Information Processing & Management最新文献_第5页

Linguistic patterns in social media content from crisis and non-crisis zones: A case study of Hurricane Ian

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-20 DOI: 10.1016/j.ipm.2025.104061

Ly Dinh, Steven Walczak

Social media platforms, particularly Twitter, play a vital role in crisis response by delivering real-time information about affected populations. To enhance the accurate detection of crisis-relevant content, this study investigates linguistic distinctions between crisis and non-crisis zones. By analyzing over 263,000 tweets from within and outside the 2022 Hurricane Ian’s impact zone, we examine normalized word frequency, syntactic categories (nouns, verbs, adjectives, adverbs), sentiment, and user interaction patterns in the tweet networks. Our findings reveal a consistent power-law distribution in the relative differences of word use between crisis and non-crisis zones. Syntactic categories differences, particularly in adjectives, highlight the crisis zone’s emphasis on the hurricane’s path and impact, while the non-crisis zone’s vocabulary centers on current news topics such as sports, politics, and leisure. Syntactic analyses show that 36% (N = 20,967) of words are used in both crisis and non-crisis zones, 29% (N = 17,168) are unique to the crisis zone, and another 35% (N = 20,101) are unique to the non-crisis zone, highlighting the broader range of topics discussed in the non-crisis zone compared to the crisis zone. Sentiment analysis indicates comparable distributions of neutral words (

\sim

99%), followed by negative words (

\sim

0.4%) and positive words (

\sim

0.4%). However, the use of profanity, indicating strong negative sentiment, occurred 19% more frequently in non-crisis zone tweets than in crisis zone tweets. Network analysis and network modeling show that the crisis zone network is denser and more cohesive, reflecting tight-knit communities during crises, whereas the non-crisis zone network is larger and more fragmented, indicating diverse user engagements. Our study’s contributions include providing insights into the distinctive usage of words in crisis and non-crisis zones, hence facilitating the evaluation of crisis-relevant language patterns. Ultimately, the findings may be used to aid responders in prioritizing urgent tweets originating from a crisis zone.

{"title":"Linguistic patterns in social media content from crisis and non-crisis zones: A case study of Hurricane Ian","authors":"Ly Dinh, Steven Walczak","doi":"10.1016/j.ipm.2025.104061","DOIUrl":"10.1016/j.ipm.2025.104061","url":null,"abstract":"<div><div>Social media platforms, particularly Twitter, play a vital role in crisis response by delivering real-time information about affected populations. To enhance the accurate detection of crisis-relevant content, this study investigates linguistic distinctions between crisis and non-crisis zones. By analyzing over 263,000 tweets from within and outside the 2022 Hurricane Ian’s impact zone, we examine normalized word frequency, syntactic categories (nouns, verbs, adjectives, adverbs), sentiment, and user interaction patterns in the tweet networks. Our findings reveal a consistent power-law distribution in the relative differences of word use between crisis and non-crisis zones. Syntactic categories differences, particularly in adjectives, highlight the crisis zone’s emphasis on the hurricane’s path and impact, while the non-crisis zone’s vocabulary centers on current news topics such as sports, politics, and leisure. Syntactic analyses show that 36% (N = 20,967) of words are used in both crisis and non-crisis zones, 29% (N = 17,168) are unique to the crisis zone, and another 35% (N = 20,101) are unique to the non-crisis zone, highlighting the broader range of topics discussed in the non-crisis zone compared to the crisis zone. Sentiment analysis indicates comparable distributions of neutral words (<span><math><mo>∼</mo></math></span> 99%), followed by negative words (<span><math><mo>∼</mo></math></span> 0.4%) and positive words (<span><math><mo>∼</mo></math></span> 0.4%). However, the use of profanity, indicating strong negative sentiment, occurred 19% more frequently in non-crisis zone tweets than in crisis zone tweets. Network analysis and network modeling show that the crisis zone network is denser and more cohesive, reflecting tight-knit communities during crises, whereas the non-crisis zone network is larger and more fragmented, indicating diverse user engagements. Our study’s contributions include providing insights into the distinctive usage of words in crisis and non-crisis zones, hence facilitating the evaluation of crisis-relevant language patterns. Ultimately, the findings may be used to aid responders in prioritizing urgent tweets originating from a crisis zone.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104061"},"PeriodicalIF":7.4,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploiting diffusion-based structured learning for item interactions representations in multimodal recommender systems

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-20 DOI: 10.1016/j.ipm.2025.104075

