Procedia Computer Science最新文献_第9页

Construction of Question Answering System Based on English Pre-Trained Language Model Enhanced by Knowledge Graph 基于知识图增强的英语预训练语言模型的问答系统构建

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.256

Pei Song

With the continuous development of technology, question answering systems based on pre trained language models have become an important component of intelligent applications. However, traditional question-answering systems often face challenges in terms of understanding accuracy and insufficient knowledge coverage when dealing with open-domain questions. To this end, this paper uses a method for building a question-answering system based on a knowledge graph-enhanced BERT pre-trained language model. First, this paper trains the basic model based on a large-scale BERT pre-trained language model. Then, this paper adopts knowledge graph technology to introduce structured knowledge information into the model by integrating domain-specific knowledge bases. Finally, in order to effectively integrate knowledge graph information, this paper uses graph neural network (GNN) to model graph data, and combines the self-attention mechanism in the BERT model to optimize the weighted fusion process of knowledge graph information. Experimental results show that the BERT model enhanced with knowledge graph performs well in multiple question-answering tasks. On the simple question of the first experiment, the average accuracy of the enhanced model increased from 84.6% of the standard BERT model to 89.8%, and the F1 score increased from 0.86 to 0.91. In complex reasoning tasks, the BERT+knowledge graph model demonstrates stronger reasoning ability and higher knowledge coverage. In the experimental conclusion, the introduction of the knowledge graph significantly improves the model’s reasoning ability and knowledge coverage, especially in professional field problems and multi-step reasoning tasks, the enhanced model shows stronger capabilities. This research provides a new construction method for question-answering systems, demonstrates the great potential of knowledge graphs in natural language processing, and has broad application prospects.

随着技术的不断发展，基于预训练语言模型的问答系统已经成为智能应用的重要组成部分。然而，传统的问答系统在处理开放领域问题时往往面临理解准确性和知识覆盖不足的挑战。为此，本文采用了一种基于知识图增强的BERT预训练语言模型构建问答系统的方法。首先，本文基于大规模BERT预训练语言模型对基本模型进行训练。然后，通过集成领域知识库，采用知识图技术将结构化的知识信息引入到模型中。最后，为了有效整合知识图信息，利用图神经网络（GNN）对图数据进行建模，并结合BERT模型中的自关注机制对知识图信息的加权融合过程进行优化。实验结果表明，经知识图增强的BERT模型在多个问答任务中表现良好。在第一个实验的简单问题上，增强模型的平均准确率从标准BERT模型的84.6%提高到89.8%，F1分数从0.86提高到0.91。在复杂的推理任务中，BERT+知识图模型表现出更强的推理能力和更高的知识覆盖率。在实验结论中，知识图的引入显著提高了模型的推理能力和知识覆盖率，特别是在专业领域问题和多步推理任务中，增强后的模型表现出更强的能力。本研究为问答系统的构建提供了一种新的方法，展示了知识图在自然语言处理中的巨大潜力，具有广阔的应用前景。

{"title":"Construction of Question Answering System Based on English Pre-Trained Language Model Enhanced by Knowledge Graph","authors":"Pei Song","doi":"10.1016/j.procs.2025.04.256","DOIUrl":"10.1016/j.procs.2025.04.256","url":null,"abstract":"<div><div>With the continuous development of technology, question answering systems based on pre trained language models have become an important component of intelligent applications. However, traditional question-answering systems often face challenges in terms of understanding accuracy and insufficient knowledge coverage when dealing with open-domain questions. To this end, this paper uses a method for building a question-answering system based on a knowledge graph-enhanced BERT pre-trained language model. First, this paper trains the basic model based on a large-scale BERT pre-trained language model. Then, this paper adopts knowledge graph technology to introduce structured knowledge information into the model by integrating domain-specific knowledge bases. Finally, in order to effectively integrate knowledge graph information, this paper uses graph neural network (GNN) to model graph data, and combines the self-attention mechanism in the BERT model to optimize the weighted fusion process of knowledge graph information. Experimental results show that the BERT model enhanced with knowledge graph performs well in multiple question-answering tasks. On the simple question of the first experiment, the average accuracy of the enhanced model increased from 84.6% of the standard BERT model to 89.8%, and the F1 score increased from 0.86 to 0.91. In complex reasoning tasks, the BERT+knowledge graph model demonstrates stronger reasoning ability and higher knowledge coverage. In the experimental conclusion, the introduction of the knowledge graph significantly improves the model’s reasoning ability and knowledge coverage, especially in professional field problems and multi-step reasoning tasks, the enhanced model shows stronger capabilities. This research provides a new construction method for question-answering systems, demonstrates the great potential of knowledge graphs in natural language processing, and has broad application prospects.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 647-655"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124807","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Test Scenario Design and Optimization of Automated Driving Lane Keeping System Based On PCA and Intelligent Algorithm 基于PCA和智能算法的自动驾驶车道保持系统测试场景设计与优化

