Journal of Data Science and Artificial Intelligence

EconVNNewsBot-2: Enhancing Economic News Retrieval with Retrieval-Augmented Thoughts for Improved Question-Answering in Vietnam

2024-12-28T01:33:16-05:00

The rapid expansion of Vietnamese economic news presents challenges for investors and entrepreneurs seeking accurate, domain-specific information. Traditional search engines struggle with complex queries, revealing a gap in effective information retrieval. In our previous work (Paper ID: 22 in ATAC 2024), we introduced EconVNNewsBot, a specialized Question-Answering (QA) system for Vietnamese economic news, where we used Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT) reasoning separately. Although effective, these methods treated retrieval and reasoning as distinct steps, limiting system efficiency and coherence. This paper extends our prior work by employing a more advanced approach: Retrieval-Augmented Thoughts (RAT). RAT unifies retrieval and reasoning into a single integrated framework, providing more precise and context-sensitive responses with improved efficiency. Our new experiments show that the RAT-based approach enhances user engagement, reduces retrieval time, and simplifies the information-gathering process. These findings suggest that RAT can transform how economic news is accessed, benefiting both information seekers and news agencies. The research is supported by the unique EconVNNews dataset, compiled using the EconVNNewsCrawl tool developed by our team, ensuring the system’s robust performance in the economic domain.

Detecting Referee Whistle Sounds in Soccer Videos Using Machine Learning Models

2025-02-20T21:31:09-05:00

In this study, we propose a simple and efficient approach for detecting referee whistle sounds in soccer match footage. Our method combines a fixed-length sliding window with feature extraction based on the short-time fourier transform (STFT). Specifically, we use a 1-second sliding window with 50% overlap to divide the audio input into smaller segments, which are analyzed to identify whistles characterized by prominent frequency components in the range of 3300–4200 Hz. For classification, we employ well-known machine learning algorithms, including Naïve Bayes, Support Vector Machines, and Neural Networks. To improve system performance, we integrate a post-processing technique based on Power Spectral Density (PSD) and predefined thresholds to reduce misclassification. Experimental results demonstrate that our method is highly effective at detecting whistles during soccer matches.

Applying deep learning with VGG16 to classify durian leaf diseases

2025-04-17T21:13:06-04:00

Durian is a high-value tropical fruit widely cultivated in Southeast Asia, particularly in Vietnam, Thailand, Malaysia, and Indonesia. However, durian trees are vulnerable to numerous leaf diseases that can significantly reduce yield and fruit quality. In this study, we propose an artificial intelligence-based approach using the VGG16 convolutional neural network to classify diseases on durian leaves. A dataset of 5,603 images representing 10 common diseases and 1 non-disease class was collected and augmented to enhance model generalization. The model achieved a classification accuracy of 94.13% and a loss value of 0.5419, demonstrating its effectiveness compared to ResNet-50. Furthermore, a web-based application was developed to allow farmers to upload leaf images for instant disease diagnosis and receive treatment recommendations. This approach provides a practical, accessible tool to support early disease detection and precision agriculture, contributing to improved productivity and sustainable farming practices.

Efficient Vietnamese Name Retrieval using Highly Discriminative N-Grams

2025-03-08T00:49:22-05:00

Retrieving Vietnamese names from large global databases is crucial for fostering connections within professional Vietnamese communities. However, the use of Latin characters in Vietnamese names often causes ambiguity, as they can resemble names from other countries. This paper introduces Highly Discriminative N-grams (HDNs), a novel query method designed to efficiently retrieve Vietnamese names from diverse datasets. Experimental results show that HDNs significantly outperform traditional unigram queries, achieving superior precision, recall, and cost-effectiveness. This innovative approach improves the accuracy and efficiency of Vietnamese name retrieval, supporting efforts to connect the Vietnamese diaspora with global opportunities.

