To employ Pre-trained Language Models (PLMs) as knowledge containers in niche domains it is important to gauge the knowledge of these PLMs about facts in these domains. It is also an important pre-requisite to know how much enrichment effort is required to make them better. As part of this work, we aim to gauge and enrich small PLMs for knowledge of world geography. Firstly, we develop a moderately sized dataset of masked sentences covering 24 different fact types about world geography to estimate knowledge of PLMs on these facts. We hypothesize that for this niche domain, smaller PLMs may not be well equipped. Secondly, we enrich PLMs with this knowledge through fine-tuning and check if the knowledge in the dataset is infused sufficiently. We further hypothesize that linguistic variability in the manual templates used to embed the knowledge in masked sentences does not affect the knowledge infusion. Finally, we demonstrate the application of PLMs to tourism blog search and Wikidata KB augmentation. In both applications, we aim at showing the effectiveness of using PLMs to achieve competitive performance.
{"title":"Gauging, enriching and applying geography knowledge in Pre-trained Language Models","authors":"Nitin Ramrakhiyani , Vasudeva Varma , Girish Keshav Palshikar , Sachin Pawar","doi":"10.1016/j.ipm.2024.103892","DOIUrl":"10.1016/j.ipm.2024.103892","url":null,"abstract":"<div><div>To employ Pre-trained Language Models (PLMs) as knowledge containers in niche domains it is important to gauge the knowledge of these PLMs about facts in these domains. It is also an important pre-requisite to know how much enrichment effort is required to make them better. As part of this work, we aim to gauge and enrich small PLMs for knowledge of world geography. Firstly, we develop a moderately sized dataset of masked sentences covering 24 different fact types about world geography to estimate knowledge of PLMs on these facts. We hypothesize that for this niche domain, smaller PLMs may not be well equipped. Secondly, we enrich PLMs with this knowledge through fine-tuning and check if the knowledge in the dataset is infused sufficiently. We further hypothesize that linguistic variability in the manual templates used to embed the knowledge in masked sentences does not affect the knowledge infusion. Finally, we demonstrate the application of PLMs to tourism blog search and Wikidata KB augmentation. In both applications, we aim at showing the effectiveness of using PLMs to achieve competitive performance.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103892"},"PeriodicalIF":7.4,"publicationDate":"2024-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142326351","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}
Event prediction aims to forecast future events by analyzing the inherent development patterns of historical events. A desirable event prediction system should learn new event knowledge, and adapt to new domains or tasks that arise in real-world application scenarios. However, continuous training can lead to catastrophic forgetting of the model. While existing continuous learning methods can retain characteristic knowledge from previous domains, they ignore potential shared knowledge in subsequent tasks. To tackle these challenges, we propose a novel event prediction method based on graph structural commonality and domain characteristic prompts, which not only avoids forgetting but also facilitates bi-directional knowledge transfer across domains. Specifically, we mitigate model forgetting by designing domain characteristic-oriented prompts in a continuous task stream with frozen the backbone pre-trained model. Building upon this, we further devise a commonality-based adaptive updating algorithm by harnessing a unique structural commonality prompt to inspire implicit common features across domains. Our experimental results on two public benchmark datasets for event prediction demonstrate the effectiveness of our proposed continuous learning event prediction method compared to state-of-the-art baselines. In tests conducted on the IED-Stream, DST’s ET-TA metric significantly improved by 5.6% over the current best baseline model, while the ET-MD metric, which reveals forgetting, decreased by 5.8%.