Nikhat Khan, Dilip Singh Sisodia

Multimodal Recommender Systems (MRS) enhance the performance of recommendations by utilizing different item information, such as text, images, and audio. Existing non-graph-based MRS techniques combine embeddings (i.e., id and multimodal embedding) but ignore indirect and higher-order interactions. Graph-based MRS approaches use graph sparsification (GS) to construct item graphs and graph convolutional networks (GCNs) for higher-order interactions. However, GS reduces the item graph size, while GCNs ignore specific information due to their predefined weights. Hence, to mitigate the mentioned issues, this study proposes a Diffusion-based Structured Learning technique for Multimodal Recommender Systems (DSL-MRS) that improves the latent item graph information flow while maintaining its structure. Additionally, we used a graph attention neural network (GANN) to represent complex higher-order item-item interactions and implemented an attention mechanism to prioritize relevant nodes by assigning weights to neighbour. Also, for optimization, a Weighted Approximate-Rank pairwise (WARP) loss function has been used to prioritize predictions for observed items over those for unspecified items. To demonstrate the advantage of DSL-MRS, we conducted extensive experiments on three publicly available categories of Amazon datasets. The experimental findings showed that the proposed approach led to an average improvement of 5.8 % in R@20, 8.7 % in precision@20,7.8 % in NDCG@20 and 8.8 % in F-score@20 compared to the baseline model. Ablation studies demonstrate the value and efficacy of DSL-MRS, as its components degrade performance when removed.

{"title":"Exploiting diffusion-based structured learning for item interactions representations in multimodal recommender systems","authors":"Nikhat Khan, Dilip Singh Sisodia","doi":"10.1016/j.ipm.2025.104075","DOIUrl":"10.1016/j.ipm.2025.104075","url":null,"abstract":"<div><div>Multimodal Recommender Systems (MRS) enhance the performance of recommendations by utilizing different item information, such as text, images, and audio. Existing non-graph-based MRS techniques combine embeddings (i.e., id and multimodal embedding) but ignore indirect and higher-order interactions. Graph-based MRS approaches use graph sparsification (GS) to construct item graphs and graph convolutional networks (GCNs) for higher-order interactions. However, GS reduces the item graph size, while GCNs ignore specific information due to their predefined weights. Hence, to mitigate the mentioned issues, this study proposes a Diffusion-based Structured Learning technique for Multimodal Recommender Systems (DSL-MRS) that improves the latent item graph information flow while maintaining its structure. Additionally, we used a graph attention neural network (GANN) to represent complex higher-order item-item interactions and implemented an attention mechanism to prioritize relevant nodes by assigning weights to neighbour. Also, for optimization, a Weighted Approximate-Rank pairwise (WARP) loss function has been used to prioritize predictions for observed items over those for unspecified items. To demonstrate the advantage of DSL-MRS, we conducted extensive experiments on three publicly available categories of Amazon datasets. The experimental findings showed that the proposed approach led to an average improvement of 5.8 % in R@20, 8.7 % in precision@20,7.8 % in NDCG@20 and 8.8 % in F-score@20 compared to the baseline model. Ablation studies demonstrate the value and efficacy of DSL-MRS, as its components degrade performance when removed.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104075"},"PeriodicalIF":7.4,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139107","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Amplifying commonsense knowledge via bi-directional relation integrated graph-based contrastive pre-training from large language models

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-20 DOI: 10.1016/j.ipm.2025.104068

Liu Yu, Fenghui Tian, Ping Kuang, Fan Zhou

Commonsense knowledge graph acquisition (CKGA) is vital in numerous knowledge-intensive applications such as question-answering and knowledge reasoning. Conventional CKGA methods rely on node-level and unidirectional relations, making them suffer from a shallow grasp of between entities and relations. Moreover, they also demand expensive, labor-intensive human annotations, and the yielding CK lacks diversity and quality. Existing commonsense knowledge bases such as ConceptNet or ATOMIC often struggle with significant scarcity and pose a major challenge in meeting the high demand for a vast amount of commonsense information. Given the recent momentum of large language models (LLMs), there is growing interest in leveraging them to overcome the above challenges.