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.194

Linfeng Hao

With the rapid development of autonomous driving technology, lane keeping system (LKS) has become an important function to improve vehicle driving safety. It effectively reduces the risk of accidents caused by lane departure by ensuring stable running of vehicles in the lane. Therefore, before large-scale application, it is particularly important to ensure its safety and reliability in complex environments. Aiming at the shortcomings of existing test methods in efficiency and scene quality, this paper proposes a test scene design and optimization method based on principal component analysis (PCA) and intelligent optimization algorithm to improve the comprehensiveness and scientificity of the test. Through the analysis of LKS system function characteristics, automatic classification and operation domain, the test requirements are defined, four types of typical test scenarios are designed, and hierarchical mathematical models are established. Based on real accident data, the key variables of discrete scene elements were extracted by PCA method to enhance scene authenticity. Aiming at the elements of the continuous scene, an intelligent optimization model based on safety boundary was constructed, which was solved by combining genetic algorithm (GA) and particle swarm optimization algorithm (PSO), and finally formed a key test scenario set covering 14 lane keeping, 34 front vehicle stationary, 34 front vehicle braking and 17 neighboring vehicles entering the scene. Through the hardware-in-the-loop test platform to verify the effectiveness of the scenario, the experimental results show that the proposed method significantly improves the test efficiency and scene quality, and provides a strong support for the full-scale LKS test and large-scale application of autonomous driving technology.

随着自动驾驶技术的快速发展，车道保持系统（LKS）已成为提高车辆行驶安全性的重要功能。它通过保证车道内车辆的稳定运行，有效地降低了车道偏离引起的事故风险。因此，在大规模应用之前，确保其在复杂环境下的安全性和可靠性就显得尤为重要。针对现有测试方法在效率和场景质量方面的不足，本文提出了一种基于主成分分析（PCA）和智能优化算法的测试场景设计与优化方法，以提高测试的全面性和科学性。通过对LKS系统功能特点、自动分类和运行域的分析，定义了测试需求，设计了四种典型测试场景，建立了分层数学模型。基于真实事故数据，采用主成分分析法提取离散场景要素的关键变量，增强场景真实性。针对连续场景要素，构建基于安全边界的智能优化模型，结合遗传算法（GA）和粒子群优化算法（PSO）进行求解，最终形成包含14辆车道保持、34辆前车静止、34辆前车制动和17辆相邻车辆进入场景的关键测试场景集。通过硬件在环测试平台验证场景的有效性，实验结果表明，所提方法显著提高了测试效率和场景质量，为自动驾驶技术的全尺寸LKS测试和大规模应用提供了有力支持。

{"title":"Test Scenario Design and Optimization of Automated Driving Lane Keeping System Based On PCA and Intelligent Algorithm","authors":"Linfeng Hao","doi":"10.1016/j.procs.2025.04.194","DOIUrl":"10.1016/j.procs.2025.04.194","url":null,"abstract":"<div><div>With the rapid development of autonomous driving technology, lane keeping system (LKS) has become an important function to improve vehicle driving safety. It effectively reduces the risk of accidents caused by lane departure by ensuring stable running of vehicles in the lane. Therefore, before large-scale application, it is particularly important to ensure its safety and reliability in complex environments. Aiming at the shortcomings of existing test methods in efficiency and scene quality, this paper proposes a test scene design and optimization method based on principal component analysis (PCA) and intelligent optimization algorithm to improve the comprehensiveness and scientificity of the test. Through the analysis of LKS system function characteristics, automatic classification and operation domain, the test requirements are defined, four types of typical test scenarios are designed, and hierarchical mathematical models are established. Based on real accident data, the key variables of discrete scene elements were extracted by PCA method to enhance scene authenticity. Aiming at the elements of the continuous scene, an intelligent optimization model based on safety boundary was constructed, which was solved by combining genetic algorithm (GA) and particle swarm optimization algorithm (PSO), and finally formed a key test scenario set covering 14 lane keeping, 34 front vehicle stationary, 34 front vehicle braking and 17 neighboring vehicles entering the scene. Through the hardware-in-the-loop test platform to verify the effectiveness of the scenario, the experimental results show that the proposed method significantly improves the test efficiency and scene quality, and provides a strong support for the full-scale LKS test and large-scale application of autonomous driving technology.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 237-246"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124818","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Design of an Intelligent Navigation System Integrating Reinforcement Learning and Computer Vision Algorithms 集成强化学习和计算机视觉算法的智能导航系统设计

