Information Processing & Management最新文献

{"title":"An empirical study of LLMs via in-context learning for stance classification","authors":"Lida Shi ,&nbsp;Fausto Giunchiglia ,&nbsp;Ran Luo ,&nbsp;Daqian Shi ,&nbsp;Rui Song ,&nbsp;Xiaolei Diao ,&nbsp;Hao Xu","doi":"10.1016/j.ipm.2025.104322","DOIUrl":"10.1016/j.ipm.2025.104322","url":null,"abstract":"<div><div>The rapid advancement of large language models (LLMs) creates new research opportunities in stance classification. However, existing studies often lack a systematic evaluation and empirical analysis of the performance of mainstream large models. In this paper, we systematically evaluate the performance of 5 SOTA large language models, including LLaMA, DeepSeek, Qwen, GPT, and Gemini, on stance classification using 13 benchmark datasets. We explore the effectiveness of two strategies — random selection and semantic similarity selection — within the framework of in-context learning. By comparing these approaches through cross-domain and in-domain experiments, we reveal how they impact model performance and provide insights for future optimization. Overall, this study clarifies the influence of different models and sampling strategies on stance classification performance and suggests directions for further research. Our code is available at: <span><span>https://github.com/shilida/In-context4Stance</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104322"},"PeriodicalIF":6.9,"publicationDate":"2025-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144809371","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}

{"title":"Iterative update scheme for nonnegative and sparse linear autoencoders in recommendation","authors":"Xuan Li,&nbsp;Shifei Ding","doi":"10.1016/j.ipm.2025.104314","DOIUrl":"10.1016/j.ipm.2025.104314","url":null,"abstract":"<div><div>Linear autoencoder models with nonnegative constraints and L1 regularization, such as the sparse linear method (SLIM), have shown remarkable performance while maintaining interpretability. However, their practicality is limited by computationally expensive training processes. This paper proposes a simple yet effective training framework for nonnegative and sparse linear autoencoders. We first develop a simple iterative update scheme (IUS) for SLIM and present a theoretical analysis of its convergence and correctness. To enhance computational efficiency, we then introduce a filtering step that prunes insignificant parameters at each iteration in practice. Based on this training scheme, we derive two model variants by removing the zero-diagonal constraint and utilizing random dropout denoising to replace L2 regularization (i.e., the dropout-based regularization in DLAE), respectively. Experimental results demonstrate that the proposed IUS algorithm reduces training time by 53.8–68.5% and memory usage by 55.6% compared to the alternating direction method of multipliers (ADMM) across six benchmark datasets. The proposed model variants achieve comparable or superior performance to state-of-the-art collaborative filtering models on all real-world datasets. These findings validate the proposed training framework’s capability to enable feasible deployment of SLIM-like models in efficiency-critical and resource-constrained environments.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104314"},"PeriodicalIF":6.9,"publicationDate":"2025-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144780915","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}

{"title":"A fine-grained evaluation framework for language models: Combining pointwise grading and pairwise comparison","authors":"Yijie Li ,&nbsp;Yuan Sun","doi":"10.1016/j.ipm.2025.104270","DOIUrl":"10.1016/j.ipm.2025.104270","url":null,"abstract":"<div><div>Automated evaluation of Large Language Models (LLMs) responses face fundamental challenges: evaluation bias, protocol inflexibility, and the trade-off between quality and accessibility. Current paradigms either rely heavily on expensive proprietary models or suffer from systematic biases and limited evaluation modes. We introduce MELD, an 8B-parameter evaluation model designed to overcome these limitations via systematic bias mitigation and multi-protocol adaptability. MELD is trained on a comprehensive dataset covering eight categories and 50 subcategories, each with tailored evaluation criteria. It supports both pointwise grading and pairwise comparison through model merging, achieving robust performance across protocols. Experiments show MELD consistently outperforms open-source baselines and matches or surpasses GPT-4 in human alignment. Notably, MELD reduces bias in position, length, and content. The framework includes a lightweight quantized deployment option, enabling high-quality evaluation in resource-constrained settings. This work provides a practical, cost-effective solution for LLM evaluation. Resources are available at: <span><span>https://github.com/Bound2-2/MELD-Eval</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104270"},"PeriodicalIF":6.9,"publicationDate":"2025-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144772953","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}