{"title":"DST: Continual event prediction by decomposing and synergizing the task commonality and specificity","authors":"Yuxin Zhang , Songlin Zhai , Yongrui Chen , Shenyu Zhang , Sheng Bi , Yuan Meng , Guilin Qi","doi":"10.1016/j.ipm.2024.103899","DOIUrl":"10.1016/j.ipm.2024.103899","url":null,"abstract":"<div><div>Event prediction aims to forecast future events by analyzing the inherent development patterns of historical events. A desirable event prediction system should learn new event knowledge, and adapt to new domains or tasks that arise in real-world application scenarios. However, continuous training can lead to catastrophic forgetting of the model. While existing continuous learning methods can retain characteristic knowledge from previous domains, they ignore potential shared knowledge in subsequent tasks. To tackle these challenges, we propose a novel event prediction method based on graph structural commonality and domain characteristic prompts, which not only avoids forgetting but also facilitates bi-directional knowledge transfer across domains. Specifically, we mitigate model forgetting by designing domain characteristic-oriented prompts in a continuous task stream with frozen the backbone pre-trained model. Building upon this, we further devise a commonality-based adaptive updating algorithm by harnessing a unique structural commonality prompt to inspire implicit common features across domains. Our experimental results on two public benchmark datasets for event prediction demonstrate the effectiveness of our proposed continuous learning event prediction method compared to state-of-the-art baselines. In tests conducted on the IED-Stream, DST’s ET-TA metric significantly improved by 5.6% over the current best baseline model, while the ET-MD metric, which reveals forgetting, decreased by 5.8%.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103899"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323907","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}
Noisy annotations have become a key issue limiting Document-level Relation Extraction (DocRE). Previous research explored the problem through manual re-annotation. However, the handcrafted strategy is of low efficiency, incurs high human costs and cannot be generalized to large-scale datasets. To address the problem, we construct a confidence-based Revision framework for DocRE (ReD), aiming to achieve high-quality automatic data revision. Specifically, we first introduce a denoising training module to recognize relational facts and prevent noisy annotations. Second, a confidence-based data revision module is equipped to perform adaptive data revision for long-tail distributed relational facts. After the data revision, we design an iterative training module to create a virtuous cycle, which transforms the revised data into useful training data to support further revision. By capitalizing on ReD, we propose ReD-DocRED, which consists of 101,873 revised annotated documents from DocRED. ReD-DocRED has introduced 57.1% new relational facts, and concurrently, models trained on ReD-DocRED have achieved significant improvements in F1 scores, ranging from 6.35 to 16.55. The experimental results demonstrate that ReD can achieve high-quality data revision and, to some extent, replace manual labeling.1
{"title":"An adaptive confidence-based data revision framework for Document-level Relation Extraction","authors":"Chao Jiang , Jinzhi Liao , Xiang Zhao , Daojian Zeng , Jianhua Dai","doi":"10.1016/j.ipm.2024.103909","DOIUrl":"10.1016/j.ipm.2024.103909","url":null,"abstract":"<div><div>Noisy annotations have become a key issue limiting <strong>Doc</strong>ument-level <strong>R</strong>elation <strong>E</strong>xtraction <strong>(DocRE)</strong>. Previous research explored the problem through manual re-annotation. However, the handcrafted strategy is of low efficiency, incurs high human costs and cannot be generalized to large-scale datasets. To address the problem, we construct a confidence-based <strong>Re</strong>vision framework for <strong>D</strong>ocRE (<strong>ReD</strong>), aiming to achieve high-quality automatic data revision. Specifically, we first introduce a denoising training module to recognize relational facts and prevent noisy annotations. Second, a confidence-based data revision module is equipped to perform adaptive data revision for long-tail distributed relational facts. After the data revision, we design an iterative training module to create a virtuous cycle, which transforms the revised data into useful training data to support further revision. By capitalizing on ReD, we propose <strong>ReD-DocRED</strong>, which consists of 101,873 revised annotated documents from DocRED. ReD-DocRED has introduced 57.1% new relational facts, and concurrently, models trained on ReD-DocRED have achieved significant improvements in F1 scores, ranging from 6.35 to 16.55. The experimental results demonstrate that ReD can achieve high-quality data revision and, to some extent, replace manual labeling.<span><span><sup>1</sup></span></span></div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103909"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323905","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}
Pub Date : 2024-09-26DOI: 10.1016/j.ipm.2024.103907
Ante Wang , Linfeng Song , Zijun Min , Ge Xu , Xiaoli Wang , Junfeng Yao , Jinsong Su
Conversational query generation aims at producing search queries from dialogue histories, which are then used to retrieve relevant knowledge from a search engine to help knowledge-based dialogue systems. Trained to maximize the likelihood of gold queries, previous models suffer from the data hunger issue, and they tend to both drop important concepts from dialogue histories and generate irrelevant concepts at inference time. We attribute these issues to the over-association phenomenon where a large number of gold queries are indirectly related to the dialogue topics, because annotators may unconsciously perform reasoning with their background knowledge when generating these gold queries. We carefully analyze the negative effects of this phenomenon on pretrained Seq2seq query producers and then propose effective instance-level weighting strategies for training to mitigate these issues from multiple perspectives. Experiments on two benchmarks, Wizard-of-Internet and DuSinc, show that our strategies effectively alleviate the negative effects and lead to significant performance gains (2% 5% across automatic metrics and human evaluation). Further analysis shows that our model selects better concepts from dialogue histories and is 10 times more data efficient than the baseline.