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.411

Lili Wang

Traditional static guidance systems have problems such as poor interactivity and low path planning efficiency. This paper designs an intelligent guidance system to achieve efficient, accurate and personalized navigation services. This paper constructs an intelligent guidance system that integrates reinforcement learning and computer vision algorithms, and adopts a multi-level architecture: the perception layer collects environmental data, the data processing layer uses YOLO and semantic segmentation to extract features, the decision layer uses deep Q network (DQN) to plan and optimize the path, and the interaction layer provides intuitive navigation and user feedback mechanism. The system effectively solves the limitations of traditional guidance systems in complex environments and improves navigation efficiency and user experience. In terms of path planning efficiency, the average path planning time of the intelligent guidance system is shorter than that of the traditional system; in terms of path navigation accuracy, the average accuracy of the intelligent guidance system reaches 99.1%, which is much higher than the 95.2% of the traditional system. These data fully prove the effectiveness of the intelligent guidance system proposed in this paper in improving the quality of navigation services and user experience.

传统的静态制导系统存在交互性差、路径规划效率低等问题。为了实现高效、准确、个性化的导航服务，本文设计了一种智能导航系统。本文构建了一个集成强化学习和计算机视觉算法的智能制导系统，采用多层架构：感知层收集环境数据，数据处理层使用YOLO和语义分割提取特征，决策层使用深度Q网络（DQN）规划和优化路径，交互层提供直观的导航和用户反馈机制。该系统有效地解决了传统制导系统在复杂环境下的局限性，提高了导航效率和用户体验。在路径规划效率方面，智能制导系统的平均路径规划时间比传统系统短；在路径导航精度方面，智能制导系统的平均精度达到99.1%，远高于传统系统的95.2%。这些数据充分证明了本文提出的智能导航系统在提高导航服务质量和用户体验方面的有效性。

{"title":"Design of an Intelligent Navigation System Integrating Reinforcement Learning and Computer Vision Algorithms","authors":"Lili Wang","doi":"10.1016/j.procs.2025.04.411","DOIUrl":"10.1016/j.procs.2025.04.411","url":null,"abstract":"<div><div>Traditional static guidance systems have problems such as poor interactivity and low path planning efficiency. This paper designs an intelligent guidance system to achieve efficient, accurate and personalized navigation services. This paper constructs an intelligent guidance system that integrates reinforcement learning and computer vision algorithms, and adopts a multi-level architecture: the perception layer collects environmental data, the data processing layer uses YOLO and semantic segmentation to extract features, the decision layer uses deep Q network (DQN) to plan and optimize the path, and the interaction layer provides intuitive navigation and user feedback mechanism. The system effectively solves the limitations of traditional guidance systems in complex environments and improves navigation efficiency and user experience. In terms of path planning efficiency, the average path planning time of the intelligent guidance system is shorter than that of the traditional system; in terms of path navigation accuracy, the average accuracy of the intelligent guidance system reaches 99.1%, which is much higher than the 95.2% of the traditional system. These data fully prove the effectiveness of the intelligent guidance system proposed in this paper in improving the quality of navigation services and user experience.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 829-837"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124840","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Path Planning and Control of Building Robots Based on Reinforcement Learning Algorithm in Intelligent Construction 智能建筑中基于强化学习算法的建筑机器人路径规划与控制

