Information Processing & Management最新文献_第7页

Knowledge enhanced graph contrastive learning for match outcome prediction

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

Information Processing & Management

Pub Date : 2024-12-12 DOI: 10.1016/j.ipm.2024.104010

Junji Jiang , Likang Wu , Zhipeng Hu , Runze Wu , Xudong Shen , Hongke Zhao

With the booming popularity, Multiplayer Online Battle Arena (MOBA) Game has become one of the mainstream games, in which players are divided into two teams, and the goal is to destroy the base of the opponent. Accurate prediction of the match outcome enables the operators to improve the players’ game experience through balance matching. However, existing methods usually model the players’ combat effectiveness based on the profile data, neglecting the evolution of proficiency of different heroes and their performance against different opponents. To model the evolution of players’ abilities, we propose the Knowledge Enhanced Graph Contrastive Learning (KEGC) framework that reinforces the predictor with the information of the knowledge graph and match sequence graphs. Specifically, we construct a knowledge graph that reflects the static system information on the cooperation and confrontation of heroes by game developers. Meanwhile, the match sequence of each player is converted into a sequence graph for representing dynamic ability, which builds edges in similar matches to capture player skills evolution. Further, we propose a coupled contrastive training framework to adaptively fuse the information from the static and dynamic views. Considering the uncertainty of players’ performance in different games, the conditional variational mechanism is introduced to KEGC. Besides, we also adopt the auxiliary task, i.e., the match balance, and design the joint loss function to suppress the noise in the training data. Extensive experiments on two real-world datasets from well-known MOBA games demonstrate the superiority of KEGC compared to the state-of-the-art methods.

{"title":"Knowledge enhanced graph contrastive learning for match outcome prediction","authors":"Junji Jiang , Likang Wu , Zhipeng Hu , Runze Wu , Xudong Shen , Hongke Zhao","doi":"10.1016/j.ipm.2024.104010","DOIUrl":"10.1016/j.ipm.2024.104010","url":null,"abstract":"<div><div>With the booming popularity, Multiplayer Online Battle Arena (MOBA) Game has become one of the mainstream games, in which players are divided into two teams, and the goal is to destroy the base of the opponent. Accurate prediction of the match outcome enables the operators to improve the players’ game experience through balance matching. However, existing methods usually model the players’ combat effectiveness based on the profile data, neglecting the evolution of proficiency of different heroes and their performance against different opponents. To model the evolution of players’ abilities, we propose the Knowledge Enhanced Graph Contrastive Learning (KEGC) framework that reinforces the predictor with the information of the knowledge graph and match sequence graphs. Specifically, we construct a knowledge graph that reflects the static system information on the cooperation and confrontation of heroes by game developers. Meanwhile, the match sequence of each player is converted into a sequence graph for representing dynamic ability, which builds edges in similar matches to capture player skills evolution. Further, we propose a coupled contrastive training framework to adaptively fuse the information from the static and dynamic views. Considering the uncertainty of players’ performance in different games, the conditional variational mechanism is introduced to KEGC. Besides, we also adopt the auxiliary task, i.e., the match balance, and design the joint loss function to suppress the noise in the training data. Extensive experiments on two real-world datasets from well-known MOBA games demonstrate the superiority of KEGC compared to the state-of-the-art methods.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104010"},"PeriodicalIF":7.4,"publicationDate":"2024-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138906","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

Dynamic analysis of pattern and optimal control research of rumor propagation model on different networks

