检索结果-内蒙古大学图书馆

Hierarchically-Trusted Evidential Fusion Method with Consistency Learning for Multimodal Language Understanding

学校读者我要写书评

暂无评论

SSRN 2024年

作者： Yang, Ying Yan-Qiu, Yang Ren, Gang School of Management Hefei University of Technology Hefei230009 China Key Laboratory of Process Optimization and Intelligent Decision-making Ministry of Education Hefei230009 China Ministry of Education Engineering Research Centre for Intelligent Decision-Making & Information System Technologies Hefei230009 China

Human communication in the real world is based on multimodal data. Multimodal language understanding tasks such as multimodal intention recognition and multimodal emotion recognition are key steps in many applications, such as dialogue systems and social media. Although many studies have explored multimodal fusion methods and applied these methods to multimodal language understanding tasks including intent recognition and sentiment analysis, it is still a knotty problem whether the classification results are trusted. Therefore, this paper proposes a hierarchically-trusted evidential fusion method (HTEF) for multimodal language understanding. We construct multiple views from multi-modal data in a hierarchical manner, and extract multiple pieces of evidence from them. Then, we introduce Dempster-Shafer evidence theory(DST) to fuse evidence obtained from multimodal data to ensure the reduction of overall uncertainty. In addition, we explore the feature-level consistency and decision-level consistency of multi-modal data to avoid the conflict across information captured from different modalities. The method is evaluated on the public multimodal intent recognition dataset MIntRec and the emotion recognition dataset MELD. The experimental results show that our model outperforms both existing single-modal and multimodal baseline models across multiple evaluation metrics, which verifies the effectiveness of the proposed method. © 2024, The Authors. All rights reserved.

关键词： Emotion Recognition

Pricing decisions under manufacturer's component open-supply strategy

学校读者我要写书评

暂无评论

arXiv

arXiv 2021年

作者： Zhu, Peiya Qian, Xiaofei Liu, Xinbao Lu, Shaojun School of Management Hefei University of Technology Hefei230009 China Key Laboratory of Process Optimization and Intelligent Decision-making Ministry of Education Hefei230009 China Ministry of Education Engineering Research Center for Intelligent Decision-Making and Information System Technologies Hefei230009 China

Faced with huge market potential and increasing competition in emerging industries, product manufacturers with key technologies tend to consider whether to implement a component open-supply strategy. This study focuses on a pricing game induced by the component open-supply strategy between a vertically integrated manufacturer (who produces key components and end products) and an exterior product manufacturer (who produces end products using purchased key components) with different customer perceived value and different cost structure. This study first establishes a three-stage pricing game model and proposes demand functions by incorporating relative customer perceived value. Based on the demand functions, we obtain feasible regions of the exterior manufacturer's sourcing decision and the optimal price decision in each region. Then the effects of relative customer perceived value, cost structure, and market structure on price decisions and optimal profits of the vertically integrated manufacturer are demonstrated. Finally, as for the optimal component supply strategy, we present a generalized closed-supply Pareto zone and establish supply strategy Pareto zones under several specific configurations. Copyright © 2021, The Authors. All rights reserved.

关键词： Game theory

Predicting Helpfulness of Multimodal Reviews with Customer Confirmation Bias: A Hierarchically-Trusted Multi-View Deep Learning Method

学校读者我要写书评

暂无评论

SSRN 2024年

作者： Yang, Ying Tang, Si Ren, Gang School of Management Hefei University of Technology Hefei230009 China Key Laboratory of Process Optimization and Intelligent Decision-making Ministry of Education Hefei230009 China Ministry of Education Engineering Research Center for Intelligent Decision-Making & Information System Technologies Hefei230009 China