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.255

Rendong Jin

At present, the construction industry is actively promoting new construction methods such as intelligent and object-intensive construction. Among them, mobile operation robots, as one of the important solutions in the intelligent construction environment, involve key technologies such as obstacle avoidance, path planning, positioning, navigation, sensing and communication, and motion control and path planning problems are considered to be its most complex tasks. This paper studies the algorithm principle and optimization method based on reinforcement learning for the path planning and control problem of mobile operation robots in the intelligent construction environment. Reinforcement learning realizes strategy iteration through a reward and punishment mechanism, and can adaptively find the best course of action in an unfamiliar setting. Q-learning, as a classic algorithm, seeks to maximize long-term rewards by updating the value function. Nevertheless, the conventional Q-learning algorithm has issues with sparse rewards, sluggish convergence, and a propensity to enter local optimality when Q values are initialized. To this end, this paper introduces the artificial potential field method, uses gravitational and repulsive potential fields to optimize path planning, and improves the Q value function by integrating the reward mechanism of potential fields. The gravitational potential field attracts the robot to approach the target point, while the repulsive potential field avoids collision with obstacles. Experiments show that this method effectively solves the issue of path planning in intricate settings and offers a fresh concept for intelligent navigation of construction robots.

目前，建筑行业正在积极推广智能化、对象密集型施工等新型施工方式。其中，移动作业机器人作为智能建筑环境的重要解决方案之一，涉及到避障、路径规划、定位、导航、传感和通信等关键技术，运动控制和路径规划问题被认为是其最复杂的任务。本文研究了智能建筑环境下移动作业机器人路径规划与控制问题的算法原理和基于强化学习的优化方法。强化学习通过奖惩机制实现策略迭代，能够在不熟悉的环境中自适应地找到最佳行动方案。Q-learning作为一种经典算法，通过更新价值函数来寻求长期回报最大化。然而，传统的Q-learning算法存在奖励稀疏、收敛缓慢以及在初始化Q值时倾向于进入局部最优的问题。为此，本文引入人工势场法，利用引力和斥力势场对路径规划进行优化，并通过整合势场奖励机制对Q值函数进行改进。重力势场吸引机器人向目标点靠近，斥力势场避免与障碍物碰撞。实验表明，该方法有效地解决了复杂环境下的路径规划问题，为建筑机器人的智能导航提供了新的思路。

{"title":"Path Planning and Control of Building Robots Based on Reinforcement Learning Algorithm in Intelligent Construction","authors":"Rendong Jin","doi":"10.1016/j.procs.2025.04.255","DOIUrl":"10.1016/j.procs.2025.04.255","url":null,"abstract":"<div><div>At present, the construction industry is actively promoting new construction methods such as intelligent and object-intensive construction. Among them, mobile operation robots, as one of the important solutions in the intelligent construction environment, involve key technologies such as obstacle avoidance, path planning, positioning, navigation, sensing and communication, and motion control and path planning problems are considered to be its most complex tasks. This paper studies the algorithm principle and optimization method based on reinforcement learning for the path planning and control problem of mobile operation robots in the intelligent construction environment. Reinforcement learning realizes strategy iteration through a reward and punishment mechanism, and can adaptively find the best course of action in an unfamiliar setting. Q-learning, as a classic algorithm, seeks to maximize long-term rewards by updating the value function. Nevertheless, the conventional Q-learning algorithm has issues with sparse rewards, sluggish convergence, and a propensity to enter local optimality when Q values are initialized. To this end, this paper introduces the artificial potential field method, uses gravitational and repulsive potential fields to optimize path planning, and improves the Q value function by integrating the reward mechanism of potential fields. The gravitational potential field attracts the robot to approach the target point, while the repulsive potential field avoids collision with obstacles. Experiments show that this method effectively solves the issue of path planning in intricate settings and offers a fresh concept for intelligent navigation of construction robots.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 637-646"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124858","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Social Media Data Mining and Online Consumer Behavior Analysis 社交媒体数据挖掘与在线消费者行为分析