{"title":"Mitigating the negative impact of over-association for conversational query production","authors":"Ante Wang , Linfeng Song , Zijun Min , Ge Xu , Xiaoli Wang , Junfeng Yao , Jinsong Su","doi":"10.1016/j.ipm.2024.103907","DOIUrl":"10.1016/j.ipm.2024.103907","url":null,"abstract":"<div><div>Conversational query generation aims at producing search queries from dialogue histories, which are then used to retrieve relevant knowledge from a search engine to help knowledge-based dialogue systems. Trained to maximize the likelihood of gold queries, previous models suffer from the data hunger issue, and they tend to both drop important concepts from dialogue histories and generate irrelevant concepts at inference time. We attribute these issues to the <em>over-association</em> phenomenon where a large number of gold queries are indirectly related to the dialogue topics, because annotators may unconsciously perform reasoning with their background knowledge when generating these gold queries. We carefully analyze the negative effects of this phenomenon on pretrained Seq2seq query producers and then propose effective instance-level weighting strategies for training to mitigate these issues from multiple perspectives. Experiments on two benchmarks, Wizard-of-Internet and DuSinc, show that our strategies effectively alleviate the negative effects and lead to significant performance gains (2%<!--> <span><math><mo>∼</mo></math></span> <!--> <!-->5% across automatic metrics and human evaluation). Further analysis shows that our model selects better concepts from dialogue histories and is <em>10 times</em> more data efficient than the baseline.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103907"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323903","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}
Pub Date : 2024-09-26DOI: 10.1016/j.ipm.2024.103900
Miquel Miró-Nicolau, Antoni Jaume-i-Capó, Gabriel Moyà-Alcover
The use of eXplainable Artificial Intelligence (XAI) systems has introduced a set of challenges that need resolution. Herein, we focus on how to correctly select an XAI method, an open questions within the field. The inherent difficulty of this task is due to the lack of a ground truth. Several authors have proposed metrics to approximate the fidelity of different XAI methods. These metrics lack verification and have concerning disagreements. In this study, we proposed a novel methodology to verify fidelity metrics, using transparent models. These models allowed us to obtain explanations with perfect fidelity. Our proposal constitutes the first objective benchmark for these metrics, facilitating a comparison of existing proposals, and surpassing existing methods. We applied our benchmark to assess the existing fidelity metrics in two different experiments, each using public datasets comprising 52,000 images. The images from these datasets had a size a 128 by 128 pixels and were synthetic data that simplified the training process. We identified that two fidelity metrics, Faithfulness Estimate and Faithfulness Correlation, obtained the expected perfect results for linear models, showing their ability to approximate fidelity for this kind of methods. However, when present with non-linear models, as the ones most used in the state-of-the-art,all metric values, indicated a lack of fidelity, with the best one showing a 30% deviation from the expected values for perfect explanation. Our experimentation led us to conclude that the current fidelity metrics are not reliable enough to be used in real scenarios. From this finding, we deemed it necessary to development new metrics, to avoid the detected problems, and we recommend the usage of our proposal as a benchmark within the scientific community to address these limitations.
