Pub Date : 2025-01-01DOI: 10.1109/TBDATA.2024.3524839
Biao Wang;Zhao Li;Zenghui Xu;Ji Zhang
Predicting the popularity of information in social networks is crucial for effective social marketing and recommendation systems. However, accurately comprehending the complex dynamics of information diffusion remains a challenging task. Existing methods, including feature-based approaches, point process models, and deep learning techniques, often fail to capture the fine-grained features of information cascades, such as dynamic diffusion patterns, cascade statistics, and the interplay between spatial and temporal information. To address these limitations, we propose Casformer, a novel graph-based Transformer architecture that effectively learns both micro-level time-aware structural information and macro-level long-term influence along the information propagation process. Casformer employs a cascade attention network (CAT) to capture the micro-level features and a Transformer model to learn the macro-level influence. Furthermore, we introduce an adaptive cascade graph sampling strategy based on the temporal diffusion pattern and cascade statistics of information to obtain the most informative cascade graph sequence. By leveraging multi-level fine-grained evolving features of information cascades, Casformer achieves high accuracy in information popularity prediction. Experimental results on real-world social network and scientific citation network datasets demonstrate the effectiveness and superiority of Casformer compared to state-of-the-art methods in information popularity prediction.
{"title":"Casformer: Information Popularity Prediction With Adaptive Cascade Sampling and Graph Transformer in Social Networks","authors":"Biao Wang;Zhao Li;Zenghui Xu;Ji Zhang","doi":"10.1109/TBDATA.2024.3524839","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3524839","url":null,"abstract":"Predicting the popularity of information in social networks is crucial for effective social marketing and recommendation systems. However, accurately comprehending the complex dynamics of information diffusion remains a challenging task. Existing methods, including feature-based approaches, point process models, and deep learning techniques, often fail to capture the fine-grained features of information cascades, such as dynamic diffusion patterns, cascade statistics, and the interplay between spatial and temporal information. To address these limitations, we propose Casformer, a novel graph-based Transformer architecture that effectively learns both micro-level time-aware structural information and macro-level long-term influence along the information propagation process. Casformer employs a cascade attention network (CAT) to capture the micro-level features and a Transformer model to learn the macro-level influence. Furthermore, we introduce an adaptive cascade graph sampling strategy based on the temporal diffusion pattern and cascade statistics of information to obtain the most informative cascade graph sequence. By leveraging multi-level fine-grained evolving features of information cascades, Casformer achieves high accuracy in information popularity prediction. Experimental results on real-world social network and scientific citation network datasets demonstrate the effectiveness and superiority of Casformer compared to state-of-the-art methods in information popularity prediction.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"1652-1663"},"PeriodicalIF":7.5,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Top-k Personalized PageRank (PPR) is a graph analysis method used to determine the $k$ most important nodes with respect to a source node. To realize fast Top-k PPR computation, indexing for each node is effective. When we apply the index-based Top-k PPR methods to dynamic graphs, the index becomes stale with edge updates, and index correction is required. Although the existing methods perform index correction for every update to guarantee Top-k PPR accuracy, they involve heavy re-indexing computation or significant memory overhead. This paper proposes a method that achieves comparable accuracy to guaranteed methods while significantly reducing re-indexing by focusing on the fact that index references are concentrated on the nodes whose index is unlikely to change due to edge updates. In particular, our method omits re-indexing as long as we achieve comparable accuracy. Furthermore, our method involves the minimum memory overhead among the existing index-based methods. The space complexity of the index is $Theta (n + m)$, where $n$ and $m$ are the number of nodes and edges of the graph, respectively. The evaluation results using real-world datasets show that our method achieves more than 0.999 Normalized Discounted Cumulative Gain until 20% of edges are updated from index generation.