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.220

Hongxin Li

This study explores the impact of social media marketing activities on consumer behavior, emotional attitudes, and purchasing decisions, aiming to optimize brand marketing strategies. As social media becomes increasingly important in consumer decision-making, how to improve marketing effectiveness through sentiment analysis and behavioral analysis has become a pressing issue to be addressed. Research methods include dynamic sentiment tracking, user behavior cluster analysis, and purchase decision path modeling. Firstly, the study uses sentiment analysis to track consumers’ sentiment fluctuations on social media in real time and analyze the impact of sentiment changes on purchase intention and brand awareness. Secondly, cluster analysis is used to segment users and identify the behavioral characteristics and purchase preferences of different groups to support personalized marketing strategies. Finally, a consumer purchase decision model is constructed through path analysis to explore the role of factors such as emotional tendencies, interactive behaviors and social influences in the decision-making process. The experimental results show that brand activities significantly increase users’ interaction frequency, sharing frequency and purchase frequency. After the activity, the user’s interaction index increased by an average of 47.6, the average number of shares increased by 6.3 times, and the average purchase frequency increased by 5.5 times. In particular, users with high interaction and positive emotional tendencies showed stronger purchase intentions and brand loyalty. Research has found that social media activities not only increase brand exposure but also promote consumers’ emotional identification and brand loyalty. Therefore, brands should flexibly adjust marketing strategies based on the portraits of different user groups, and use sentiment analysis and social influence to optimize the purchase decision path, thereby improving marketing effectiveness.

本研究探讨社会化媒体营销活动对消费者行为、情感态度和购买决策的影响，旨在优化品牌营销策略。随着社交媒体在消费者决策中的作用越来越重要，如何通过情感分析和行为分析来提高营销效果已经成为一个亟待解决的问题。研究方法包括动态情感跟踪、用户行为聚类分析和购买决策路径建模。首先，本研究采用情绪分析法，实时跟踪消费者在社交媒体上的情绪波动，分析情绪变化对购买意愿和品牌认知的影响。其次，利用聚类分析对用户进行细分，识别不同群体的行为特征和购买偏好，为个性化营销策略提供支持。最后，通过路径分析构建消费者购买决策模型，探讨情感倾向、互动行为、社会影响等因素在决策过程中的作用。实验结果表明，品牌活动显著提高了用户的互动频率、分享频率和购买频率。活动结束后，用户互动指数平均增长47.6，平均分享次数增长6.3倍，平均购买频率增长5.5倍。尤其是互动程度高、情绪倾向积极的用户，其购买意愿和品牌忠诚度更强。研究发现，社交媒体活动不仅增加了品牌曝光率，还促进了消费者的情感认同和品牌忠诚度。因此，品牌应根据不同用户群体的画像，灵活调整营销策略，并利用情感分析和社会影响力来优化购买决策路径，从而提高营销效果。

{"title":"Social Media Data Mining and Online Consumer Behavior Analysis","authors":"Hongxin Li","doi":"10.1016/j.procs.2025.04.220","DOIUrl":"10.1016/j.procs.2025.04.220","url":null,"abstract":"<div><div>This study explores the impact of social media marketing activities on consumer behavior, emotional attitudes, and purchasing decisions, aiming to optimize brand marketing strategies. As social media becomes increasingly important in consumer decision-making, how to improve marketing effectiveness through sentiment analysis and behavioral analysis has become a pressing issue to be addressed. Research methods include dynamic sentiment tracking, user behavior cluster analysis, and purchase decision path modeling. Firstly, the study uses sentiment analysis to track consumers’ sentiment fluctuations on social media in real time and analyze the impact of sentiment changes on purchase intention and brand awareness. Secondly, cluster analysis is used to segment users and identify the behavioral characteristics and purchase preferences of different groups to support personalized marketing strategies. Finally, a consumer purchase decision model is constructed through path analysis to explore the role of factors such as emotional tendencies, interactive behaviors and social influences in the decision-making process. The experimental results show that brand activities significantly increase users’ interaction frequency, sharing frequency and purchase frequency. After the activity, the user’s interaction index increased by an average of 47.6, the average number of shares increased by 6.3 times, and the average purchase frequency increased by 5.5 times. In particular, users with high interaction and positive emotional tendencies showed stronger purchase intentions and brand loyalty. Research has found that social media activities not only increase brand exposure but also promote consumers’ emotional identification and brand loyalty. Therefore, brands should flexibly adjust marketing strategies based on the portraits of different user groups, and use sentiment analysis and social influence to optimize the purchase decision path, thereby improving marketing effectiveness.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 406-413"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Application Research of Deep Learning in Financial Time Series Prediction 深度学习在金融时间序列预测中的应用研究

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.172

Lin Zou

The market’s every twitch has the power to ripple through the entire economic landscape. Think of stock indices as the market’s temperature gauge, signaling whether it’s heating up or cooling down. For the current methods of predicting stock indices using deep learning models, the computational complexity of deep learning models is relatively high. Therefore, how to reduce computational costs without compromising prediction performance has become a key and difficult point for future work. This study only conducted predictive research on market data and did not apply the two deep learning models proposed by our institute to time series data in other fields. Therefore, applying the deep learning model proposed in this article to time series in other non-financial fields is a key direction for future research. This study not only provides a new perspective for predicting markets, but also lays the foundation for further application of deep learning technology in the financial field.