{"title":"A comprehensive study on fidelity metrics for XAI","authors":"Miquel Miró-Nicolau, Antoni Jaume-i-Capó, Gabriel Moyà-Alcover","doi":"10.1016/j.ipm.2024.103900","DOIUrl":"10.1016/j.ipm.2024.103900","url":null,"abstract":"<div><div>The use of eXplainable Artificial Intelligence (XAI) systems has introduced a set of challenges that need resolution. Herein, we focus on how to correctly select an XAI method, an open questions within the field. The inherent difficulty of this task is due to the lack of a ground truth. Several authors have proposed metrics to approximate the fidelity of different XAI methods. These metrics lack verification and have concerning disagreements. In this study, we proposed a novel methodology to verify fidelity metrics, using transparent models. These models allowed us to obtain explanations with perfect fidelity. Our proposal constitutes the first objective benchmark for these metrics, facilitating a comparison of existing proposals, and surpassing existing methods. We applied our benchmark to assess the existing fidelity metrics in two different experiments, each using public datasets comprising 52,000 images. The images from these datasets had a size a 128 by 128 pixels and were synthetic data that simplified the training process. We identified that two fidelity metrics, Faithfulness Estimate and Faithfulness Correlation, obtained the expected perfect results for linear models, showing their ability to approximate fidelity for this kind of methods. However, when present with non-linear models, as the ones most used in the state-of-the-art,all metric values, indicated a lack of fidelity, with the best one showing a 30% deviation from the expected values for perfect explanation. Our experimentation led us to conclude that the current fidelity metrics are not reliable enough to be used in real scenarios. From this finding, we deemed it necessary to development new metrics, to avoid the detected problems, and we recommend the usage of our proposal as a benchmark within the scientific community to address these limitations.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103900"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323906","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-09-26DOI: 10.1016/j.ipm.2024.103893
Pavel Milei , Nadezhda Votintseva , Angel Barajas
As business data grows in volume and complexity, there is an increasing demand for efficient, accurate, and scalable methods to analyse and classify business models. This study introduces and validates a novel approach for the automated identification of business models through content analysis of company reports. Our method builds on the semantic operationalisation of the business model that establishes a detailed structure of business model elements along with the dictionary of associated keywords. Through several refinement steps, we calibrate theory-derived keywords and obtain a final dictionary that totals 318 single words and collocations. We then run dictionary-based content analysis on a dataset of 363 annual reports from young public companies. The results are presented via a web-based software prototype, available online, that enables researchers and practitioners to visualise the structure and magnitude of business model elements based on the annual reports. Furthermore, we conduct a cluster analysis of the obtained data and combine the results with the extant theory to derive 5 categories of business models in young companies.