In this study, we propose a new paradigm to amplify commonsense knowledge via bi-directional relation integrated graph-based contrastive pre-training (BIRGHT) from the newest foundation models. BRIGHT is an integral and closed-loop framework composed of corpora construction, further contrastive pre-training, task-driven instruction tuning, filtering strategy, and an evaluation system. The key of BRIGHT is to leverage reverse relations to create a symmetric graph and transform the bi-directional relations into sentence-level ones. The reverse sentences are considered positive examples for forward sentences, and three types of negatives are introduced to ensure efficient contrastive learning, which mitigates the “reversal curse” issue as evidenced in experiments. Empirical results demonstrate that BRIGHT is able to generate novel knowledge (up to 397K) and that the GPT-4 acceptance rate is high quality, with up to 90.51% (ATOMIC) and 85.59% (ConceptNet) accuracy at top 1, which approaches human performance for these resources. Our BRIGHT is publicly available at https://github.com/GreyHuu/BRIGHT/tree/main.

{"title":"Amplifying commonsense knowledge via bi-directional relation integrated graph-based contrastive pre-training from large language models","authors":"Liu Yu, Fenghui Tian, Ping Kuang, Fan Zhou","doi":"10.1016/j.ipm.2025.104068","DOIUrl":"10.1016/j.ipm.2025.104068","url":null,"abstract":"<div><div>Commonsense knowledge graph acquisition (CKGA) is vital in numerous knowledge-intensive applications such as question-answering and knowledge reasoning. Conventional CKGA methods rely on node-level and unidirectional relations, making them suffer from a shallow grasp of between entities and relations. Moreover, they also demand expensive, labor-intensive human annotations, and the yielding CK lacks diversity and quality. Existing commonsense knowledge bases such as ConceptNet or ATOMIC often struggle with significant scarcity and pose a major challenge in meeting the high demand for a vast amount of commonsense information. Given the recent momentum of large language models (LLMs), there is growing interest in leveraging them to overcome the above challenges.</div><div>In this study, we propose a new paradigm to amplify commonsense knowledge via <u>b</u>i-di<u>r</u>ect<u>i</u>onal relation integrated <u>g</u>rap<u>h</u>-based con<u>t</u>rastive pre-training (<strong>BIRGHT</strong>) from the newest foundation models. BRIGHT is an integral and closed-loop framework composed of corpora construction, further contrastive pre-training, task-driven instruction tuning, filtering strategy, and an evaluation system. The key of BRIGHT is to leverage reverse relations to create a symmetric graph and transform the bi-directional relations into sentence-level ones. The reverse sentences are considered positive examples for forward sentences, and three types of negatives are introduced to ensure efficient contrastive learning, which mitigates the “reversal curse” issue as evidenced in experiments. Empirical results demonstrate that BRIGHT is able to generate novel knowledge (up to 397K) and that the GPT-4 acceptance rate is high quality, with up to 90.51% (ATOMIC) and 85.59% (ConceptNet) accuracy at top 1, which approaches human performance for these resources. Our BRIGHT is publicly available at <span><span>https://github.com/GreyHuu/BRIGHT/tree/main</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104068"},"PeriodicalIF":7.4,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139097","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Beyond expression: Comprehensive visualization of knowledge triplet facts

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-18 DOI: 10.1016/j.ipm.2025.104062

Wei Liu , Yixue He , Chao Wang , Shaorong Xie , Weimin Li

Multi-modal Knowledge Graphs (KGs) enhance traditional KGs by incorporating multi-modal data to bridge the information gap in natural language processing (NLP) tasks. One direct method to incorporate multi-modal data is to associate structured KG with corresponding image modalities, thereby visualizing entities and triplet facts. However, existing visualization methods for triplet facts often exclude triplet facts containing abstract entities and non-visual relations, resulting in their disassociation from corresponding image modalities. This exclusion compromises the completeness and utility of multi-modal KGs. In this paper, we aim to construct a comprehensive multi-modal KG that includes abstract entities and non-visual relations, ensuring complete visualization of every triplet fact. To achieve this purpose, we propose a method for the integration of image Retrieval-Generation-Editing (RGE) to completely and accurately visualize each triplet fact. Initially, we correct the triplet facts by integrating a Large Language Model (LLM) with a retrieved knowledge database about triplet facts. Subsequently, by providing appropriate contextual examples to the LLM, we generate visual elements of relations, enriching the semantics of the triplet facts. We then employ image retrieval to obtain images that reflect the semantics of each triplet fact. For those triplet facts for which images cannot be directly retrieved, we utilize image generation and editing to create and modify images that can express the semantics of the triplet facts. Through the RGE method, we construct a multi-modal KG named DB15kFact, which includes 86,722 triplet facts, 274 relations, 12,842 entities, and 387,096 images. The construction of DB15kFact has resulted in a fourfold increase in the number of relations compared to the previous multi-modal KG, ImgFact. In experiments, both automatic and manual evaluations confirm the quality of DB15kFact. The results demonstrate that the DB15kFact significantly enhances model performance in link prediction and relation classification. Notably, in link prediction, the model optimized with DB15kFact achieves a 7.12% improvement in the H@10 metric compared to existing solutions.