市场的每一次波动都有波及整个经济格局的力量。可以把股指看作是市场的温度表，表明市场是在升温还是在降温。对于目前使用深度学习模型预测股票指数的方法，深度学习模型的计算复杂度较高。因此，如何在不影响预测性能的前提下降低计算成本成为未来工作的重点和难点。本研究仅对市场数据进行了预测研究，未将研究所提出的两种深度学习模型应用于其他领域的时间序列数据。因此，将本文提出的深度学习模型应用于其他非金融领域的时间序列是未来研究的重点方向。本研究不仅为市场预测提供了新的视角，也为深度学习技术在金融领域的进一步应用奠定了基础。

引用次数: 0

Application of Computer Image Recognition Algorithm in Agricultural Product Quality Inspection 计算机图像识别算法在农产品质量检测中的应用

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.04.237

Xiangyu Lu

In agricultural product quality inspection, subtle defects of agricultural products are easily disturbed by complex background and lighting changes, making them difficult to identify. To solve this problem, this paper uses Vision Transformer (ViT) to increase the precision of identifying minute flaws on agricultural items’ surfaces. The training data set is enlarged using data augmentation technology, which rotates, crops, and modifies brightness to enhance the model’s capacity to adjust to various changes. Using a self-attention mechanism, the ViT model detects minor surface flaws globally and captures long-range dependencies in the image. Combining transfer learning and fine-tuning strategies, the ViT model pre-trained on a large-scale image dataset was optimized on a specific agricultural product dataset, further improving recognition accuracy and robustness. Experimental results indicate that the ViT model in this work has F1-score of 0.89 and a precision of 92.5% when compared to other models. This shows that ViT has high accuracy and robustness in agricultural product quality detection. The results of this paper have made a great contribution to the development of future agricultural automation detection.

在农产品质量检测中，农产品的细微缺陷容易受到复杂背景和光照变化的干扰，难以识别。为了解决这一问题，本文利用视觉变压器（Vision Transformer, ViT）来提高农产品表面微小缺陷的识别精度。利用数据增强技术对训练数据集进行放大，通过旋转、裁剪和修改亮度来增强模型对各种变化的适应能力。使用自注意机制，ViT模型可以全局检测微小的表面缺陷，并捕获图像中的长期依赖关系。将迁移学习和微调策略相结合，在大规模图像数据集上进行预训练的ViT模型在特定农产品数据集上进行优化，进一步提高识别精度和鲁棒性。实验结果表明，与其他模型相比，本文的ViT模型的f1得分为0.89，精度为92.5%。这表明ViT在农产品质量检测中具有较高的准确性和鲁棒性。本文的研究成果对未来农业自动化检测的发展有很大的贡献。

{"title":"Application of Computer Image Recognition Algorithm in Agricultural Product Quality Inspection","authors":"Xiangyu Lu","doi":"10.1016/j.procs.2025.04.237","DOIUrl":"10.1016/j.procs.2025.04.237","url":null,"abstract":"<div><div>In agricultural product quality inspection, subtle defects of agricultural products are easily disturbed by complex background and lighting changes, making them difficult to identify. To solve this problem, this paper uses Vision Transformer (ViT) to increase the precision of identifying minute flaws on agricultural items’ surfaces. The training data set is enlarged using data augmentation technology, which rotates, crops, and modifies brightness to enhance the model’s capacity to adjust to various changes. Using a self-attention mechanism, the ViT model detects minor surface flaws globally and captures long-range dependencies in the image. Combining transfer learning and fine-tuning strategies, the ViT model pre-trained on a large-scale image dataset was optimized on a specific agricultural product dataset, further improving recognition accuracy and robustness. Experimental results indicate that the ViT model in this work has F1-score of 0.89 and a precision of 92.5% when compared to other models. This shows that ViT has high accuracy and robustness in agricultural product quality detection. The results of this paper have made a great contribution to the development of future agricultural automation detection.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"261 ","pages":"Pages 485-493"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144124646","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Comparison of Machine Learning Algorithms for Heart Disease Prediction 心脏疾病预测的机器学习算法比较