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

Information Processing & Management

Pub Date : 2024-12-10 DOI: 10.1016/j.ipm.2024.104016

Haoyan Sha, Linhe Zhu

With the continuous development of network media, various rumors will inevitably appear in the process of information exchange and dissemination due to inadequate communication, and rumors are also easy to be more widely spread along the network public platform. At this time, it is particularly necessary to study and control information transmission. Based on the secondary transmission mechanism and the discrete characteristics of the network in real life, we use Laplacian matrix to build a reaction–diffusion model of information transmission. Then, in order to explore the distribution state of various types of people after perturbation is applied at the equilibrium point of the system, we use the correlated theory of Turing pattern to analyze the Turing instability of homogeneous network and heterogeneous network respectively to determine the necessary conditions for the emergence of Turing pattern. According to the specific situation of the appearance of different morphological patterns, the amplitude equation of the model is obtained by multi-scale analysis method. In addition, the theory is verified reasonably and effectively by numerical simulation. In view of different network structures, we first study the general pattern, and make a reasonable analysis of the propagation on the shortest path and the influence of government control. Second, in the simulation process of parameter identification based on optimal control theory, we can find that the algorithm type and network structure will affect the final convergence effect of parameters. Meanwhile, it can be found that connected with the selection of parameters, in order to achieve information control, we need both the intervention of government departments and the improvement of our own discrimination ability in real life. We can effectively control the crowd departure rate near the target value to achieve the ideal distribution state, which is of great significance for information control. Based on the above, we also have conducted a fitting analysis of the model combined with the relevant data of the number of epidemic-related tweets, which can explain the rationality of the model construction.

{"title":"Dynamic analysis of pattern and optimal control research of rumor propagation model on different networks","authors":"Haoyan Sha, Linhe Zhu","doi":"10.1016/j.ipm.2024.104016","DOIUrl":"10.1016/j.ipm.2024.104016","url":null,"abstract":"<div><div>With the continuous development of network media, various rumors will inevitably appear in the process of information exchange and dissemination due to inadequate communication, and rumors are also easy to be more widely spread along the network public platform. At this time, it is particularly necessary to study and control information transmission. Based on the secondary transmission mechanism and the discrete characteristics of the network in real life, we use Laplacian matrix to build a reaction–diffusion model of information transmission. Then, in order to explore the distribution state of various types of people after perturbation is applied at the equilibrium point of the system, we use the correlated theory of Turing pattern to analyze the Turing instability of homogeneous network and heterogeneous network respectively to determine the necessary conditions for the emergence of Turing pattern. According to the specific situation of the appearance of different morphological patterns, the amplitude equation of the model is obtained by multi-scale analysis method. In addition, the theory is verified reasonably and effectively by numerical simulation. In view of different network structures, we first study the general pattern, and make a reasonable analysis of the propagation on the shortest path and the influence of government control. Second, in the simulation process of parameter identification based on optimal control theory, we can find that the algorithm type and network structure will affect the final convergence effect of parameters. Meanwhile, it can be found that connected with the selection of parameters, in order to achieve information control, we need both the intervention of government departments and the improvement of our own discrimination ability in real life. We can effectively control the crowd departure rate near the target value to achieve the ideal distribution state, which is of great significance for information control. Based on the above, we also have conducted a fitting analysis of the model combined with the relevant data of the number of epidemic-related tweets, which can explain the rationality of the model construction.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104016"},"PeriodicalIF":7.4,"publicationDate":"2024-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138808","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

Combining association-rule-guided sequence augmentation with listwise contrastive learning for session-based recommendation

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

Information Processing & Management

Pub Date : 2024-12-10 DOI: 10.1016/j.ipm.2024.103999

Xiangkui Lu , Jun Wu

Sequence augmentation based contrastive learning (SACL) plays a critical role in user behavior modeling towards sequential recommendation tasks. However, SACL cannot work well in the scenario of session-based recommendation (SBR), where the anonymous user behavior sequences (known as sessions) are very short (e.g., with no more than 5 interactions), making most augmentation techniques ineffective. In this paper, we propose a novel method named LCAA (Listwise Contrastive learning with Association-rule-based sequence Augmentation), which lengthens the current session with association rules to create an augmented session, and then leverages a corresponding listwise contrastive loss to maximize the agreement of two recommendation lists generated from the original session and its augmentation. Remarkably, LCAA is a model-agnostic method that can be easily plugged into a wide range of existing SBR models towards better accuracy. To evaluate the effectiveness of LCAA, we implement it with five SBR models utilizing various deep learning techniques (NARM, STAMP, SRGNN, CORE, and ATTMIX) and then compare the performance of each SBR baseline with its LCAA-modified version. Extensive experiments on three datasets (Diginetica, Nowplaying, and Tmall) demonstrate that LCAA yields the average improvement of around 5% on the complete testing sets and around 3% on the short session testing sets in terms of HR and MRR metrics. The code is publicly available.¹