{"title":"Spectral-constrained global and local feature learning for hyperspectral anomaly detection","authors":"Zhe Zhao,&nbsp;Jiangluqi Song,&nbsp;Huixin Zhou,&nbsp;Yong Zhu,&nbsp;Jiajia Zhang","doi":"10.1016/j.ipm.2025.104313","DOIUrl":"10.1016/j.ipm.2025.104313","url":null,"abstract":"<div><div>Hyperspectral anomaly detection (HAD) plays a crucial role in remote sensing image processing. Many autoencoder (AE)-based algorithms often face limitations due to insufficient spectral properties and inadequate integration of global and local features within the hyperspectral image (HSI). To address these challenges, a spectral-constrained global and local feature learning network (SGLNet) is proposed for HAD. Firstly, SGLNet employs three sub-networks to extract the global features, local features and spectral low-rank features from the encoding features, respectively. Specifically, a memory matrix in the low-rank representation branch can capture the global low-rank characteristics of HSI. For the global feature extraction branch, we employ graph convolution to effectively mine global information, thereby enhancing the capability of SGLNet for background modeling. Then, to make full use of the extracted features, a spectral-guided feature fusion module (SFFM) is designed to integrate the features. The SFFM can dynamically adjust local and global features while reducing spatial and spectral information redundancy, thereby enabling effective feature fusion. Next, the fused features are used to predict the background of HSI. Finally, abnormal scores are obtained by combining the RX detection result on the input HSI and the detection result using Mahalanobis distance on the residual image. Comparative experiments conducted on four real hyperspectral datasets demonstrate the effectiveness and superiority of the proposed method, surpassing previous AE-based methods by an average of 0.16%, 0.38%, 0.01%, and 0.98% in <span><math><msub><mrow><mi>AUC</mi></mrow><mrow><mi>(D,F)</mi></mrow></msub></math></span> values. This indicates that effectively utilizing both local and global information, along with spectral properties, can enhance the accuracy of anomaly detection. The code of this work will be released at: <span><span>https://github.com/xautzhaozhe/SGLNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104313"},"PeriodicalIF":6.9,"publicationDate":"2025-08-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144758002","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}

{"title":"Efficient Table Embeddings via Self-Supervised Structural-Semantic Graph Autoencoder","authors":"Jinlong Tian ,&nbsp;Shixuan Liu ,&nbsp;Ruochun Jin ,&nbsp;Mengmeng Li ,&nbsp;Yanfang Zhou ,&nbsp;Xinhai Xu ,&nbsp;Yuhua Tang","doi":"10.1016/j.ipm.2025.104298","DOIUrl":"10.1016/j.ipm.2025.104298","url":null,"abstract":"<div><div>Representing tabular data effectively is difficult due to its structural complexity and semantic nuances. Existing models either inadequately capture these features or suffer from computational inefficiency. This paper presents TEA (Table Embedding Autoencoder), a self-supervised learning framework for tabular data embedding. TEA utilizes a Contextual Tabular Graph representation incorporating crucial table relationships and a specialized Table Graph Autoencoder (TGAE) with multi-facet reconstruction (feature/edge/degree). This design ensures efficient learning of comprehensive structural and semantic embeddings. On eight benchmark datasets, TEA surpasses SOTA tabular models and LLMs, achieving average F-measure improvements of 14 (entity resolution). Crucially, TEA is 4x more computationally efficient than the SOTA model, facilitating large-scale data processing.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104298"},"PeriodicalIF":6.9,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757999","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}