Accurately identifying high quality online-reviews is beneficial to both consumers and e-commerce platforms. Consumers usually exhibit confirmation bias phenomenon when evaluating the helpfulness of product reviews. However, there has been limited research on simulating and incorporating this information processing bias into predicting multimodal review helpfulness. To address this challenge, we combine confirmation bias theory with deep learning techniques, and propose a hierarchically-trusted multi-view deep learning method for predicting the helpfulness of multimodal reviews. Drawing upon the elaborate likelihood model and confirmation bias theory, multiple views are designed to represent consumers’ initial beliefs, central cues and peripheral cues of multimodal reviews. A hierarchically-trusted multi-view fusion strategy is designed to dynamically aggregate the different views considering their interactions. Experimental results on the *** review dataset demonstrate that the proposed method effectively improves the accuracy of evaluating the values of online reviews. This research contributes to the improvement of platform service operations and enhances consumer experience. © 2024, The Authors. All rights reserved.

关键词： Deep learning

A Scheduling Strategy for OLAM Tasks and Its Application in a Financial BI System

学校读者我要写书评

暂无评论

A Scheduling Strategy for OLAM Tasks and Its Application in ...

The Second International Conference on Business Intelligence and Financial Engineering(BIFE 2009)(第二届商务智能与金融工程国际会议)

作者： Yazhuo Gao Zhiwei Ni Yuxiao Zhao School of Management Hefei Univ.of Technology Key Lab.of Process Optimization and Intelligent Decision-making Ministry of Education Hefei 230009 China

This paper proposes a scheduling strategy which divides OLAM (Online Analytical Mining) tasks into query tasks, mining tasks and updating tasks in order to improve the efficiency of Bl (Business Intelligence) systems. OLAM mechanism is considered as the seamless integration of OLAP (Online Analytical processing) and data mining on data cubes, according to which the scheduling strategy focuses on. This paper emphasizes the interaction of OLAP, data mining and data cube updating, illustrates the details of the OLAM scheduling strategy and demonstrates its efficiency by performing it in a financial Bl system.

关键词： OLAM data mining OLAP data cube BI scheduling strategy

Generative Adversarial Network-Based Optimal Energy Storage Capacity Configuration for Newly–Build Microgrids

学校读者我要写书评

暂无评论

SSRN 2024年

作者： Zhou, Kunshu Zhou, Kaile Yang, Shanlin School of Management Hefei University of Technology Hefei230009 China Key Laboratory of Process Optimization and Intelligent Decision-making of Ministry of Education Hefei University of Technology Hefei230009 China

In the initial stage of microgrid construction, historical data is limited. It is challenging to address the issues of the small amount of data and high uncertainty of renewable energy when proceeding with capacity configuration. This study performs Wasserstein distance-based generative adversarial network with gradient penalty (WGAN-GP) to enhance the original 30 days PV power generation scenario data. Furthermore, the improved data is used to build a distributionally robust optimization model based on various discrete scenarios to determine the optimal capacity configuration of energy storage in newly–built microgrids. The results obtained from the solution are compared with those of traditional stochastic optimization and robust optimization, and the robustness of the results is verified by sensitivity analysis. The comparison results demonstrate that the strategy employed in this study can better address the issues of limited data and uncertainty in renewable energy in the capacity configuration process of newly-built microgrids. © 2024, The Authors. All rights reserved.

关键词： Sensitivity analysis

Optimal Government Subsidy Policy for Photovoltaic Project Investment decision making: A Real Options Method

学校读者我要写书评

暂无评论

SSRN 2021年

作者： Zhu, Xing Liao, Baoyu Yang, Shanlin School of Management Hefei University of Technology Hefei230009 China Key Lab of Process Optimization and Intelligent Decision-making Ministry of Education Hefei230009 China

In an uncertain environment, it is important to investigate whether to postpone, abandon or immediately invest in photovoltaic (PV) projects. This paper applies a real options model to explore the optimal investment decision for investors and the government’s optimal incentive strategy in China’s distributed PV market. The uncertainties of feed-in tariffs (FIT) and investment costs are included in this study as critical factors affecting the options value of PV projects. Multiple scenarios are designed from different dimensions, including various subsidy durations, diverse FIT contracts, different resource areas, and distinct regulatory policies. The results indicate that the investment decision for investors and government subsidy policies are affected by many factors, such as current FIT, investment costs, FIT drift rate, investment cost drift rate, the risk-free rate. The impacts of a fixed FIT contract are similar in a regulated market and a free market, but the impacts of a free FIT contract change according to the market system. The results also reveal that appropriate FIT regulation can effectively induce investors who abandon investment to choose to delay investment, and significantly shorten the delay time. These conclusions can provide valuable insights for investors and policymakers to make effective decisions and targeted policies. © 2021, The Authors. All rights reserved.