{"title":"Combining association-rule-guided sequence augmentation with listwise contrastive learning for session-based recommendation","authors":"Xiangkui Lu , Jun Wu","doi":"10.1016/j.ipm.2024.103999","DOIUrl":"10.1016/j.ipm.2024.103999","url":null,"abstract":"<div><div>Sequence augmentation based contrastive learning (SACL) plays a critical role in user behavior modeling towards sequential recommendation tasks. However, SACL cannot work well in the scenario of session-based recommendation (SBR), where the anonymous user behavior sequences (known as sessions) are very short (e.g., with no more than 5 interactions), making most augmentation techniques ineffective. In this paper, we propose a novel method named LCAA (<strong>L</strong>istwise <strong>C</strong>ontrastive learning with <strong>A</strong>ssociation-rule-based sequence <strong>A</strong>ugmentation), which lengthens the current session with association rules to create an augmented session, and then leverages a corresponding listwise contrastive loss to maximize the agreement of two recommendation lists generated from the original session and its augmentation. Remarkably, LCAA is a model-agnostic method that can be easily plugged into a wide range of existing SBR models towards better accuracy. To evaluate the effectiveness of LCAA, we implement it with five SBR models utilizing various deep learning techniques (NARM, STAMP, SRGNN, CORE, and ATTMIX) and then compare the performance of each SBR baseline with its LCAA-modified version. Extensive experiments on three datasets (Diginetica, Nowplaying, and Tmall) demonstrate that LCAA yields the average improvement of around 5% on the complete testing sets and around 3% on the short session testing sets in terms of HR and MRR metrics. The code is publicly available.<span><span><sup>1</sup></span></span></div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 103999"},"PeriodicalIF":7.4,"publicationDate":"2024-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138905","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

Unsupervised microservice system anomaly detection via contrastive multi-modal representation clustering

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

Information Processing & Management

Pub Date : 2024-12-10 DOI: 10.1016/j.ipm.2024.104013

Peipeng Wang, Xiuguo Zhang, Yutian Chen, Zhiying Cao

Anomaly detection in microservice systems is crucial for ensuring system stability and reliability. Existing methods rely solely on a single type of monitoring data (e.g., metrics or logs) with either partial or full labels, which miss a large number of anomalies and is costly to manually tag. Therefore, we propose an unsupervised Microservice system Anomaly Detection method via Contrastive Multi-modal representation Clustering (MAD-CMC) to tackle these issues. MAD-CMC first adopts a hierarchical architecture to simultaneously explore the spatial–temporal correlation of metrics and log context information. Next, to facilitate metrics-logs interaction, MAD-CMC introduces a cross-modal Transformer, which outputs multi-modal representation for clustering. During the clustering process, we design a multi-grained contrastive learning approach. Benefiting from the clustering results, MAD-CMC bring intra-cluster representation closer while pushing inter-cluster representation farther away at both inter- and intra-modality aspect. Considering that normal samples are simpler and far more numerous than abnormal sample, we propose a dynamic weighting formula, and apply it to contrastive loss to improve the model’s discrimination ability. Sufficient experiments on public dataset show that MAD-CMC outperforms state-of-the-art methods.