{"title":"Automated Identification of Business Models","authors":"Pavel Milei , Nadezhda Votintseva , Angel Barajas","doi":"10.1016/j.ipm.2024.103893","DOIUrl":"10.1016/j.ipm.2024.103893","url":null,"abstract":"<div><div>As business data grows in volume and complexity, there is an increasing demand for efficient, accurate, and scalable methods to analyse and classify business models. This study introduces and validates a novel approach for the automated identification of business models through content analysis of company reports. Our method builds on the semantic operationalisation of the business model that establishes a detailed structure of business model elements along with the dictionary of associated keywords. Through several refinement steps, we calibrate theory-derived keywords and obtain a final dictionary that totals 318 single words and collocations. We then run dictionary-based content analysis on a dataset of 363 annual reports from young public companies. The results are presented via a web-based software prototype, available online, that enables researchers and practitioners to visualise the structure and magnitude of business model elements based on the annual reports. Furthermore, we conduct a cluster analysis of the obtained data and combine the results with the extant theory to derive 5 categories of business models in young companies.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103893"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A sentence is composed of linguistically linked units, such as words or phrases. The dependencies between them compose the linguistic structures of a sentence, which indicates the meanings of linguistic units and encodes the syntactic or semantic relationships between them. Therefore, it is important to learn the linguistic structures of a sentence for entity relation extraction or other natural language processing (NLP) tasks. In related works, manual rules or dependency trees are usually adopted to capture the linguistic structures. These methods heavily depend on prior knowledge or external toolkits. In this paper, we introduce a Supervised Graph Autoencoder Network (SGAN) model to automatically learn the linguistic structures of a sentence. Unlike traditional graph neural networks that use a fixed adjacency matrix initialized with prior knowledge, the SGAN model contains a learnable adjacency matrix that is dynamically tuned by a task-relevant learning objective. It can automatically learn linguistic structures from raw input sentences. After being evaluated on seven public datasets, the SGAN achieves state-of-the-art (SOTA) performance, outperforming all compared models. The results show that automatically learned linguistic structures have better performance than manually designed linguistic patterns. It exhibits great potential for supporting entity relation extraction and other NLP tasks.
{"title":"Automatically learning linguistic structures for entity relation extraction","authors":"Weizhe Yang , Yanping Chen , Jinling Xu , Yongbin Qin , Ping Chen","doi":"10.1016/j.ipm.2024.103904","DOIUrl":"10.1016/j.ipm.2024.103904","url":null,"abstract":"<div><div>A sentence is composed of linguistically linked units, such as words or phrases. The dependencies between them compose the linguistic structures of a sentence, which indicates the meanings of linguistic units and encodes the syntactic or semantic relationships between them. Therefore, it is important to learn the linguistic structures of a sentence for entity relation extraction or other natural language processing (NLP) tasks. In related works, manual rules or dependency trees are usually adopted to capture the linguistic structures. These methods heavily depend on prior knowledge or external toolkits. In this paper, we introduce a Supervised Graph Autoencoder Network (SGAN) model to automatically learn the linguistic structures of a sentence. Unlike traditional graph neural networks that use a fixed adjacency matrix initialized with prior knowledge, the SGAN model contains a learnable adjacency matrix that is dynamically tuned by a task-relevant learning objective. It can automatically learn linguistic structures from raw input sentences. After being evaluated on seven public datasets, the SGAN achieves state-of-the-art (SOTA) performance, outperforming all compared models. The results show that automatically learned linguistic structures have better performance than manually designed linguistic patterns. It exhibits great potential for supporting entity relation extraction and other NLP tasks.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103904"},"PeriodicalIF":7.4,"publicationDate":"2024-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142323908","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}
Graph-based anomaly detection aims to identify anomalous vertices in graph-structured data. It relies on the ability of graph neural networks (GNNs) to capture both relational and attribute information within graphs. However, previous GNN-based methods exhibit two critical shortcomings. Firstly, GNN is inherently a low-pass filter that tends to lead similar representations of neighboring vertices, which may result in the loss of critical anomalous information, termed as low-frequency constraints. Secondly, anomalous vertices that deliberately mimic normal vertices in features and structures are hard to detect, especially when the distribution of labels is unbalanced. To address these defects, we propose a Universal Adaptive Algorithm for Graph Anomaly Detection (U-AGAD), which employs enhanced high frequency filters to overcome the low-frequency constraints, as well as aggregating both -nearest neighbor (NN) and -farthest neighbor (FN) to resolve the vertices’ camouflage problem. Extensive experiments demonstrated the effectiveness and universality of our proposed U-AGAD and its constituent components, achieving improvements of up to 6% and an average increase of 2% on AUC-PR over the state-of-the-art methods. The source codes, and parameter setting details can be found at https://github.com/LIyvqi/U-A2GAD.