{"title":"Language model collaboration for relation extraction from classical Chinese historical documents","authors":"Xuemei Tang ,&nbsp;Linxu Wang ,&nbsp;Jun Wang","doi":"10.1016/j.ipm.2025.104286","DOIUrl":"10.1016/j.ipm.2025.104286","url":null,"abstract":"<div><div>Classical Chinese historical documents are invaluable for Chinese cultural heritage and history research, while they remain underexplored within natural language processing (NLP) due to limited annotated resources and linguistic evolution spanning thousands of years. Addressing the challenges presented by this low annotated resource domain, we develop a relation extraction (RE) corpus that preserves the characteristics of classical Chinese documents. Utilizing this corpus, we explore RE in classical Chinese documents through a collaboration framework that integrates small pre-trained language models (SLMs), such as BERT, with large language models (LLMs) like GPT-3.5. SLMs can quickly adapt to specific tasks given sufficient supervised data but often struggle with few-shot scenarios. Conversely, LLMs leverage broad domain knowledge to handle few-shot challenges but face limitations when processing lengthy input sequences. Combining these complementary strengths, we propose a “train-guide-predict” collaboration framework, where a small language model corporate with a large language model (SLCoLM). This framework enables SLMs to capture task-specific knowledge for head relation categories, while LLMs offer insights for few-shot relation categories. Experimental results show that SLCoLM outperforms both fine-tuned SLMs and LLMs using in-context learning (ICL). It also helps mitigate the long-tail problem in classical Chinese historical documents.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104286"},"PeriodicalIF":6.9,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144757998","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}

{"title":"Misinformation detection on online social networks using pretrained language models","authors":"Pir Noman Ahmad ,&nbsp;Adnan Muhammad Shah ,&nbsp;KangYoon Lee ,&nbsp;Wazir Muhammad","doi":"10.1016/j.ipm.2025.104342","DOIUrl":"10.1016/j.ipm.2025.104342","url":null,"abstract":"<div><div>The growing prevalence of online misinformation poses substantial threats, with notable examples including the undermined integrity of democratic processes and decreased effectiveness of public health efforts. The effectiveness of existing solutions, such as user education and content removal, remains unclear, primarily because confirmation bias and peer pressure hinder the identification of noncredible information by users. To address these challenges posed by online misinformation, this study proposes a state-of-the-art approach that leverages transformer-based models, including bidirectional encoder representation from transformers (BERT), GPT-2, and XLNet. These models leverage attention mechanisms to simultaneously process and capture contextual subtleties in documents, enabling highly accurate misinformation detection and classification in dynamic and complex online narratives. A transformer-based pretrained language model is used to analyze, a large corpus of tweets related to misinformation events concerning the 2020 U.S. election. Although isolated interventions are found to be ineffective, a synergistic approach is shown to reduce misinformation prevalence by 87.9 % within a 40-min delay based on a credibility interval of 80 %. These findings highlight the potential of empirical models to inform policies, enhance content moderation practices, and strengthen public resilience against misinformation.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104342"},"PeriodicalIF":6.9,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144758000","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}

{"title":"Multi-Scale Transformers with dual attention and adaptive masking for sequential recommendation","authors":"Haiqin Li ,&nbsp;Yuhan Yang ,&nbsp;Jun Zeng ,&nbsp;Min Gao ,&nbsp;Junhao Wen","doi":"10.1016/j.ipm.2025.104318","DOIUrl":"10.1016/j.ipm.2025.104318","url":null,"abstract":"<div><div>Sequential recommendation focuses on modeling and predicting a user’s next actions based on their sequential behavior patterns, using the temporal order and dynamics of user actions to provide more personalized and contextual suggestions. Sequential recommendation models rely on limited temporal scales, making it challenging to explicitly capture diverse user behaviors spanning multiple scales. Motivated by this challenge, this paper introduces ScaleRec, an advanced Multi-Scale Transformer architecture augmented with dual attention mechanisms and adaptive masking for sequential recommendation. ScaleRec integrates interaction granularity and context through multi-scale division, segmenting user behavior sequences into patches of varying lengths. Dual attention explicitly models fine-grained interests and coarse-grained preferences, including intra-patch cross-attention and inter-patch self-attention. Specifically, intra-patch cross-attention employs a learnable Gaussian kernel to introduce locality-based inductive biases, capturing fine-grained behavioral dynamics. The inter-patch self-attention is further enhanced by a Context-adaptive Preferences Aggregator, which dynamically selects and integrates relevant long-term user preferences. Additionally, we introduce an adaptive masking fusion strategy to filter redundant information dynamically. Extensive experiments on six benchmark datasets show that ScaleRec achieves state-of-the-art performance, improving the recommendation performance by up to 24.95% in terms of HR@5. The code of the proposed model is available at: <span><span>https://github.com/gangtann/ScaleRec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104318"},"PeriodicalIF":6.9,"publicationDate":"2025-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144748641","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}