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.03.172

Ujjwal Daharwal , Indrasen Singh , Ganesh Khekare

Heart disease remains a significant health concern globally, prompting the exploration of advanced methodologies for its early prediction and intervention. In this research, we use a comprehensive method to predict heart disease by using the capability of various machine learning algorithms. A dataset comprising diverse patient attributes and medical indicators is utilized for training and testing the models. Our study explores the effectiveness of various prominent machine learning algorithms such as Random Forest, and K-Nearest Neighbors (KNN) in predicting heart disease risk. Through meticulous feature selection and engineering, we enhance the algorithm’s ability to discern patterns and relationships within the data. To improve model interpretability and generalizability, feature selection and optimization procedures are examined. Furthermore, a comparison between contrast the strengths and limitations of these algorithms to identify the most suitable model for heart disease prediction has been presented. The findings of this research contribute to the ongoing efforts to develop accurate and reliable predictive tools for early detection of cardiovascular diseases, thereby facilitating timely intervention and improving patient outcomes. Integration of machine learning into cardiovascular risk assessment holds great promise for advancing personalized medicine and preventive healthcare strategies.

心脏病仍然是全球重大的健康问题，促使探索其早期预测和干预的先进方法。在这项研究中，我们利用各种机器学习算法的能力，使用一种综合的方法来预测心脏病。使用包含不同患者属性和医疗指标的数据集来训练和测试模型。我们的研究探讨了各种著名的机器学习算法，如随机森林和k -近邻（KNN）在预测心脏病风险方面的有效性。通过细致的特征选择和工程，我们增强了算法在数据中识别模式和关系的能力。为了提高模型的可解释性和泛化性，研究了特征选择和优化过程。此外，还比较了这些算法的优势和局限性，以确定最适合心脏病预测的模型。这项研究的发现有助于开发准确可靠的预测工具，以早期发现心血管疾病，从而促进及时干预和改善患者的预后。将机器学习整合到心血管风险评估中，对于推进个性化医疗和预防性医疗策略具有很大的前景。

{"title":"Comparison of Machine Learning Algorithms for Heart Disease Prediction","authors":"Ujjwal Daharwal , Indrasen Singh , Ganesh Khekare","doi":"10.1016/j.procs.2025.03.172","DOIUrl":"10.1016/j.procs.2025.03.172","url":null,"abstract":"<div><div>Heart disease remains a significant health concern globally, prompting the exploration of advanced methodologies for its early prediction and intervention. In this research, we use a comprehensive method to predict heart disease by using the capability of various machine learning algorithms. A dataset comprising diverse patient attributes and medical indicators is utilized for training and testing the models. Our study explores the effectiveness of various prominent machine learning algorithms such as Random Forest, and K-Nearest Neighbors (KNN) in predicting heart disease risk. Through meticulous feature selection and engineering, we enhance the algorithm’s ability to discern patterns and relationships within the data. To improve model interpretability and generalizability, feature selection and optimization procedures are examined. Furthermore, a comparison between contrast the strengths and limitations of these algorithms to identify the most suitable model for heart disease prediction has been presented. The findings of this research contribute to the ongoing efforts to develop accurate and reliable predictive tools for early detection of cardiovascular diseases, thereby facilitating timely intervention and improving patient outcomes. Integration of machine learning into cardiovascular risk assessment holds great promise for advancing personalized medicine and preventive healthcare strategies.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"260 ","pages":"Pages 12-21"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144139173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Safeguarding the Landscape of Mental Wellness: Analyzing Cyber Threats and Mitigation Strategies in Digital Healthcare 保护心理健康：分析数字医疗保健中的网络威胁和缓解策略

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.03.173

Cyrus Mehra , Arvind K. Sharma

With the pervasiveness of digital technologies across various sectors, mental health care has not been left out, where the use of IoT has enhanced therapeutic practice through better monitoring of patients, collection of data, and proper alignment of treatment measures. However, the data collected for mental health is synonymous with high sensitivity; the information collected via IoT devices is prone to cybersecurity challenges which may compromise patient privacy or interfere with the treatment process. This review paper seeks to synthesize the two sides of IoT application in mental health, which are; advantages and cybersecurity threats. This paper aims at conducting a deep review of sources to ascertain the level of cybersecurity threats surrounding IoT in mental health, the performance of the current security measures, and possible future measures. It aims to identify the best measures to curb insecurity tendencies in the use of IoT in mental health. The review will contribute to the existing sources in the quest to secure health data in the rapidly evolving digital health field hence helping to create trust in secure information sharing.