{"title":"Reducing Re-Indexing for Top-k Personalized PageRank Computation on Dynamic Graphs","authors":"Tsuyoshi Yamashita;Naoki Matsumoto;Kunitake Kaneko","doi":"10.1109/TBDATA.2024.3524833","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3524833","url":null,"abstract":"Top-k Personalized PageRank (PPR) is a graph analysis method used to determine the <inline-formula><tex-math>$k$</tex-math></inline-formula> most important nodes with respect to a source node. To realize fast Top-k PPR computation, indexing for each node is effective. When we apply the index-based Top-k PPR methods to dynamic graphs, the index becomes stale with edge updates, and index correction is required. Although the existing methods perform index correction for every update to guarantee Top-k PPR accuracy, they involve heavy re-indexing computation or significant memory overhead. This paper proposes a method that achieves comparable accuracy to guaranteed methods while significantly reducing re-indexing by focusing on the fact that index references are concentrated on the nodes whose index is unlikely to change due to edge updates. In particular, our method omits re-indexing as long as we achieve comparable accuracy. Furthermore, our method involves the minimum memory overhead among the existing index-based methods. The space complexity of the index is <inline-formula><tex-math>$Theta (n + m)$</tex-math></inline-formula>, where <inline-formula><tex-math>$n$</tex-math></inline-formula> and <inline-formula><tex-math>$m$</tex-math></inline-formula> are the number of nodes and edges of the graph, respectively. The evaluation results using real-world datasets show that our method achieves more than 0.999 Normalized Discounted Cumulative Gain until 20% of edges are updated from index generation.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"1707-1719"},"PeriodicalIF":7.5,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10819623","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144598067","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01DOI: 10.1109/TBDATA.2024.3524828
Khondhaker Al Momin;Arif Mohaimin Sadri;Kristin Olofsson;K.K. Muraleetharan;Hugh Gladwin
In an era increasingly affected by natural and human-caused disasters, the role of social media in disaster communication has become ever more critical. Despite substantial research on social media use during crises, a significant gap remains in detecting crisis-related misinformation. Detecting deviations in information is fundamental for identifying and curbing the spread of misinformation. This study introduces a novel Information Switching Pattern Model to identify dynamic shifts in perspectives among users who mention each other in crisis-related narratives on social media. These shifts serve as evidence of crisis misinformation affecting user-mention network interactions. The study utilizes advanced natural language processing, network science, and census data to analyze geotagged tweets related to compound disaster events in Oklahoma in 2022. The impact of misinformation is revealed by distinct engagement patterns among various user types, such as bots, private organizations, non-profits, government agencies, and news media throughout different disaster stages. These patterns show how different disasters influence public sentiment, highlight the heightened vulnerability of mobile home communities, and underscore the importance of education and transportation access in crisis response. Understanding these engagement patterns is crucial for detecting misinformation and leveraging social media as an effective tool for risk communication during disasters.
{"title":"Information Switching Patterns of Risk Communication in Social Media During Disasters","authors":"Khondhaker Al Momin;Arif Mohaimin Sadri;Kristin Olofsson;K.K. Muraleetharan;Hugh Gladwin","doi":"10.1109/TBDATA.2024.3524828","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3524828","url":null,"abstract":"In an era increasingly affected by natural and human-caused disasters, the role of social media in disaster communication has become ever more critical. Despite substantial research on social media use during crises, a significant gap remains in detecting crisis-related misinformation. Detecting deviations in information is fundamental for identifying and curbing the spread of misinformation. This study introduces a novel <italic>Information Switching Pattern Model</i> to identify dynamic shifts in perspectives among users who mention each other in crisis-related narratives on social media. These shifts serve as evidence of crisis misinformation affecting user-mention network interactions. The study utilizes advanced natural language processing, network science, and census data to analyze geotagged tweets related to compound disaster events in Oklahoma in 2022. The impact of misinformation is revealed by distinct engagement patterns among various user types, such as bots, private organizations, non-profits, government agencies, and news media throughout different disaster stages. These patterns show how different disasters influence public sentiment, highlight the heightened vulnerability of mobile home communities, and underscore the importance of education and transportation access in crisis response. Understanding these engagement patterns is crucial for detecting misinformation and leveraging social media as an effective tool for risk communication during disasters.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"1733-1744"},"PeriodicalIF":7.5,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10820023","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144606221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Large Language models (LLMs) usually rely on extensive training datasets. In the financial domain, creating numerical reasoning datasets that include a mix of tables and long text often involves substantial manual annotation expenses. To address the limited data resources and reduce the annotation cost, we introduce FinLLMs, a method for generating financial question-answering (QA) data based on common financial formulas using LLMs. First, we compile a list of common financial formulas and construct a graph based on the variables these formulas employ. We then augment the formula set by combining those that share identical variables as new elements. Specifically, we explore formulas obtained by manual annotation and merge those formulas with shared variables by traversing the constructed graph. Finally, utilizing LLMs, we generate financial QA data that encompasses both tabular information and long textual content, building on the collected formula set. Our experiments demonstrate that the synthetic data generated by FinLLMs effectively enhances the performance of various numerical reasoning models in the financial domain, including both pre-trained language models (PLMs) and fine-tuned LLMs. This performance surpasses that of two established benchmark financial QA datasets.