{"title":"Beyond expression: Comprehensive visualization of knowledge triplet facts","authors":"Wei Liu , Yixue He , Chao Wang , Shaorong Xie , Weimin Li","doi":"10.1016/j.ipm.2025.104062","DOIUrl":"10.1016/j.ipm.2025.104062","url":null,"abstract":"<div><div>Multi-modal Knowledge Graphs (KGs) enhance traditional KGs by incorporating multi-modal data to bridge the information gap in natural language processing (NLP) tasks. One direct method to incorporate multi-modal data is to associate structured KG with corresponding image modalities, thereby visualizing entities and triplet facts. However, existing visualization methods for triplet facts often exclude triplet facts containing abstract entities and non-visual relations, resulting in their disassociation from corresponding image modalities. This exclusion compromises the completeness and utility of multi-modal KGs. In this paper, we aim to construct a comprehensive multi-modal KG that includes abstract entities and non-visual relations, ensuring complete visualization of every triplet fact. To achieve this purpose, we propose a method for the integration of image <strong>R</strong>etrieval-<strong>G</strong>eneration-<strong>E</strong>diting (RGE) to completely and accurately visualize each triplet fact. Initially, we correct the triplet facts by integrating a Large Language Model (LLM) with a retrieved knowledge database about triplet facts. Subsequently, by providing appropriate contextual examples to the LLM, we generate visual elements of relations, enriching the semantics of the triplet facts. We then employ image retrieval to obtain images that reflect the semantics of each triplet fact. For those triplet facts for which images cannot be directly retrieved, we utilize image generation and editing to create and modify images that can express the semantics of the triplet facts. Through the RGE method, we construct a multi-modal KG named <span>DB15kFact</span>, which includes 86,722 triplet facts, 274 relations, 12,842 entities, and 387,096 images. The construction of <span>DB15kFact</span> has resulted in a fourfold increase in the number of relations compared to the previous multi-modal KG, ImgFact. In experiments, both automatic and manual evaluations confirm the quality of <span>DB15kFact</span>. The results demonstrate that the <span>DB15kFact</span> significantly enhances model performance in link prediction and relation classification. Notably, in link prediction, the model optimized with <span>DB15kFact</span> achieves a 7.12% improvement in the H@10 metric compared to existing solutions.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104062"},"PeriodicalIF":7.4,"publicationDate":"2025-01-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139414","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Historical facts learning from Long-Short Terms with Language Model for Temporal Knowledge Graph Reasoning

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-17 DOI: 10.1016/j.ipm.2024.104047

Wenjie Xu , Ben Liu , Miao Peng , Zihao Jiang , Xu Jia , Kai Liu , Lei Liu , Min Peng

Temporal Knowledge Graph Reasoning (TKGR) aims to reason the missing parts in TKGs based on historical facts from different time periods. Traditional GCN-based TKGR models depend on structured relations between entities. To utilize the rich linguistic information in TKGs, some models have focused on applying pre-trained language models (PLMs) to TKGR. However, previous PLM-based models still face some issues: (1) they did not mine the associations in relations; (2) they did not differentiate the impact of historical facts from different time periods. (3) they introduced external knowledge to enhance the performance without fully utilizing the inherent reasoning capabilities of PLMs. To deal with these issues, we propose HFL: Historical Facts Learning from Long-Short Terms with Language Model for TKGR. Firstly, we construct time tokens for different types of time intervals to use timestamps and input the historical facts relevant to the query into the PLMs to learn the associations in relations. Secondly, we take a multi-perspective sampling strategy to learn from different time periods and use the original text information in TKGs or even no text information to learn reasoning abilities without any external knowledge. Finally, we perform HFL on four TKGR benchmarks, and the experiment results demonstrate that HFL has great competitiveness compared to both graph-based and PLM-based models. Additionally, we design a variant that applies HFL to LLMs and evaluate the performance of different LLMs.