{"title":"A Universal Adaptive Algorithm for Graph Anomaly Detection","authors":"Yuqi Li, Guosheng Zang, Chunyao Song, Xiaojie Yuan","doi":"10.1016/j.ipm.2024.103905","DOIUrl":"10.1016/j.ipm.2024.103905","url":null,"abstract":"<div><div>Graph-based anomaly detection aims to identify anomalous vertices in graph-structured data. It relies on the ability of graph neural networks (GNNs) to capture both relational and attribute information within graphs. However, previous GNN-based methods exhibit two critical shortcomings. Firstly, GNN is inherently a low-pass filter that tends to lead similar representations of neighboring vertices, which may result in the loss of critical anomalous information, termed as low-frequency constraints. Secondly, anomalous vertices that deliberately mimic normal vertices in features and structures are hard to detect, especially when the distribution of labels is unbalanced. To address these defects, we propose a <strong>U</strong>niversal <strong>A</strong>daptive <strong>A</strong>lgorithm for <strong>G</strong>raph <strong>A</strong>nomaly <strong>D</strong>etection (<strong>U-A</strong><span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span><strong>GAD</strong>), which employs enhanced high frequency filters to overcome the low-frequency constraints, as well as aggregating both <span><math><mi>k</mi></math></span>-nearest neighbor (<span><math><mi>k</mi></math></span>NN) and <span><math><mi>k</mi></math></span>-farthest neighbor (<span><math><mi>k</mi></math></span>FN) to resolve the vertices’ camouflage problem. Extensive experiments demonstrated the effectiveness and universality of our proposed <strong>U-A</strong><span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span><strong>GAD</strong> and its constituent components, achieving improvements of up to 6% and an average increase of 2% on AUC-PR over the state-of-the-art methods. The source codes, and parameter setting details can be found at <span><span>https://github.com/LIyvqi/U-A2GAD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103905"},"PeriodicalIF":7.4,"publicationDate":"2024-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142319166","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}
Pub Date : 2024-09-25DOI: 10.1016/j.ipm.2024.103896
Haoyan Fu, Zhida Qin, Wenhao Xue, Gangyi Ding
Session-based recommendation (SBR) predicts the next item in user sequences. Existing research focuses on item transition patterns, neglecting semantic information dependencies crucial for understanding users’ preferences. Incorporating semantic characteristics is vital for accurate recommendations, especially in applications like user purchase sequences. In this paper, to tackle the above issue, we novelly propose a framework that hierarchically fuses temporal and semantic dependencies. Technically, we present the Item Transition Dependency Module and Semantic Dependency Module based on the whole session set: (i) Item Transition Dependency Module is exclusively to learn the item embeddings through temporal relations and utilizes item transitions from both global and local levels; (ii) Semantic Dependency Module develops mutually independent embeddings of both sessions and items via stable interaction relations. In addition, under the unified organization of the Cross View, semantic information is adaptively incorporated into the temporal dependency learning and used to improve the performance of SBR. Extensive experiments on three large-scale real-world datasets show the superiority of our framework over current state-of-the-art methods. In particular, our model improves its performance over SOTA on all three datasets, with 5.5%, 0.2%, and 3.0% improvements on Recall@20, and 5.8%, 4.6%, and 2.0% improvements on MRR@20, respectively.