{"title":"FinLLMs: A Framework for Financial Reasoning Dataset Generation With Large Language Models","authors":"Ziqiang Yuan;Kaiyuan Wang;Shoutai Zhu;Ye Yuan;Jingya Zhou;Yanlin Zhu;Wenqi Wei","doi":"10.1109/TBDATA.2024.3524083","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3524083","url":null,"abstract":"Large Language models (LLMs) usually rely on extensive training datasets. In the financial domain, creating numerical reasoning datasets that include a mix of tables and long text often involves substantial manual annotation expenses. To address the limited data resources and reduce the annotation cost, we introduce FinLLMs, a method for generating financial question-answering (QA) data based on common financial formulas using LLMs. First, we compile a list of common financial formulas and construct a graph based on the variables these formulas employ. We then augment the formula set by combining those that share identical variables as new elements. Specifically, we explore formulas obtained by manual annotation and merge those formulas with shared variables by traversing the constructed graph. Finally, utilizing LLMs, we generate financial QA data that encompasses both tabular information and long textual content, building on the collected formula set. Our experiments demonstrate that the synthetic data generated by FinLLMs effectively enhances the performance of various numerical reasoning models in the financial domain, including both pre-trained language models (PLMs) and fine-tuned LLMs. This performance surpasses that of two established benchmark financial QA datasets.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 5","pages":"2264-2277"},"PeriodicalIF":5.7,"publicationDate":"2024-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144990274","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-12-30DOI: 10.1109/TBDATA.2024.3524081
Jinsong Chen;Chang Liu;Kaiyuan Gao;Gaichao Li;Kun He
Graph Transformers, emerging as a new architecture for graph representation learning, suffer from the quadratic complexity and can only handle graphs with at most thousands of nodes. To this end, we propose a Neighborhood Aggregation Graph Transformer (NAGphormer) that treats each node as a sequence containing a series of tokens constructed by our proposed Hop2Token module. For each node, Hop2Token aggregates the neighborhood features from different hops into different representations, producing a sequence of token vectors as one input. In this way, NAGphormer could be trained in a mini-batch manner and thus could scale to large graphs with millions of nodes. To further enhance the model's generalization, we propose NAGphormer+, an extended model of NAGphormer with a novel data augmentation method called Neighborhood Augmentation (NrAug). Based on the output of Hop2Token, NrAug simultaneously augments the features of neighborhoods from global as well as local views. In this way, NAGphormer+ can fully utilize the neighborhood information of multiple nodes, thereby undergoing more comprehensive training and improving the model's generalization capability. Extensive experiments on benchmark datasets from small to large demonstrate the superiority of NAGphormer+ against existing graph Transformers and mainstream GNNs, as well as the original NAGphormer.
{"title":"NAGphormer+: A Tokenized Graph Transformer With Neighborhood Augmentation for Node Classification in Large Graphs","authors":"Jinsong Chen;Chang Liu;Kaiyuan Gao;Gaichao Li;Kun He","doi":"10.1109/TBDATA.2024.3524081","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3524081","url":null,"abstract":"Graph Transformers, emerging as a new architecture for graph representation learning, suffer from the quadratic complexity and can only handle graphs with at most thousands of nodes. To this end, we propose a Neighborhood Aggregation Graph Transformer (NAGphormer) that treats each node as a sequence containing a series of tokens constructed by our proposed Hop2Token module. For each node, Hop2Token aggregates the neighborhood features from different hops into different representations, producing a sequence of token vectors as one input. In this way, NAGphormer could be trained in a mini-batch manner and thus could scale to large graphs with millions of nodes. To further enhance the model's generalization, we propose NAGphormer+, an extended model of NAGphormer with a novel data augmentation method called Neighborhood Augmentation (NrAug). Based on the output of Hop2Token, NrAug simultaneously augments the features of neighborhoods from global as well as local views. In this way, NAGphormer+ can fully utilize the neighborhood information of multiple nodes, thereby undergoing more comprehensive training and improving the model's generalization capability. Extensive experiments on benchmark datasets from small to large demonstrate the superiority of NAGphormer+ against existing graph Transformers and mainstream GNNs, as well as the original NAGphormer.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2085-2098"},"PeriodicalIF":7.5,"publicationDate":"2024-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144598062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Multi-view multi-label classification is a crucial machine learning paradigm aimed at building robust multi-label predictors by integrating heterogeneous features from various sources while addressing multiple correlated labels. However, in real-world applications, concerns over data confidentiality and security often prevent data exchange or fusion across different sources, leading to the challenging issue of data islands. To tackle this problem, we propose a general federated multi-view multi-label classification method, FMVML, which integrates a novel multi-view multi-label classification technique into a federated learning framework. This approach enables cross-view feature fusion and multi-label semantic classification while preserving the data privacy of each independent source. Within this federated framework, we first extract view-specific information from each individual client to capture unique characteristics and then consolidate consensus information from different views on the global server to represent shared features. Unlike previous methods, our approach enhances cross-view fusion and semantic expression by jointly capturing both feature and semantic aspects of specificity and commonality. The final label predictions are generated by combining the view-specific predictions from individual clients and the consensus predictions from the global server. Extensive experiments across various applications demonstrate that FMVML fully leverages multi-view data in a privacy-preserving manner and consistently outperforms state-of-the-art methods.