{"title":"Unsupervised microservice system anomaly detection via contrastive multi-modal representation clustering","authors":"Peipeng Wang, Xiuguo Zhang, Yutian Chen, Zhiying Cao","doi":"10.1016/j.ipm.2024.104013","DOIUrl":"10.1016/j.ipm.2024.104013","url":null,"abstract":"<div><div>Anomaly detection in microservice systems is crucial for ensuring system stability and reliability. Existing methods rely solely on a single type of monitoring data (e.g., metrics or logs) with either partial or full labels, which miss a large number of anomalies and is costly to manually tag. Therefore, we propose an unsupervised Microservice system Anomaly Detection method via Contrastive Multi-modal representation Clustering (MAD-CMC) to tackle these issues. MAD-CMC first adopts a hierarchical architecture to simultaneously explore the spatial–temporal correlation of metrics and log context information. Next, to facilitate metrics-logs interaction, MAD-CMC introduces a cross-modal Transformer, which outputs multi-modal representation for clustering. During the clustering process, we design a multi-grained contrastive learning approach. Benefiting from the clustering results, MAD-CMC bring intra-cluster representation closer while pushing inter-cluster representation farther away at both inter- and intra-modality aspect. Considering that normal samples are simpler and far more numerous than abnormal sample, we propose a dynamic weighting formula, and apply it to contrastive loss to improve the model’s discrimination ability. Sufficient experiments on public dataset show that MAD-CMC outperforms state-of-the-art methods.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104013"},"PeriodicalIF":7.4,"publicationDate":"2024-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138907","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

Analysis of user experience in low-resource languages: A case study of the Uzbek language Google Play reviews

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

Information Processing & Management

Pub Date : 2024-12-08 DOI: 10.1016/j.ipm.2024.104015

Aizihaierjiang Yusufu , Abidan Ainiwaer , Bobo Li , Fei Li , Aizierguli Yusufu , Donghong Ji

Understanding user experience is crucial for business success, yet analyzing user reviews in low-resource languages presents significant challenges due to the scarcity of annotated data. To address this gap, we conducted an in-depth analysis of 27,985 Uzbek reviews from the Google Play Store, focusing on the six key aspects of the User Experience Honeycomb model. Our study meticulously annotated these reviews, comprising a total of 43,712 sentences, to assess the sentiment polarity across these six dimensions. To benchmark this task, we propose an integrated framework that leverages pre-trained models along with GCN to capture semantic relationships, thereby enhancing the accuracy of sentiment analysis. Our approach demonstrated superior performance, achieving an absolute improvement of 0.30 in the F1 score for multi-classification tasks and 0.43 for binary classification tasks compared to existing baseline methods. These results underscore the effectiveness of our proposed framework in understanding user experience in low-resource language contexts, offering valuable insights for businesses and researchers alike.

{"title":"Analysis of user experience in low-resource languages: A case study of the Uzbek language Google Play reviews","authors":"Aizihaierjiang Yusufu , Abidan Ainiwaer , Bobo Li , Fei Li , Aizierguli Yusufu , Donghong Ji","doi":"10.1016/j.ipm.2024.104015","DOIUrl":"10.1016/j.ipm.2024.104015","url":null,"abstract":"<div><div>Understanding user experience is crucial for business success, yet analyzing user reviews in low-resource languages presents significant challenges due to the scarcity of annotated data. To address this gap, we conducted an in-depth analysis of 27,985 Uzbek reviews from the Google Play Store, focusing on the six key aspects of the User Experience Honeycomb model. Our study meticulously annotated these reviews, comprising a total of 43,712 sentences, to assess the sentiment polarity across these six dimensions. To benchmark this task, we propose an integrated framework that leverages pre-trained models along with GCN to capture semantic relationships, thereby enhancing the accuracy of sentiment analysis. Our approach demonstrated superior performance, achieving an absolute improvement of 0.30 in the F1 score for multi-classification tasks and 0.43 for binary classification tasks compared to existing baseline methods. These results underscore the effectiveness of our proposed framework in understanding user experience in low-resource language contexts, offering valuable insights for businesses and researchers alike.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104015"},"PeriodicalIF":7.4,"publicationDate":"2024-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138904","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

Predicting counseling behavioral propensity based on temporal return visits patterns and current perceived intensity with chronic conditions management