关键词： Investments

Analysis on Medicine Compounding for Stroke Prevention Treated by Xin'an Physicians Based on Association Rules

学校读者我要写书评

暂无评论

Analysis on Medicine Compounding for Stroke Prevention Treat...

International Conference on intelligent Computation Technology and Automation, ICICTA

作者： Luan Gao Feng-gang Li Jian Wang Yu Liang Yu Li Anhui Chinese Medical College of Chinese Medicine Clinic Institute Hefei China Key Laboratory of Process Optimization and Intelligent Decision-making Ministry of Education Hefei University of Technology Hefei China

The purpose of this paper is to probe into the rules of medicine compounding for stroke prevention treated by Xin'an physicians by data mining. The method is in two steps. First step is to build the database of the Xin'an physical stroke prevention. Second step is to use the method of association rules contained in professional data mining software and probe into the medicine rules for stroke prevention treated by Xin'an physicians. The result is that Chenpi, Fuling, Gancao, Luxiancao, Xiqiancao, Baijili, and Shinanye which are of the seven common Chinese medical compounding prescriptions can be used for Xin'an physicians' stroke prevention. The conclusion is that we should utilize the eliminating phlegm, calming wind and clearing the channels when Xin'an physicians treat stroke problem.

关键词： Medical diagnostic imaging Association rules Itemsets Diseases Drugs

Disagreement-based semi-supervised learning method for enterprise credit risk assessment

学校读者我要写书评

暂无评论

ICIC Express Letters 2015年第6期9卷 1671-1676页

作者： Wang, Gang Li, Ningning School of Management Hefei University of Technology No. 193 Tunxi Road Hefei China Key Laboratory of Process Optimization and Intelligent Decision-Making Ministry of Education Hefei University of Technology No. 193 Tunxi Road Hefei China

The enterprise credit risk assessment problem has long been regarded as an important and widely studied issue in both academia and industry. However, unla-beled data problem is paid less attention to in the credit risk assessment community. To fill in this research gap, one of popular disagreement semi-supervised learning methods, co-training, is introduced to exploit the unlabeled data to improve the enterprise credit risk assessment performance. Co-training allows one to start with just a few labeled instances to produce an initial weak classifier and then use unlabeled data to improve the performance. Experiments based on two credit datasets reveal that co-training SVM&DT (Support Vector Machine & decision Tree) gets the better performance than the compared methods. Based on these results, we can conclude that co-training SVM&DT can be used as an alternative method for enterprise credit risk assessment. © 2015 ICIC International.

关键词： decision trees

CO-RS: A new method for opinion mining based on semi-supervised learning and ensemble learning

学校读者我要写书评

暂无评论

ICIC Express Letters 2015年第7期9卷 1803-1808页

In recent years many machine learning methods have been proposed for opinion mining and the effectiveness of applying them to opinion mining has also been approved theoretically. However, machine learning methods encounter many problems when they are used in practice. One of the problems is to construct learning systems with strong generalization ability. In this study, we propose, Co-RS, a semi-supervised method for opinion mining based on co-training and Random Subspace. Co-RS tries to get the strong generalization ability by integrating the strength of two machine learning paradigms, i.e., semi-supervised learning and ensemble learning. Moreover, four public opinion datasets were investigated to verify the effectiveness of proposed method. Empirical results reveal that Co-RS gets the best result among the compared methods. These results illustrate that Co-RS can be used as a viable method for opinion mining. © 2015, ISSN 1881-803X.

关键词： Sentiment analysis