvement in all engagement metrics that marketers rely on, including 46% growth in open rates, 56% higher click-through rates, and an 51% boost in conversion rates, all compared to human generated content. The unified approach presented by this paper outperforms standalone models and human-generated content in terms of engagement, as the comparative analysis shows. We also discuss the ethical considerations related to content bias and personalization boundaries, alongside challenges faced in this type of projects, such as computation demands and probable solutions. Finally, this paper proposes future directions to be taken, including expansion to other digital marketing channels, the use of other advanced artificial intelligence techniques, and the development of real-time content adaptation mechanisms based on user feedback.

The rapid advancements in artificial intelligence (AI) and robotics have paved the way for innovative autonomous systems capable of performing complex tasks. This project integrates robotics with Large Language Models (LLMs) to develop an intelligent, versatile and user-friendly robotic system. The robot is designed to interpret structured commands, make real-time decisions, and navigate autonomously in dynamic environments, addressing key challenges faced by traditional autonomous systems. Central to the system is a Raspberry Pi 4, which serves as the main processing unit, integrating components such as a webcam for visual data capture, an L298N motor driver for motor control, and a Bluetooth speaker for real-time feedback. The LLM API enables the robot to process natural language commands, providing context-aware task execution and adaptability to changing scenarios. Testing has demonstrated the system’s ability to perform autonomous navigation, detect obstacles, and execute tasks effectively. This research offers a foundation for various industries, including logistics, healthcare, education, and hazardous environment operations. By incorporating LLMs the robot overcomes limitations of traditional rule-based systems, enhancing dynamic decision-making and user interaction. With its modular design and scalability, it bridges the gap between human-like intelligence and mechanical precision, setting the stage for future advancements in AI-driven robotics.

Access to fast and efficient information is crucial in today's digital era, especially for teenagers in obtaining mental health services. The manual method used by Youth Information and Counselling Centre (PIK R) to provide mental health information requires significant time and effort. This research presents an AI-based solution by developing a chatbot system using retrieval-augmented generation (RAG) and large language models (LLM). This chatbot is designed to provide accurate and effective mental health information for teenagers throughout the day. An analysis of a dataset consisting of articles on teenage mental health and data from the Alodokter website was used as the basis for the development of this chatbot. The research results show that the chatbot is capable of providing relevant and accurate information, with evaluations using the recall-oriented understudy for gisting evaluation (ROUGE) score method yielding an average of ROUGE-1 with a precision of 87.8%, recall of 83.0%, and F1-measure of 84.0%; ROUGE-2 with a precision of 82.8%, recall of 76.8%, and F1-measure of 78.2%; and ROUGE-L with a precision of 88.0%, recall of 82.6%, and F1-measure of 83.4%. These findings indicate the potential use of chatbots as an effective tool to support the mental health of adolescents.

In recent times, remarkable performance has been shown by large language models (LLMs) in a range of natural language processing (NLP) such as questioning, responding, document production, and translating languages. In today's competitive business landscape, understanding consumer behaviour in online buying is crucial for the success of e-commerce platforms. The work proposes a novel Five-Tier service-oriented BI architecture (FSOBIA) that leverages advanced tuned decision tree (ATDT) techniques for predicting online buying behaviour. The proposed FSOBIA offers e-commerce platforms a scalable and adaptable solution for gaining insights into consumer preferences and making informed business decisions. The goal of FSOBIA's design and implementation is to meet the needs of evolving users and quicker service. Experimental evaluations on real-world datasets in FSOBIA achieved over 95% prediction accuracy, outperforming traditional models: Decision trees (82%), and XGBoost (91%), while offering better scalability and computational efficiency.

Sentiment analysis can be a powerful tool in safeguarding public health. This allows authorities to investigate and take action before a foodborne illness outbreak spreads. This paper introduces a novel system that proactively empowers restaurants to identify potential food safety hazards and hygiene regulation violations. The system leverages the power of natural language processing (NLP) to analyze Arabic restaurant reviews left by customers. By fine-tuning a pre-trained BERT mini-Arabic model on three targeted datasets: Sentiment Twitter Corpus, an Algerian dialect dataset, and an Arabic restaurant dataset, the system achieves an impressive accuracy of 91%. Additionally, the system caters to spoken feedback by accepting audio reviews. We utilized Whisper AI for accurate text transcription, followed by classification using a fine-tuned Gemini model from Google on Algerian local comments and others generated using large language models (LLMs) through few-shot learning techniques, reaching an accuracy of 93%. Notably, both models operate independently and concurrently. Leveraging RESTful APIs, the system integrates the solved sub-solutions from each microservice into a fusion layer for a comprehensive restaurant evaluation. This multifaceted approach delivers remarkable results for both modern standard Arabic (MSA) and the Algerian dialect, demonstrating its effectiveness in addressing restaurant food safety concerns.