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

Information Processing & Management

Pub Date : 2024-12-07 DOI: 10.1016/j.ipm.2024.104024

Qinkai Luo, Chao Yang, Jun Yang

The healthcare demand for online medical counseling has increased. Management of chronic conditions through Internet hospitals and online platforms has shifted from cure-centered approaches to service-oriented counseling. Previous studies have given less attention to the combined behavioral and perceived determinants that influence subsequent counseling, particularly in terms of types and timing. Our research identifies key determinants including sequential intertemporal behavioral patterns of return visits, perceived current counseling intensity measured by perceived usefulness and emotional attitudes, as well as diagnosis-oriented clustering enhancements suitable for chronic conditions specialties. The Diagnosis-specific Interval-scaled Perception-sensitive (DIPs) framework integrates nearly 380 thousand real dialogues from Chinese electronic healthcare records (CEHRs) and auxiliary information. Performance evaluations using the receiver operating characteristic (ROC) curve and the precision-recall curve (PRC) yield high scores of 0.95 Area Under the ROC Curve (AUROC) and 0.72 Area Under the Precision-Recall Curve (AUPRC), which is significant in unbalanced multi-classification tasks offering a solution to chronic online counseling behavioral propensity estimation and widespread adoption in real-world settings. DIPs framework's credible quantitative interpretability provides insights into prioritizing behavioral and perceptual impacts over computational features in counseling propensity predictions. Platforms and physicians can facilitate targeted interventions that align patients’ expectations with the sustainable delivery of on-demand services in chronic conditions management.

{"title":"Predicting counseling behavioral propensity based on temporal return visits patterns and current perceived intensity with chronic conditions management","authors":"Qinkai Luo, Chao Yang, Jun Yang","doi":"10.1016/j.ipm.2024.104024","DOIUrl":"10.1016/j.ipm.2024.104024","url":null,"abstract":"<div><div>The healthcare demand for online medical counseling has increased. Management of chronic conditions through Internet hospitals and online platforms has shifted from cure-centered approaches to service-oriented counseling. Previous studies have given less attention to the combined behavioral and perceived determinants that influence subsequent counseling, particularly in terms of types and timing. Our research identifies key determinants including sequential intertemporal behavioral patterns of return visits, perceived current counseling intensity measured by perceived usefulness and emotional attitudes, as well as diagnosis-oriented clustering enhancements suitable for chronic conditions specialties. The Diagnosis-specific Interval-scaled Perception-sensitive (DIPs) framework integrates nearly 380 thousand real dialogues from Chinese electronic healthcare records (CEHRs) and auxiliary information. Performance evaluations using the receiver operating characteristic (ROC) curve and the precision-recall curve (PRC) yield high scores of 0.95 Area Under the ROC Curve (AUROC) and 0.72 Area Under the Precision-Recall Curve (AUPRC), which is significant in unbalanced multi-classification tasks offering a solution to chronic online counseling behavioral propensity estimation and widespread adoption in real-world settings. DIPs framework's credible quantitative interpretability provides insights into prioritizing behavioral and perceptual impacts over computational features in counseling propensity predictions. Platforms and physicians can facilitate targeted interventions that align patients’ expectations with the sustainable delivery of on-demand services in chronic conditions management.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104024"},"PeriodicalIF":7.4,"publicationDate":"2024-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138903","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

CLAVE: A deep learning model for source code authorship verification with contrastive learning and transformer encoders