{"title":"Historical facts learning from Long-Short Terms with Language Model for Temporal Knowledge Graph Reasoning","authors":"Wenjie Xu , Ben Liu , Miao Peng , Zihao Jiang , Xu Jia , Kai Liu , Lei Liu , Min Peng","doi":"10.1016/j.ipm.2024.104047","DOIUrl":"10.1016/j.ipm.2024.104047","url":null,"abstract":"<div><div>Temporal Knowledge Graph Reasoning (TKGR) aims to reason the missing parts in TKGs based on historical facts from different time periods. Traditional GCN-based TKGR models depend on structured relations between entities. To utilize the rich linguistic information in TKGs, some models have focused on applying pre-trained language models (PLMs) to TKGR. However, previous PLM-based models still face some issues: (1) they did not mine the associations in relations; (2) they did not differentiate the impact of historical facts from different time periods. (3) they introduced external knowledge to enhance the performance without fully utilizing the inherent reasoning capabilities of PLMs. To deal with these issues, we propose HFL: <strong>H</strong>istorical <strong>F</strong>acts <strong>L</strong>earning from Long-Short Terms with Language Model for TKGR. Firstly, we construct time tokens for different types of time intervals to use timestamps and input the historical facts relevant to the query into the PLMs to learn the associations in relations. Secondly, we take a multi-perspective sampling strategy to learn from different time periods and use the original text information in TKGs or even no text information to learn reasoning abilities without any external knowledge. Finally, we perform HFL on four TKGR benchmarks, and the experiment results demonstrate that HFL has great competitiveness compared to both graph-based and PLM-based models. Additionally, we design a variant that applies HFL to LLMs and evaluate the performance of different LLMs.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104047"},"PeriodicalIF":7.4,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

User identification network with contrastive clustering for shared-account recommendation

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-15 DOI: 10.1016/j.ipm.2024.104055

Xinhua Wang , Houping Yue , Lei Guo , Feng Guo , Chen He , Xiaohui Han

The Shared-Account Recommendation (SAR) aims to accurately identify and accommodate the varied preferences of multiple users sharing a single account by analyzing their aggregated interactions. SAR faces challenges in preference identification when multiple users share an account. Existing Shared-Account Modeling (SAM) methods assume overly simplistic conditions and overlook the robustness of representations, leading to inaccurate embeddings that are susceptible to disturbances. To address limitations in existing SAR methods, we introduce the Contrastive Clustering User Identification Network (CCUI-Net) framework to enhance SAR. This framework employs graph-based transformations and node representation learning to refine user embeddings, utilizes hierarchical contrastive clustering for improved user identification and robustness against data noise, and leverages an attention mechanism to dynamically balance contributions from various users. These innovations significantly boost the precision and reliability of recommendations. Experimental results across four domains from the HVIDEO and HAMAZON datasets (E-domain and V-domain in HVIDEO, M-domain and B-domain in HAMAZON) demonstrate that CCUI-Net exceeds the performance of many existing available methods on the metrics MRR@5, MRR@20, Recall@5, and Recall@20. Specifically, the improvements in the M-domain and B-domain for Recall@5 and Recall@20 are 14.64%, 8.55%, 18.67%, and 9.59% respectively.

{"title":"User identification network with contrastive clustering for shared-account recommendation","authors":"Xinhua Wang , Houping Yue , Lei Guo , Feng Guo , Chen He , Xiaohui Han","doi":"10.1016/j.ipm.2024.104055","DOIUrl":"10.1016/j.ipm.2024.104055","url":null,"abstract":"<div><div>The Shared-Account Recommendation (SAR) aims to accurately identify and accommodate the varied preferences of multiple users sharing a single account by analyzing their aggregated interactions. SAR faces challenges in preference identification when multiple users share an account. Existing Shared-Account Modeling (SAM) methods assume overly simplistic conditions and overlook the robustness of representations, leading to inaccurate embeddings that are susceptible to disturbances. To address limitations in existing SAR methods, we introduce the <strong>C</strong>ontrastive <strong>C</strong>lustering <strong>U</strong>ser <strong>I</strong>dentification Network (CCUI-Net) framework to enhance SAR. This framework employs graph-based transformations and node representation learning to refine user embeddings, utilizes hierarchical contrastive clustering for improved user identification and robustness against data noise, and leverages an attention mechanism to dynamically balance contributions from various users. These innovations significantly boost the precision and reliability of recommendations. Experimental results across four domains from the HVIDEO and HAMAZON datasets (E-domain and V-domain in HVIDEO, M-domain and B-domain in HAMAZON) demonstrate that CCUI-Net exceeds the performance of many existing available methods on the metrics MRR@5, MRR@20, Recall@5, and Recall@20. Specifically, the improvements in the M-domain and B-domain for Recall@5 and Recall@20 are 14.64%, 8.55%, 18.67%, and 9.59% respectively.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104055"},"PeriodicalIF":7.4,"publicationDate":"2025-01-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139103","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A unified framework for multi-modal rumor detection via multi-level dynamic interaction with evolving stances