Large language models (LLMs) are ubiquitous today with major usage in the fields of industry, research, and academia. LLMs involve unsupervised learning with large natural language data, obtained mostly from the internet. There are several challenges that arise because of these data sources. One such challenge is with respect to domain-specific knowledge and context. This paper deals with the major challenges faced by LLMs due to data sources, such as, lack of domain expertise, understanding specialized terminology, contextual understanding, data bias, and the limitations of transfer learning. This paper also discusses some solutions for the mitigation of these challenges such as pre-training LLMs on domain-specific corpora, expert annotations, improving transformer models with enhanced attention mechanisms, memory-augmented models, context-aware loss functions, balanced datasets, and the use of knowledge distillation techniques.

This study explores the development and implementation of a retrieval-augmented generation (RAG) system using the large language model (LLM) to enhance the learning of hadith through a chat interface for high school students. This study addresses challenges in optimizing RAG configurations and problems associated with traditional educational methods that lack interactivity. In addition, the RAG system was designed to replace real teacher interactions, offering a chat feature that provides contextual answers to real-life scenarios related to Hadith. Various configurations were tested, with a focus on the Matn component, achieving a high accuracy score with a mean of .754 and demonstrating efficiency in context relevance with a mean of .797. Results indicated significant accessibility using our RAG system for learning hadith via WhatsApp’s chat interface. Hence, this study highlights the potential of RAG systems in transforming educational environments and offers insights into the development of technology for interactive Hadith learning solutions.

This study explores text summarization, a critical component of natural language processing (NLP), specifically targeting scientific documents. Traditional extractive summarization, which relies on the original wording, often results in disjointed sequences of sentences and fails to convey key ideas concisely. To address these issues and ensure comprehensive inclusion of relevant details, our research aims to improve the coherence and completeness of summaries. We employed 25 different large language models (LLMs) to evaluate their performance in generating abstractive summaries of scholarly scientific documents. A recall-oriented evaluation of the generated summaries revealed that LLMs such as 'Claude v2.1,' 'PPLX 70B Online,' and 'Mistral 7B Instruct' demonstrated exceptional performance with ROUGE-1 scores of 0.92, 0.88, and 0.85, respectively, supported by high precision and recall values from bidirectional encoder representations from transformers (BERT) scores (0.902, 0.894, and 0.888). These findings offer valuable insights for NLP researchers, laying the foundation for future advancements in LLMs for summarization. The study highlights potential improvements in text summarization techniques, benefiting various NLP applications.

This study aims to examine teachers’ perceptions of the effectiveness of large language models (LLM) in supporting data-driven decision-making in educational contexts. Specifically, the study explores the influence of technological pedagogical knowledge, technological content knowledge, and technological pedagogical content knowledge on teachers’ utilization of LLMs for informed decision-making. Additionally, it investigates the moderating role of ethical considerations in the use of LLMs. A survey-based methodology was employed to collect data from university teachers in Chengdu, Sichuan, China, resulting in a sample of 319 respondents, which was analyzed using Smart PLS 4. The findings indicate that technological pedagogical knowledge, technological content knowledge, and technological pedagogical content knowledge for LLM use significantly impact data-driven decision-making in teaching. Moreover, ethical considerations were found to significantly moderate the relationship between these knowledge domains and data-driven decision-making. This study contributes novel insights by addressing the evaluation and application of LLM effectiveness from teachers’ perspectives, ultimately fostering the advancement of quality education.

Recent advancements in large language models (LLMs) have led to opportunities for improving applications across various domains. However, existing LLMs fine-tuned for Swahili or other African languages often rely on pre-trained multilingual models, resulting in a relatively small portion of training data dedicated to Swahili. In this study, we compare the performance of monolingual and multilingual models in Swahili natural language inference tasks using the cross-lingual natural language inference (XNLI) dataset. Our research demonstrates the superior effectiveness of dedicated Swahili monolingual models, achieving an accuracy rate of 69%. These monolingual models exhibit significantly enhanced precision, recall, and F1 scores, particularly in predicting contradiction and neutrality. Overall, the findings in this article emphasize the critical importance of using monolingual models in low-resource language processing contexts, providing valuable insights for developing more efficient and tailored natural language processing systems that benefit languages facing similar resource constraints.