{"title":"Federated Multi-View Multi-Label Classification","authors":"Hongdao Meng;Yongjian Deng;Qiyu Zhong;Yipeng Wang;Zhen Yang;Gengyu Lyu","doi":"10.1109/TBDATA.2024.3522812","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3522812","url":null,"abstract":"Multi-view multi-label classification is a crucial machine learning paradigm aimed at building robust multi-label predictors by integrating heterogeneous features from various sources while addressing multiple correlated labels. However, in real-world applications, concerns over data confidentiality and security often prevent data exchange or fusion across different sources, leading to the challenging issue of data islands. To tackle this problem, we propose a general federated multi-view multi-label classification method, FMVML, which integrates a novel multi-view multi-label classification technique into a federated learning framework. This approach enables cross-view feature fusion and multi-label semantic classification while preserving the data privacy of each independent source. Within this federated framework, we first extract view-specific information from each individual client to capture unique characteristics and then consolidate consensus information from different views on the global server to represent shared features. Unlike previous methods, our approach enhances cross-view fusion and semantic expression by jointly capturing both feature and semantic aspects of specificity and commonality. The final label predictions are generated by combining the view-specific predictions from individual clients and the consensus predictions from the global server. Extensive experiments across various applications demonstrate that FMVML fully leverages multi-view data in a privacy-preserving manner and consistently outperforms state-of-the-art methods.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2072-2084"},"PeriodicalIF":7.5,"publicationDate":"2024-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144598078","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Semantic expertise remains a reliable foundation for industrial decision-making, while Large Language Models (LLMs) can augment the often limited empirical knowledge by generating domain-specific insights, though the quality of this generative knowledge is uncertain. Integrating LLMs with the collective wisdom of multiple stakeholders could enhance the quality and scale of knowledge, yet this integration might inadvertently raise privacy concerns for stakeholders. In response to this challenge, Federated Learning (FL) is harnessed to improve the knowledge base quality by cryptically leveraging other stakeholders’ knowledge, where knowledge base is represented in Knowledge Graph (KG) form. Initially, a multi-field hyperbolic (MFH) graph embedding method vectorizes entities, furnishing mathematical representations in lieu of solely semantic meanings. The FL framework subsequently encrypted identifies and fuses common entities, whereby the updated entities’ embedding can refine other private entities’ embedding locally, thus enhancing the overall KG quality. Finally, the KG complement method refines and clarifies triplets to improve the overall quality of the KG. An experiment assesses the proposed approach across different industrial KGs, confirming its effectiveness as a viable solution for collaborative KG creation, all while maintaining data security.