{"title":"Human-guided collective LLM intelligence for strategic planning via two-stage information retrieval","authors":"Sangyeop Kim ,&nbsp;Junguk Ha ,&nbsp;Hangyeul Lee ,&nbsp;Sohhyung Park ,&nbsp;Sungzoon Cho","doi":"10.1016/j.ipm.2025.104288","DOIUrl":"10.1016/j.ipm.2025.104288","url":null,"abstract":"<div><div>Modern businesses face increasing challenges in strategic planning due to the immense volume of digital information. The rapid growth of available data sources – from market trends and competitor activities to real-time economic indicators – makes comprehensive analysis within tight timeframes arduous. To address these challenges, large language models (LLMs) have emerged as potential tools, efficiently analyzing extensive information across diverse domains. However, LLMs face critical limitations: they cannot access proprietary information or real-time data and cannot engage in collaborative refinement processes that human experts traditionally use to develop and improve strategic analyses. This study introduces the Collective Intelligence of AI Consultants (CIAIC) framework, where specialized AI agents function as individual consultants, collaborating like a consulting team to enhance strategic analysis. The framework combines real-time data integration with collaborative AI mechanisms in a five-stage process: (1) human-guided objective definition, (2) retrieval-augmented draft generation, (3) supplementary data retrieval through multi-agents, (4) draft revision via collective intelligence, and (5) multi-perspective strategic plan compilation. Experimental evaluations using PESTEL and SWOT analyses demonstrate the effectiveness of this collective approach through both quantitative metrics and human preference assessments.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104288"},"PeriodicalIF":6.9,"publicationDate":"2025-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144748640","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}

{"title":"Anomaly detection in UAV-captured crowd images using cumulative frame segmentation and adversarial learning","authors":"Fangfang Ye ,&nbsp;Jinming Wang ,&nbsp;Cao Shuhua ,&nbsp;Zhou Dong ,&nbsp;Ting Wang ,&nbsp;Ezzeddine Touti","doi":"10.1016/j.ipm.2025.104320","DOIUrl":"10.1016/j.ipm.2025.104320","url":null,"abstract":"<div><div>Anomaly detection in crowds using unmanned aerial vehicle (UAV) captured images is preceded by computer-aided analysis and intelligent learning algorithms. The study is pursued using conventional image processing steps and detection methods. This article introduces a novel anomaly object-detecting method, utilizing the Cumulative Frame Segmentation (AODM-CFS) approach to identify abnormalities in UAV-captured images based on variations in pixel intensity, the data from the VisDrone dataset, and UAV Anomaly Detection. The proposed method segments the maximum intensity varying pixels by examining different pixel occurrences. The cumulative frames are segmented using the maximum repeated intensity pixels to identify objects with maximum feature diversity. The pixel repetition is verified using concatenated adversarial learning, generating repeated and dissimilar pixel maps for various identified frames. These frames are updated using the pixels discovered towards the end of the image. The training for the network map is repeated using segmented frames that rely on maximum feature diversions. Therefore, the abnormal object/ human in the image is identified using the maximum dispersion frame. The proposed method increased detection accuracy by 13.46 %, segmentation precision by 14.08 %, sensitivity by 12.7 %, and specificity by 12.88 %, resulting in an 11.92 % reduction in segmentation error compared to other existing models.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 1","pages":"Article 104320"},"PeriodicalIF":6.9,"publicationDate":"2025-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144738612","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}