Statistical machine translation (SMT) used parallel corpora and statistical models, to identify translation patterns and probabilities. Although this method had advantages, it had trouble with idiomatic expressions, context-specific subtleties, and intricate linguistic structures. The subsequent introduction of deep neural networks such as recurrent neural networks (RNNs), long short-term memory (LSTMs), transformers with attention mechanisms, and the emergence of large language model (LLM) frameworks has marked a paradigm shift in machine translation in recent years and has entirely replaced the traditional statistical approaches. The LLMs are able to capture complex language patterns, semantics, and context because they have been trained on enormous volumes of text data. Our study summarizes the most significant contributions in the literature related to LLM prompting, fine-tuning, retrieval augmented generation, improved transformer variants for faster translation, multilingual LLMs, and quality estimation with LLMs. This new research direction guides the development of more efficient and innovative solutions to address the current challenges of LLMs, including hallucinations, translation bias, information leakage, and inaccuracy due to language inconsistencies.

The advanced models of deep neural networks like bidirectional encoder from the transformers (BERT) and others, poses challenges in terms of computational resources and model size. In order to tackle these issues, techniques of model pruning have surfaced as the most useful methods in addressing the issues of model complexity. This research paper explores the concept of pruning BERT attention heads across the ensemble of winning tickets in order to enhance the efficiency of the model without sacrificing performance. Experimental evaluations show how effective the approach is, in achieving significant model compression while still maintaining competitive performance across different natural language processing tasks. The key findings of this study include model size that has been reduced by 36%, with our ensemble model reaching greater performance as compared to the baseline BERT model on both Stanford Sentiment Treebank v2 (SST-2) and Corpus of Linguistic Acceptability (CoLA) datasets. The results further show a F1-score of 94% and 96%, respectively, and accuracy scores of 95% and 96% on the two datasets. The findings of this research paper contribute to the ongoing efforts in enhancing the efficiency of large-scale language models.

In the evolving landscape of text generation, which has advanced rapidly in recent years, techniques for evaluating the performance and quality of the generated text lag behind relatively. Traditionally, lexical-based metrics such as bilingual evaluation understudy (BLEU), recall-oriented understudy for gisting evaluation (ROUGE), metric for evaluation of translation with explicit ordering (METEOR), consensus-based image description evaluation (CIDER), and F1 have been utilized, primarily relying on n-gram similarity for evaluation. In recent years, neural and machine-learning-based metrics, like bidirectional encoder representations from transformers (BERT) score, key phrase question answering (KPQA), and BERT supervised training of learned evaluation metric for reading comprehension (LERC) have shown superior performance over traditional metrics but suffered from a lack of generalization towards different domains and requires massive human-labeled training data. The main contribution of the current research is to investigate the use of train-free large language models (LLMs) as scoring metrics, evaluators, and judges within a questionanswering context, encompassing both closed and open-QA scenarios. To validate this idea, we employ a simple zero-shot prompting of Mixtral 8x7 B, a popular and widely used open-source LLM, to score a variety of datasets and domains. The experimental results on ten different benchmark datasets are compared against human judgments, revealing that, on average, simple LLMbased metrics outperformed sophisticated state-of-the-art statistical and neural machine-learning-based metrics by 2-8 points on answer-pairs scoring tasks and up to 15 points on contrastive preferential tasks.

This paper presents and discusses a framework that leverages large-scale language models (LLMs) for data enrichment and continuous monitoring emphasizing its essential role in optimizing the performance of deployed models. It introduces a comprehensive large language model operations (LLMOps) methodology based on continuous monitoring and continuous improvement of the data, the primary determinant of the model, in order to optimize the prediction of a given phenomenon. To this end, first we examine the use of real-time web scraping using tools such as Kafka and Spark Streaming for data acquisition and processing. In addition, we explore the integration of LLMOps for complete lifecycle management of machine learning models. Focusing on continuous monitoring and improvement, we highlight the importance of this approach for ensuring optimal performance of deployed models based on data and machine learning (ML) model monitoring. We also illustrate this methodology through a case study based on real data from several real estate listing sites, demonstrating how MLflow can be integrated into an LLMOps pipeline to guarantee complete development traceability, proactive detection of performance degradations and effective model lifecycle management.

This paper conducted a systematic review of Scopus-indexed publications on large language models (LLMs) and natural language processing (NLP) extracted in October 2023 to address the dearth of literature on their opportunities and challenges. Through bibliometric analysis, from the 1,600 relevant documents, the study explored research productivity, revealing both opportunities and challenges spanning research and real-world applications in education, medicine, and health care, citations, and keyword co-occurrence networks. Results highlighted distribution patterns and dominant players like Google LLC and Stanford University. Opportunities such as technological development in generative artificial intelligence (AI), were contrasted with challenges such as biases and ethical concerns. The intellectual structure analysis revealed prominent application areas in health and education and also emphasized issues such as AI divide and human-AI partnership. Improvement on the technology performance of LLM and NLP remains to be a challenge. Recommendations include further exploration of open research problems and bibliometric studies using other research databases given the research bias towards Scopus-indexed English publications.