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-14 DOI: 10.1016/j.ipm.2025.104066

Tiening Sun, Chengwei Liu, Lizhi Chen, Zhong Qian, Peifeng Li, Qiaoming Zhu

With the escalating dissemination of textual and visual content on the Internet, multi-modal rumor detection has garnered significant scholarly attention in recent research studies. Currently, the prevailing methods in multi-modal rumor detection tend to emphasize the information integration from source posts and images, overlooking the dynamic interaction between multi-modal sources and evolving conversational structures. Furthermore, they fail to recognize the potential advantage that introducing evolving user stances as a form of collective decision-making can improve the model’s performance in rumor classification. In this paper, we propose a novel Evolving Stance-aware Dynamic Graph Fusion Network (ESDGFN) to address the above issues. This network aims to integrate the source, the image and the dynamic conversation graph into a unified framework. Specifically, we begin by leveraging a cross-modal transformer for fine-grained feature fusion of the multi-modal sources. Simultaneously, based on the temporal attributes of posts, we construct a set of dynamically changing conversation graphs for each conversation thread, simulating and encoding the evolving stances of users towards the target event within these conversation graphs. Subsequently, we design a multi-level fusion strategy, incorporating both coarse-grained multi-modal feature guidance and fine-grained cross-modal similarity-aware fusion. This strategy aims to generate interactively enhanced multi-modal encoding and dynamic graph representations. The experimental results on both PHEME and Twitter datasets highlight the excellence of our ESDGFN model. It achieves 90.6% accuracy on PHEME, a 3.3% improvement compared to the state-of-the-art method, and 87% accuracy on Twitter, with a 2.4% improvement.

{"title":"A unified framework for multi-modal rumor detection via multi-level dynamic interaction with evolving stances","authors":"Tiening Sun, Chengwei Liu, Lizhi Chen, Zhong Qian, Peifeng Li, Qiaoming Zhu","doi":"10.1016/j.ipm.2025.104066","DOIUrl":"10.1016/j.ipm.2025.104066","url":null,"abstract":"<div><div>With the escalating dissemination of textual and visual content on the Internet, multi-modal rumor detection has garnered significant scholarly attention in recent research studies. Currently, the prevailing methods in multi-modal rumor detection tend to emphasize the information integration from source posts and images, overlooking the dynamic interaction between multi-modal sources and evolving conversational structures. Furthermore, they fail to recognize the potential advantage that introducing evolving user stances as a form of collective decision-making can improve the model’s performance in rumor classification. In this paper, we propose a novel Evolving Stance-aware Dynamic Graph Fusion Network (ESDGFN) to address the above issues. This network aims to integrate the source, the image and the dynamic conversation graph into a unified framework. Specifically, we begin by leveraging a cross-modal transformer for fine-grained feature fusion of the multi-modal sources. Simultaneously, based on the temporal attributes of posts, we construct a set of dynamically changing conversation graphs for each conversation thread, simulating and encoding the evolving stances of users towards the target event within these conversation graphs. Subsequently, we design a multi-level fusion strategy, incorporating both coarse-grained multi-modal feature guidance and fine-grained cross-modal similarity-aware fusion. This strategy aims to generate interactively enhanced multi-modal encoding and dynamic graph representations. The experimental results on both PHEME and Twitter datasets highlight the excellence of our ESDGFN model. It achieves 90.6% accuracy on PHEME, a 3.3% improvement compared to the state-of-the-art method, and 87% accuracy on Twitter, with a 2.4% improvement.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104066"},"PeriodicalIF":7.4,"publicationDate":"2025-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139415","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Confusing negative commonsense knowledge generation with hierarchy modeling and LLM-enhanced filtering

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-14 DOI: 10.1016/j.ipm.2025.104060