ASPS - AI-Driven Smart Parking Solutions with Integrated Web and Mobile Applications

2024-12-18T09:34:31-05:00

The research suggests a smart parking system using the combination of AIoT and Edge Computing, in there, Raspberry Pi
collects the data from cameras and sensors and then transfers it to the local computer for AI to process and overcome the
performance limitations of the Raspberry Pi and create the comprehensive parking system including the website for admin
and mobile application for users. Previous research on smart parking systems has evolved both simple solutions using
sensors to advanced systems that integrate AI and cameras (dual cameras system with 96.74% accuracy and fisheye camera
system with YOLOv5 reached 94.3%), however, there are still some restrictions covering brightness, weather conditions,
etc. The key features include license plate detection using YOLOv10, displaying superior results than others. This smart
parking system solves the existing issues by leveraging edge computing to handle these AI tasks. Data collected from
cameras and sensors will be processed locally, ensuring real-time performance. The web platform allows remote
management, while the mobile app provides detailed information about the driver and also the online payment function.
Future enhancements cover the slot reservation function and AI-powered available space detection besides other factors
such as reducing traffic congestion, improving parking management, and decreasing pollution. The project will offer a
scalable solution for urban areas with high demand for vehicles like Ho Chi Minh City combined with low-cost
hardware/sensors and advanced AI models, from there can contribute to the development of a smart city with efficient and
user-friendly smart parking system.

Deep Learning-Based Recognition of Traditional Medicinal Leaves for Common Cold Treatment

2025-05-13T23:29:14-04:00

The use of traditional herbal remedies, particularly medicinal leaves, remains popular in Vietnam for treating common cold-related illnesses. However, identifying the correct leaf types and applying them appropriately poses challenges for non-expert users. This study proposes a web-based system utilizing artificial intelligence to assist in recognizing medicinal leaves commonly used in traditional treatments for colds. A dataset comprising 5,958 images of 10 different leaf types was constructed, labeled, preprocessed, and augmented to enhance model performance. A fine-tuned deep learning model based on the VGG16 convolutional neural network architecture was employed for classification and achieved an accuracy of 95.23% on the test set. The model was then integrated into a user-friendly web application that enables users to upload leaf images for recognition and receive detailed medicinal usage guidance. This system not only promotes the safe and informed use of traditional herbal medicine but also represents a meaningful step toward digitizing and preserving Vietnam’s rich ethnobotanical knowledge through artificial intelligence.

SYNERGY- Smart System for Home Device Control and User Interaction

2024-12-31T12:39:56-05:00

Smart home management is becoming increasingly important, requiring high-end systems that seamlessly integrate with users' needs and preferences. We introduce SYNERGY, an intelligent system to optimally support home environment management that can integrate smart home devices through context-aware actions and natural language interactions. By integrating NLP, machine learning, and the AI2-THOR virtual environment, SYNERGY detects users' intentions and emotional signals, enabling adaptive responses. Unlike systems limited by superficial empathy, SYNERGY facilitates deeper and more meaningful interactions. To effectively address task allocation, SYNERGY uses a reward-based mechanism that optimizes device usage and minimizes user effort. Our research demonstrates high accuracy in emotion and intent recognition, along with superior performance in task allocation and real-time adaptability. Evaluation confirms its effectiveness in improving user satisfaction by providing contextually relevant responses and seamless device control. Future work aims to incorporate more user-centric features, advancing SYNERGY as a practical and empathetic smart home solution.

An Intelligent Platform for Assessing Program Learning Outcomes - Leveraging Data Processing and AI-Driven Reporting

2025-03-06T08:43:17-05:00

The EVALLOS platform introduces a transformative approach to the assessment of Program Learning Outcomes (PLOs) and Course Learning Outcomes (CLOs) in higher education. Leveraging advanced data processing, artificial intelligence (AI), and dynamic visualization tools, the system provides a unified framework for evaluating educational outcomes. EVALLOS automates key processes, including data collection, CLO-PLO mapping, and report generation, ensuring precision, efficiency, and scalability. With features such as real-time dashboards, AI-generated insights, and actionable recommendations, the platform empowers educators to refine teaching strategies, address learning gaps, and align curriculum with institutional goals. Additionally, EVALLOS facilitates accreditation compliance by delivering comprehensive reports that integrate quantitative metrics with qualitative analysis. Through its innovative design, EVALLOS advances institutional decision-making, fosters continuous improvement, and sets a benchmark for AI applications in education.