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

Information Processing & Management

Pub Date : 2024-12-06 DOI: 10.1016/j.ipm.2024.104005

David Álvarez-Fidalgo , Francisco Ortin

Source code authorship verification involves determining whether two code fragments are written by the same programmer. It has many uses, including malware authorship analysis, copyright dispute resolution and plagiarism detection. Source code authorship verification is challenging because it must generalize to code written by programmers not included in its training data. In this paper, we present CLAVE (Contrastive Learning for Authorship Verification with Encoder representations), a novel deep learning model for source code authorship verification that leverages contrastive learning and a Transformer Encoder-based architecture. We initially pre-train CLAVE on a dataset of 270,602 Python source code files extracted from GitHub. Subsequently, we fine-tune CLAVE for authorship verification using contrastive learning on Python submissions from 61,956 distinct programmers in Google Code Jam and Kick Start competitions. This approach allows the model to learn stylometric representations of source code, enabling comparison via vector distance for authorship verification. CLAVE achieves an AUC of 0.9782, reduces the error of the state-of-the-art source code authorship verification systems by at least 23.4% and improves the AUC of cutting-edge source code LLMs by 21.9% to 40%. We also evaluate the main components of CLAVE on its AUC performance improvement: pre-training (1.8%), loss function (0.2%–2.8%), input length (0.1%–0.7%), model size (0.2%), and tokenizer (0.1%–0.7%).

{"title":"CLAVE: A deep learning model for source code authorship verification with contrastive learning and transformer encoders","authors":"David Álvarez-Fidalgo , Francisco Ortin","doi":"10.1016/j.ipm.2024.104005","DOIUrl":"10.1016/j.ipm.2024.104005","url":null,"abstract":"<div><div>Source code authorship verification involves determining whether two code fragments are written by the same programmer. It has many uses, including malware authorship analysis, copyright dispute resolution and plagiarism detection. Source code authorship verification is challenging because it must generalize to code written by programmers not included in its training data. In this paper, we present CLAVE (Contrastive Learning for Authorship Verification with Encoder representations), a novel deep learning model for source code authorship verification that leverages contrastive learning and a Transformer Encoder-based architecture. We initially pre-train CLAVE on a dataset of 270,602 Python source code files extracted from GitHub. Subsequently, we fine-tune CLAVE for authorship verification using contrastive learning on Python submissions from 61,956 distinct programmers in Google Code Jam and Kick Start competitions. This approach allows the model to learn stylometric representations of source code, enabling comparison via vector distance for authorship verification. CLAVE achieves an AUC of 0.9782, reduces the error of the state-of-the-art source code authorship verification systems by at least 23.4% and improves the AUC of cutting-edge source code LLMs by 21.9% to 40%. We also evaluate the main components of CLAVE on its AUC performance improvement: pre-training (1.8%), loss function (0.2%–2.8%), input length (0.1%–0.7%), model size (0.2%), and tokenizer (0.1%–0.7%).</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104005"},"PeriodicalIF":7.4,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138901","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}

引用次数: 0

SecNet: A second order neural network for MI-EEG

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

Information Processing & Management

Pub Date : 2024-12-06 DOI: 10.1016/j.ipm.2024.104012

Wei Liang , Brendan Z. Allison , Ren Xu , Xinjie He , Xingyu Wang , Andrzej Cichocki , Jing Jin

Motor imagery-based brain-computer interfaces (MI-BCIs) hold significant potential for individuals with severe paralysis who are aware and alert by but unable to reliably control their muscles. MI-BCIs have garnered increasing attention from researchers in the field of motor function rehabilitation as a healthcare technology. However, the variability of inter-session and the difficulty in extracting energy features bring huge challenges to the information processing of such systems. To overcome this, we propose a novel architecture called SecNet, designed to capture relationships between convolutional features. SecNet utilizes multiple branches to learn spatio-temporal features of EEG signals and then pools them into a covariance. We integrate regularization and attention mechanisms to enhance the learning efficiency of features, followed by the adoption of a second order pooling method. Lastly, we employ Riemannian geometry learning to map features derived from symmetric positive definite covariance matrices. Evaluation experiments are conducted on a dataset from stroke patients and further compared its performance on two public datasets. Experimental results show that the SecNet is superior to the benchmark methods and achieves accuracy rates of 72.90%, 87.08% and 74.28% on the Stroke dataset, BCI IV 2a dataset and OpenBMI dataset, respectively. These results demonstrate its efficacy and robustness in inter-session decoding for MI-BCI, showing its practical utility for application. Our code is available at https://github.com/SecNet-mi/SecNet.