Yaqing Sheng, Weixin Zeng, Jiuyang Tang, Lihua Liu, Xiang Zhao

While most of the world’s knowledge exists in a positive and affirmative form, negative knowledge also plays a significant role by showing what is not true or what not to think, and has yet been largely overlooked. Existing negative commonsense knowledge generation methods adopt the generation-filtering paradigm, while the produced negative statements are easy to detect and fail to contribute to both human perception and task-specific algorithms that require negative samples for training. In response, we put forward CONEG, a negative commonsense knowledge generation framework that generates confusing statements, featuring hierarchy modeling in candidate generation and LLM-enhanced two-stage filtering. Specifically, in the candidate generation stage, we identify congeners for entity phrases in the commonsense knowledge base using box embeddings, which can effectively capture the hierarchical correlations among entity phrases and produce confusing candidates. In the candidate filtering stage, we design a two-stage filtering strategy, consisting of intrinsic triple confidence measuring and extrinsic refinement through large language models with group-based instructions, which can effectively filter out true facts and low-quality negative candidates. We empirically evaluate our proposal on both intrinsic assessment and downstream tasks, and the results demonstrate that CONEG and its components are effective in terms of producing confusing negative knowledge, surpassing the state-of-the-art methods.

{"title":"Confusing negative commonsense knowledge generation with hierarchy modeling and LLM-enhanced filtering","authors":"Yaqing Sheng, Weixin Zeng, Jiuyang Tang, Lihua Liu, Xiang Zhao","doi":"10.1016/j.ipm.2025.104060","DOIUrl":"10.1016/j.ipm.2025.104060","url":null,"abstract":"<div><div>While most of the world’s knowledge exists in a positive and affirmative form, negative knowledge also plays a significant role by showing what is not true or what not to think, and has yet been largely overlooked. Existing negative commonsense knowledge generation methods adopt the generation-filtering paradigm, while the produced negative statements are easy to detect and fail to contribute to both human perception and task-specific algorithms that require negative samples for training. In response, we put forward <span>CONEG</span>, a negative commonsense knowledge generation framework that generates <em>confusing</em> statements, featuring hierarchy modeling in candidate generation and LLM-enhanced two-stage filtering. Specifically, in the <em>candidate generation</em> stage, we identify <em>congeners</em> for entity phrases in the commonsense knowledge base using box embeddings, which can effectively capture the hierarchical correlations among entity phrases and produce confusing candidates. In the <em>candidate filtering</em> stage, we design a two-stage filtering strategy, consisting of <em>intrinsic</em> triple confidence measuring and <em>extrinsic</em> refinement through large language models with group-based instructions, which can effectively filter out true facts and low-quality negative candidates. We empirically evaluate our proposal on both intrinsic assessment and downstream tasks, and the results demonstrate that <span>CONEG</span> and its components are effective in terms of producing confusing negative knowledge, surpassing the state-of-the-art methods.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104060"},"PeriodicalIF":7.4,"publicationDate":"2025-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139099","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Disentangled feature graph for Hierarchical Text Classification

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-14 DOI: 10.1016/j.ipm.2025.104065

Renyuan Liu, Xuejie Zhang, Jin Wang, Xiaobing Zhou

Effectively utilizing the hierarchical relationship among labels is the core of Hierarchical Text Classification (HTC). Previous research on HTC has tended to enhance the dependencies between labels. However, they overlook some labels that may conflict with other labels because alleviating label conflicts also weakens label dependencies and reduces the model performance. Therefore, this paper focuses on the issue of label conflicts and studies methods to alleviate label conflicts without affecting the mutual support relationship between labels. To solve the abovementioned problem, we first use the feature disentanglement method to cut off all label connections. Then, the connection among labels is selectively established by constructing a hierarchical graph on disentangled features. Finally, the Graph Neural Networks (GNN) is adopted to encode the obtained Disentanglement Feature Graph (DFG) and enables only labels with connections to support each other, while labels without connections do not interfere with each other. The experimental results on the WOS, RCV1-v2, and BGC datasets show the effectiveness of DFG. In detail, the experimental results show that on the WOS dataset, the model incorporating DFG achieved a 1.07% improvement in Macro-F1, surpassing the best model by 0.27%. On the RCV1-v2 dataset, the model incorporating DFG achieved a 0.95% improvement in Micro-F1, surpassing the best model by 0.21%. On the BGC dataset, the model incorporating DFG achieved a 1.81% improvement in Micro-F1, surpassing the best model by 0.45%.