随着数字技术在各个部门的普及，精神卫生保健也没有被遗漏，物联网的使用通过更好地监测患者、收集数据和适当调整治疗措施，加强了治疗实践。然而，为心理健康收集的数据是高度敏感的代名词；通过物联网设备收集的信息容易受到网络安全挑战，这可能会损害患者隐私或干扰治疗过程。本文旨在综合物联网在心理健康领域应用的两个方面：优势和网络安全威胁。本文旨在对来源进行深入审查，以确定物联网在心理健康方面的网络安全威胁程度、当前安全措施的效果以及未来可能采取的措施。它旨在确定最佳措施，以遏制在心理健康中使用物联网的不安全倾向。该审查将有助于在快速发展的数字卫生领域寻求安全卫生数据的现有来源，从而有助于在安全信息共享方面建立信任。

{"title":"Safeguarding the Landscape of Mental Wellness: Analyzing Cyber Threats and Mitigation Strategies in Digital Healthcare","authors":"Cyrus Mehra , Arvind K. Sharma","doi":"10.1016/j.procs.2025.03.173","DOIUrl":"10.1016/j.procs.2025.03.173","url":null,"abstract":"<div><div>With the pervasiveness of digital technologies across various sectors, mental health care has not been left out, where the use of IoT has enhanced therapeutic practice through better monitoring of patients, collection of data, and proper alignment of treatment measures. However, the data collected for mental health is synonymous with high sensitivity; the information collected via IoT devices is prone to cybersecurity challenges which may compromise patient privacy or interfere with the treatment process. This review paper seeks to synthesize the two sides of IoT application in mental health, which are; advantages and cybersecurity threats. This paper aims at conducting a deep review of sources to ascertain the level of cybersecurity threats surrounding IoT in mental health, the performance of the current security measures, and possible future measures. It aims to identify the best measures to curb insecurity tendencies in the use of IoT in mental health. The review will contribute to the existing sources in the quest to secure health data in the rapidly evolving digital health field hence helping to create trust in secure information sharing.</div></div>","PeriodicalId":20465,"journal":{"name":"Procedia Computer Science","volume":"260 ","pages":"Pages 22-31"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144139174","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Framework for Customer Segmentation to Improve Marketing Strategies Using Machine Learning 使用机器学习改进营销策略的客户细分框架

Procedia Computer Science

Pub Date : 2025-01-01 DOI: 10.1016/j.procs.2025.03.240

Aya Ashraf , Christina Albert Rayed , Nancy Awadallah Awad , Heba M. Sabry

It is hard for the marketing team to set a strategy without dividing the customers into groups. Clustering is a well-known machine-learning technique that can be used to implement customer segmentation. It is an unsupervised learning method that creates clusters by dividing a dataset into many valuable subclasses. In online retail datasets, algorithms such as K-means, Mini Batch K-means, Spectral Clustering, and Fuzzy K-means are employed to categorize customers according to their Recency, Frequency, and Monetary (RFM) features. After analyzing the Silhouette Score, the K-means achieved a higher score, 0.432619, which implies that this algorithm achieved comparable cluster cohesion and separation levels. This paper aims to develop a framework for customer segmentation using machine learning to improve marketing strategies.

如果不把顾客分成不同的群体，营销团队很难制定战略。聚类是一种众所周知的机器学习技术，可用于实现客户细分。它是一种无监督学习方法，通过将数据集划分为许多有价值的子类来创建聚类。在在线零售数据集中，采用K-means、Mini Batch K-means、谱聚类和模糊K-means等算法，根据客户的recent， Frequency和Monetary （RFM）特征对客户进行分类。通过对Silhouette Score的分析，K-means获得了更高的分数0.432619，这意味着该算法达到了相当的聚类内聚和分离水平。本文旨在开发一个使用机器学习来改进营销策略的客户细分框架。

引用次数: 0