{"title":"Fusing temporal and semantic dependencies for session-based recommendation","authors":"Haoyan Fu, Zhida Qin, Wenhao Xue, Gangyi Ding","doi":"10.1016/j.ipm.2024.103896","DOIUrl":"10.1016/j.ipm.2024.103896","url":null,"abstract":"<div><div>Session-based recommendation (SBR) predicts the next item in user sequences. Existing research focuses on item transition patterns, neglecting semantic information dependencies crucial for understanding users’ preferences. Incorporating semantic characteristics is vital for accurate recommendations, especially in applications like user purchase sequences. In this paper, to tackle the above issue, we novelly propose a framework that hierarchically fuses temporal and semantic dependencies. Technically, we present the Item Transition Dependency Module and Semantic Dependency Module based on the whole session set: (i) Item Transition Dependency Module is exclusively to learn the item embeddings through temporal relations and utilizes item transitions from both global and local levels; (ii) Semantic Dependency Module develops mutually independent embeddings of both sessions and items via stable interaction relations. In addition, under the unified organization of the Cross View, semantic information is adaptively incorporated into the temporal dependency learning and used to improve the performance of SBR. Extensive experiments on three large-scale real-world datasets show the superiority of our framework over current state-of-the-art methods. In particular, our model improves its performance over SOTA on all three datasets, with 5.5%, 0.2%, and 3.0% improvements on Recall@20, and 5.8%, 4.6%, and 2.0% improvements on MRR@20, respectively.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103896"},"PeriodicalIF":7.4,"publicationDate":"2024-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142319165","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}
Pub Date : 2024-09-24DOI: 10.1016/j.ipm.2024.103895
Mohammad Zia Ur Rehman , Sufyaan Zahoor , Areeb Manzoor , Musharaf Maqbool , Nagendra Kumar
A substantial portion of offensive content on social media is directed towards women. Since the approaches for general offensive content detection face a challenge in detecting misogynistic content, it requires solutions tailored to address offensive content against women. To this end, we propose a novel multimodal framework for the detection of misogynistic and sexist content. The framework comprises three modules: the Multimodal Attention module (MANM), the Graph-based Feature Reconstruction Module (GFRM), and the Content-specific Features Learning Module (CFLM). The MANM employs adaptive gating-based multimodal context-aware attention, enabling the model to focus on relevant visual and textual information and generating contextually relevant features. The GFRM module utilizes graphs to refine features within individual modalities, while the CFLM focuses on learning text and image-specific features such as toxicity features and caption features. Additionally, we curate a set of misogynous lexicons to compute the misogyny-specific lexicon score from the text. We apply test-time augmentation in feature space to better generalize the predictions on diverse inputs. The performance of the proposed approach has been evaluated on two multimodal datasets, MAMI, and MMHS150K, with 11,000 and 13,494 samples, respectively. The proposed method demonstrates an average improvement of 11.87% and 10.82% in macro-F1 over existing multimodal methods on the MAMI and MMHS150K datasets, respectively.
{"title":"A context-aware attention and graph neural network-based multimodal framework for misogyny detection","authors":"Mohammad Zia Ur Rehman , Sufyaan Zahoor , Areeb Manzoor , Musharaf Maqbool , Nagendra Kumar","doi":"10.1016/j.ipm.2024.103895","DOIUrl":"10.1016/j.ipm.2024.103895","url":null,"abstract":"<div><div>A substantial portion of offensive content on social media is directed towards women. Since the approaches for general offensive content detection face a challenge in detecting misogynistic content, it requires solutions tailored to address offensive content against women. To this end, we propose a novel multimodal framework for the detection of misogynistic and sexist content. The framework comprises three modules: the Multimodal Attention module (MANM), the Graph-based Feature Reconstruction Module (GFRM), and the Content-specific Features Learning Module (CFLM). The MANM employs adaptive gating-based multimodal context-aware attention, enabling the model to focus on relevant visual and textual information and generating contextually relevant features. The GFRM module utilizes graphs to refine features within individual modalities, while the CFLM focuses on learning text and image-specific features such as toxicity features and caption features. Additionally, we curate a set of misogynous lexicons to compute the misogyny-specific lexicon score from the text. We apply test-time augmentation in feature space to better generalize the predictions on diverse inputs. The performance of the proposed approach has been evaluated on two multimodal datasets, MAMI, and MMHS150K, with 11,000 and 13,494 samples, respectively. The proposed method demonstrates an average improvement of 11.87% and 10.82% in macro-F1 over existing multimodal methods on the MAMI and MMHS150K datasets, respectively.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 1","pages":"Article 103895"},"PeriodicalIF":7.4,"publicationDate":"2024-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S0306457324002541/pdfft?md5=d17cb5e20a69f9c766570983bc722abc&pid=1-s2.0-S0306457324002541-main.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142314954","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号