{"title":"Disentangled feature graph for Hierarchical Text Classification","authors":"Renyuan Liu, Xuejie Zhang, Jin Wang, Xiaobing Zhou","doi":"10.1016/j.ipm.2025.104065","DOIUrl":"10.1016/j.ipm.2025.104065","url":null,"abstract":"<div><div>Effectively utilizing the hierarchical relationship among labels is the core of Hierarchical Text Classification (HTC). Previous research on HTC has tended to enhance the dependencies between labels. However, they overlook some labels that may conflict with other labels because alleviating label conflicts also weakens label dependencies and reduces the model performance. Therefore, this paper focuses on the issue of label conflicts and studies methods to alleviate label conflicts without affecting the mutual support relationship between labels. To solve the abovementioned problem, we first use the feature disentanglement method to cut off all label connections. Then, the connection among labels is selectively established by constructing a hierarchical graph on disentangled features. Finally, the Graph Neural Networks (GNN) is adopted to encode the obtained Disentanglement Feature Graph (DFG) and enables only labels with connections to support each other, while labels without connections do not interfere with each other. The experimental results on the WOS, RCV1-v2, and BGC datasets show the effectiveness of DFG. In detail, the experimental results show that on the WOS dataset, the model incorporating DFG achieved a 1.07% improvement in Macro-F1, surpassing the best model by 0.27%. On the RCV1-v2 dataset, the model incorporating DFG achieved a 0.95% improvement in Micro-F1, surpassing the best model by 0.21%. On the BGC dataset, the model incorporating DFG achieved a 1.81% improvement in Micro-F1, surpassing the best model by 0.45%.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104065"},"PeriodicalIF":7.4,"publicationDate":"2025-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139098","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Why logit distillation works: A novel knowledge distillation technique by deriving target augmentation and logits distortion

IF 7.4 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2025-01-13 DOI: 10.1016/j.ipm.2024.104056

Md Imtiaz Hossain, Sharmen Akhter, Nosin Ibna Mahbub, Choong Seon Hong, Eui-Nam Huh

Although logit distillation aims to transfer knowledge from a large teacher network to a student, the underlying mechanisms and reasons for its effectiveness are unclear. This article explains the effectiveness of knowledge distillation (KD). Based on the observations, this paper proposes a novel distillation technique called TALD-KD that performs through Target Augmentation and a novel concept of dynamic Logits Distortion technique. The proposed TALD-KD unraveled the intricate relationships of dark knowledge semantics, randomness, flexibility, and augmentation with logits-level KD via three different investigations, hypotheses, and observations. TALD-KD improved student generalization through the linear combination of the teacher logits and random noise. Among the three versions assessed (TALD-A, TALD-B, and TALD-C), TALD-B improved the performance of KD on a large-scale ImageNet-1K dataset from 68.87% to 69.58% for top-1 accuracy, and from 88.76% to 90.13% for top-5 accuracy. Similarly, for the state-of-the-art approach, DKD, the performance improvements by the TALD-B ranged from 72.05% to 72.81% for top-1 accuracy and from 91.05% to 92.04% for top-5 accuracy. The other versions revealed the secrets of logit-level KD. Extensive ablation studies confirmed the superiority of the proposed approach over existing state-of-the-art approaches in diverse scenarios.

{"title":"Why logit distillation works: A novel knowledge distillation technique by deriving target augmentation and logits distortion","authors":"Md Imtiaz Hossain, Sharmen Akhter, Nosin Ibna Mahbub, Choong Seon Hong, Eui-Nam Huh","doi":"10.1016/j.ipm.2024.104056","DOIUrl":"10.1016/j.ipm.2024.104056","url":null,"abstract":"<div><div>Although logit distillation aims to transfer knowledge from a large teacher network to a student, the underlying mechanisms and reasons for its effectiveness are unclear. This article explains the effectiveness of knowledge distillation (KD). Based on the observations, this paper proposes a novel distillation technique called TALD-KD that performs through Target Augmentation and a novel concept of dynamic Logits Distortion technique. The proposed TALD-KD unraveled the intricate relationships of dark knowledge semantics, randomness, flexibility, and augmentation with logits-level KD via three different investigations, hypotheses, and observations. TALD-KD improved student generalization through the linear combination of the teacher logits and random noise. Among the three versions assessed (TALD-A, TALD-B, and TALD-C), TALD-B improved the performance of KD on a large-scale ImageNet-1K dataset from 68.87% to 69.58% for top-1 accuracy, and from 88.76% to 90.13% for top-5 accuracy. Similarly, for the state-of-the-art approach, DKD, the performance improvements by the TALD-B ranged from 72.05% to 72.81% for top-1 accuracy and from 91.05% to 92.04% for top-5 accuracy. The other versions revealed the secrets of logit-level KD. Extensive ablation studies confirmed the superiority of the proposed approach over existing state-of-the-art approaches in diverse scenarios.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104056"},"PeriodicalIF":7.4,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143139102","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0