{"title":"SecNet: A second order neural network for MI-EEG","authors":"Wei Liang , Brendan Z. Allison , Ren Xu , Xinjie He , Xingyu Wang , Andrzej Cichocki , Jing Jin","doi":"10.1016/j.ipm.2024.104012","DOIUrl":"10.1016/j.ipm.2024.104012","url":null,"abstract":"<div><div>Motor imagery-based brain-computer interfaces (MI-BCIs) hold significant potential for individuals with severe paralysis who are aware and alert by but unable to reliably control their muscles. MI-BCIs have garnered increasing attention from researchers in the field of motor function rehabilitation as a healthcare technology. However, the variability of inter-session and the difficulty in extracting energy features bring huge challenges to the information processing of such systems. To overcome this, we propose a novel architecture called SecNet, designed to capture relationships between convolutional features. SecNet utilizes multiple branches to learn spatio-temporal features of EEG signals and then pools them into a covariance. We integrate regularization and attention mechanisms to enhance the learning efficiency of features, followed by the adoption of a second order pooling method. Lastly, we employ Riemannian geometry learning to map features derived from symmetric positive definite covariance matrices. Evaluation experiments are conducted on a dataset from stroke patients and further compared its performance on two public datasets. Experimental results show that the SecNet is superior to the benchmark methods and achieves accuracy rates of 72.90%, 87.08% and 74.28% on the Stroke dataset, BCI IV 2a dataset and OpenBMI dataset, respectively. These results demonstrate its efficacy and robustness in inter-session decoding for MI-BCI, showing its practical utility for application. Our code is available at <span><span>https://github.com/SecNet-mi/SecNet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104012"},"PeriodicalIF":7.4,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138858","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

Attention-based multi-layer network representation learning framework for network alignment

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

Information Processing & Management

Pub Date : 2024-12-06 DOI: 10.1016/j.ipm.2024.104009

Yao Li , He Cai , Huilin Liu

Network alignment, which aims at finding the node correspondences between networks, is the cornerstone of multi-network applications. Existing efforts on network alignment suffer from the alignment space misregistration problem (i.e., the alignment spaces of two networks are not matched) and the alignment inconsistency problem (i.e., the consistency assumptions they held cannot be satisfied). To tackle these problems, in this paper, we propose an Attention-based Multi-layer Network representation learning framework for network alignment, named AMN. Specifically, to tackle alignment space misregistration problem, a novel network fusion strategy is proposed. It can establish connections between networks while preserving the specific information in each network. Based on this strategy, two networks are learned simultaneously and the representation spaces of them are matched. Secondly, an attention-based multi-layer graph neural network named A-GNN is devised, in which an innovative inter-layer attention mechanism is proposed. Different from existing attention mechanisms, the proposed inter-layer attention mechanism learns vector weights, so that it can fine-tune the consistent information in each dimension. Hence, AMN can make full use of the consistent information and alleviate the influence of alignment inconsistency problem. Experiments conducted on 4 kinds of real-world datasets show that AMN outperforms 9 state-of-the-art methods by at least 0.007–0.671 in terms of precision@1.

{"title":"Attention-based multi-layer network representation learning framework for network alignment","authors":"Yao Li , He Cai , Huilin Liu","doi":"10.1016/j.ipm.2024.104009","DOIUrl":"10.1016/j.ipm.2024.104009","url":null,"abstract":"<div><div>Network alignment, which aims at finding the node correspondences between networks, is the cornerstone of multi-network applications. Existing efforts on network alignment suffer from the alignment space misregistration problem (i.e., the alignment spaces of two networks are not matched) and the alignment inconsistency problem (i.e., the consistency assumptions they held cannot be satisfied). To tackle these problems, in this paper, we propose an Attention-based Multi-layer Network representation learning framework for network alignment, named AMN. Specifically, to tackle alignment space misregistration problem, a novel network fusion strategy is proposed. It can establish connections between networks while preserving the specific information in each network. Based on this strategy, two networks are learned simultaneously and the representation spaces of them are matched. Secondly, an attention-based multi-layer graph neural network named A-GNN is devised, in which an innovative inter-layer attention mechanism is proposed. Different from existing attention mechanisms, the proposed inter-layer attention mechanism learns vector weights, so that it can fine-tune the consistent information in each dimension. Hence, AMN can make full use of the consistent information and alleviate the influence of alignment inconsistency problem. Experiments conducted on 4 kinds of real-world datasets show that AMN outperforms 9 state-of-the-art methods by at least 0.007–0.671 in terms of precision@1.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104009"},"PeriodicalIF":7.4,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138900","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