{"title":"Unlocking Large Language Model Power in Industry: Privacy-Preserving Collaborative Creation of Knowledge Graph","authors":"Liqiao Xia;Junming Fan;Ajith Parlikad;Xiao Huang;Pai Zheng","doi":"10.1109/TBDATA.2024.3522814","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3522814","url":null,"abstract":"Semantic expertise remains a reliable foundation for industrial decision-making, while Large Language Models (LLMs) can augment the often limited empirical knowledge by generating domain-specific insights, though the quality of this generative knowledge is uncertain. Integrating LLMs with the collective wisdom of multiple stakeholders could enhance the quality and scale of knowledge, yet this integration might inadvertently raise privacy concerns for stakeholders. In response to this challenge, Federated Learning (FL) is harnessed to improve the knowledge base quality by cryptically leveraging other stakeholders’ knowledge, where knowledge base is represented in Knowledge Graph (KG) form. Initially, a multi-field hyperbolic (MFH) graph embedding method vectorizes entities, furnishing mathematical representations in lieu of solely semantic meanings. The FL framework subsequently encrypted identifies and fuses common entities, whereby the updated entities’ embedding can refine other private entities’ embedding locally, thus enhancing the overall KG quality. Finally, the KG complement method refines and clarifies triplets to improve the overall quality of the KG. An experiment assesses the proposed approach across different industrial KGs, confirming its effectiveness as a viable solution for collaborative KG creation, all while maintaining data security.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2046-2060"},"PeriodicalIF":7.5,"publicationDate":"2024-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597812","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-12-26DOI: 10.1109/TBDATA.2024.3522804
Jiajun Sun;Dianliang Wu
The promising applications of mobile crowdsensing (MCS) have attracted much research interest recently, especially for the posted-pricing scenes. However, existing works mainly focus on the stationary MCS, no matter whether in a stochastic or adversarial environment, where each price (or arm) remains identical over time. However, in many realistic MCS applications such as environment monitoring and recommendation systems, stationary bandits do not model the posted-pricing sequential decision problems where the reward distributions of each price (arm) and cost distribution vary over time due to the changes in light intensity and mobile devices’ remnant energy. While in this paper, we study a more general submodular crowdsensing scene to address the non-stationary sequential pricing problems, and construct a monotonic submodular function merging the marginal reward and temporal difference errors (TD-errors) of deep reinforcement learning (DRL). Moreover, we explore a weighted budget-limited non-stationary pricing mechanism by using the deep deterministic policy gradient (DDPG) method for submodular MCS from the perspectives of the hard-drop and soft-drop weights. Our mechanism can readily be extended to non-submodular MCS or other MCS scenes. Extensive simulations demonstrate that our mechanism outweighs existing benchmarks.
{"title":"Online Non-Stationary Pricing Incentives for Budget-Limited Crowdsensing","authors":"Jiajun Sun;Dianliang Wu","doi":"10.1109/TBDATA.2024.3522804","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3522804","url":null,"abstract":"The promising applications of mobile crowdsensing (MCS) have attracted much research interest recently, especially for the posted-pricing scenes. However, existing works mainly focus on the stationary MCS, no matter whether in a stochastic or adversarial environment, where each price (or arm) remains identical over time. However, in many realistic MCS applications such as environment monitoring and recommendation systems, stationary bandits do not model the posted-pricing sequential decision problems where the reward distributions of each price (arm) and cost distribution vary over time due to the changes in light intensity and mobile devices’ remnant energy. While in this paper, we study a more general submodular crowdsensing scene to address the non-stationary sequential pricing problems, and construct a monotonic submodular function merging the marginal reward and temporal difference errors (TD-errors) of deep reinforcement learning (DRL). Moreover, we explore a weighted budget-limited non-stationary pricing mechanism by using the deep deterministic policy gradient (DDPG) method for submodular MCS from the perspectives of the hard-drop and soft-drop weights. Our mechanism can readily be extended to non-submodular MCS or other MCS scenes. Extensive simulations demonstrate that our mechanism outweighs existing benchmarks.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2025-2035"},"PeriodicalIF":7.5,"publicationDate":"2024-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-12-26DOI: 10.1109/TBDATA.2024.3522805
Liner Yang;Jiaxin Yuan;Cunliang Kong;Jingsi Yu;Ruining Chong;Zhenghao Liu;Erhong Yang
The task of complexity-controllable definition generation refers to providing definitions with different readability for words in specific contexts. This task can be utilized to help language learners eliminate reading barriers and facilitate language acquisition. However, the available training data for this task remains scarce due to the difficulty of obtaining reliable definition data and the high cost of data standardization. To tackle those challenges, we introduce a general solution from both the data-driven and method-driven perspectives. We construct a large-scale standard Chinese dataset, COMPILING, which contains both difficult and simple definitions and can serve as a benchmark for future research. Besides, we propose a multitasking framework SimpDefiner for unsupervised controllable definition generation. By designing a parameter-sharing scheme between two decoders, the framework can extract the complexity information from the non-parallel corpus. Moreover, we propose the SimpDefiner guided prompting (SGP) method, where simple definitions generated by SimpDefiner are utilized to construct prompts for GPT-4, hence obtaining more realistic and contextually appropriate definitions. The results demonstrate SimpDefiner's outstanding ability to achieve controllable generation and better results could be achieved when GPT-4 is incorporated.