DMGAE: An interpretable representation learning method for directed scale-free networks based on autoencoder and masking

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

Information Processing & Management

Pub Date : 2024-12-06 DOI: 10.1016/j.ipm.2024.104007

Qin-Cheng Yang , Kai Yang , Zhao-Long Hu , Minglu Li

Although existing graph self-supervised learning approaches have paid attention to the directed nature of networks, they have often overlooked the ubiquitous scale-free attributes. This oversight has resulted in a theoretical gap in understanding graph self-supervised learning from the perspective of network structure. In this paper, we study the degree distribution characteristics of source and target nodes in directed scale-free networks, encompassing node and edge dimensions. Our theoretical analysis reveals the relationship between the average degree of nodes and their average in-degree and out-degree, which is instrumental in discerning positive and negative edges, as well as edge directionality. Here positive edges are the ones that exists in the original graph, and negative edges are the ones that not exists in the original graph. Furthermore, we uncover negative edges connecting to central nodes and positive edges to peripheral nodes to be less predictable. Based on these crucial theoretical insights, we propose DMGAE (Directed Masked Graph Autoencoder), a novel representation learning method for directed scale-free networks that offers interpretability. The DMGAE method employs a weighted graph based on edges to replace the original graph structure. It integrates a masking approach based on the weight of the edges. Additionally, it incorporates an adaptive negative sampling method, edge decoder and a degree decoder based on the difference between the in-degree and out-degree of the node. This enhances the model’s capability to learn edges and discern their directions. Empirical studies on extensive real-world network data show that, compared to the state-of-the-art methods, DMGAE not only generally has superior learning performance on directed networks, but also performs exceedingly well on undirected networks.

{"title":"DMGAE: An interpretable representation learning method for directed scale-free networks based on autoencoder and masking","authors":"Qin-Cheng Yang , Kai Yang , Zhao-Long Hu , Minglu Li","doi":"10.1016/j.ipm.2024.104007","DOIUrl":"10.1016/j.ipm.2024.104007","url":null,"abstract":"<div><div>Although existing graph self-supervised learning approaches have paid attention to the directed nature of networks, they have often overlooked the ubiquitous scale-free attributes. This oversight has resulted in a theoretical gap in understanding graph self-supervised learning from the perspective of network structure. In this paper, we study the degree distribution characteristics of source and target nodes in directed scale-free networks, encompassing node and edge dimensions. Our theoretical analysis reveals the relationship between the average degree of nodes and their average in-degree and out-degree, which is instrumental in discerning positive and negative edges, as well as edge directionality. Here positive edges are the ones that exists in the original graph, and negative edges are the ones that not exists in the original graph. Furthermore, we uncover negative edges connecting to central nodes and positive edges to peripheral nodes to be less predictable. Based on these crucial theoretical insights, we propose DMGAE (Directed Masked Graph Autoencoder), a novel representation learning method for directed scale-free networks that offers interpretability. The DMGAE method employs a weighted graph based on edges to replace the original graph structure. It integrates a masking approach based on the weight of the edges. Additionally, it incorporates an adaptive negative sampling method, edge decoder and a degree decoder based on the difference between the in-degree and out-degree of the node. This enhances the model’s capability to learn edges and discern their directions. Empirical studies on extensive real-world network data show that, compared to the state-of-the-art methods, DMGAE not only generally has superior learning performance on directed networks, but also performs exceedingly well on undirected networks.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"62 3","pages":"Article 104007"},"PeriodicalIF":7.4,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138859","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