{"title":"Tailored Definitions With Easy Reach: Complexity-Controllable Definition Generation","authors":"Liner Yang;Jiaxin Yuan;Cunliang Kong;Jingsi Yu;Ruining Chong;Zhenghao Liu;Erhong Yang","doi":"10.1109/TBDATA.2024.3522805","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3522805","url":null,"abstract":"The task of complexity-controllable definition generation refers to providing definitions with different readability for words in specific contexts. This task can be utilized to help language learners eliminate reading barriers and facilitate language acquisition. However, the available training data for this task remains scarce due to the difficulty of obtaining reliable definition data and the high cost of data standardization. To tackle those challenges, we introduce a general solution from both the data-driven and method-driven perspectives. We construct a large-scale standard Chinese dataset, COMPILING, which contains both difficult and simple definitions and can serve as a benchmark for future research. Besides, we propose a multitasking framework SimpDefiner for unsupervised controllable definition generation. By designing a parameter-sharing scheme between two decoders, the framework can extract the complexity information from the non-parallel corpus. Moreover, we propose the SimpDefiner guided prompting (SGP) method, where simple definitions generated by SimpDefiner are utilized to construct prompts for GPT-4, hence obtaining more realistic and contextually appropriate definitions. The results demonstrate SimpDefiner's outstanding ability to achieve controllable generation and better results could be achieved when GPT-4 is incorporated.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2061-2071"},"PeriodicalIF":7.5,"publicationDate":"2024-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597737","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-12-26DOI: 10.1109/TBDATA.2024.3522817
Shicheng Cui;Deqiang Li;Jing Zhang
Graph Neural Networks (GNNs) have been proven to be useful for learning graph-based knowledge. However, one of the drawbacks of GNN techniques is that they may get stuck in the problem of over-squashing. Recent studies attribute to the message passing paradigm that it may amplify some specific local relations and distort long-range information under a certain GNN. To alleviate such phenomena, we propose a novel and general GNN framework, dubbed MC-GNN, which introduces the multi-channel neural architecture to learn and fuse multi-view graph-based information. The purpose of MC-GNN is to extract distinct channel-based graph features and adaptively adjust the importance of the features. To this end, we use the Hilbert-Schmidt Independence Criterion (HSIC) to enlarge the disparity between the embeddings encoded by each channel and follow an attention mechanism to fuse the embeddings with adaptive weight adjustment. MC-GNN can apply multiple GNN backbones, which provides a solution for learning structural relations from a multi-view perspective. Experimental results demonstrate that the proposed MC-GNN is superior to the compared state-of-the-art GNN methods.
{"title":"MC-GNN: Multi-Channel Graph Neural Networks With Hilbert-Schmidt Independence Criterion","authors":"Shicheng Cui;Deqiang Li;Jing Zhang","doi":"10.1109/TBDATA.2024.3522817","DOIUrl":"https://doi.org/10.1109/TBDATA.2024.3522817","url":null,"abstract":"Graph Neural Networks (GNNs) have been proven to be useful for learning graph-based knowledge. However, one of the drawbacks of GNN techniques is that they may get stuck in the problem of over-squashing. Recent studies attribute to the message passing paradigm that it may amplify some specific local relations and distort long-range information under a certain GNN. To alleviate such phenomena, we propose a novel and general GNN framework, dubbed MC-GNN, which introduces the multi-channel neural architecture to learn and fuse multi-view graph-based information. The purpose of MC-GNN is to extract distinct channel-based graph features and adaptively adjust the importance of the features. To this end, we use the Hilbert-Schmidt Independence Criterion (HSIC) to enlarge the disparity between the embeddings encoded by each channel and follow an attention mechanism to fuse the embeddings with adaptive weight adjustment. MC-GNN can apply multiple GNN backbones, which provides a solution for learning structural relations from a multi-view perspective. Experimental results demonstrate that the proposed MC-GNN is superior to the compared state-of-the-art GNN methods.","PeriodicalId":13106,"journal":{"name":"IEEE Transactions on Big Data","volume":"11 4","pages":"2036-2045"},"PeriodicalIF":7.5,"publicationDate":"2024-12-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144597728","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：