Shaoyong Du
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Shaoyong Du has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Xiangyang Luo
Information Engineering University, Zhengzhou, China
7 shared publications - Jingyu Hua
Nanjing University
5 shared publications - Sheng Zhong
State Key Laboratory of Novel Software Technology, Nanjing University, Nanjing, Jiangsu, China
5 shared publications - Fenlin Liu
Henan Key Laboratory of Cyberspace Situation Awareness, Information Engineering University, Zhengzhou, China
2 shared publications - Haifeng Yang
Henan Key Laboratory of Cyberspace Situation Awareness, Information Engineering University, Zhengzhou, China
2 shared publications Meng Zhang
Henan Province Key Laboratory of Cyberspace Situation Awareness, Peoples Liberation Army Information Engineering University, Zhengzhou, China
2 shared publications
Latest Publications
- 30 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-16
SHSLoc: Semi-Hard Sample Calibration and BCLPSO for Indoor Localization
Indoor localization is critical for applications in internet of thing, artificial intelligence, and location-based services. However, existing methods lack dynamic optimization mechanisms to address ambiguous boundary sample, resulting in insufficient fingerprint discrimination, high confusion, and limited accuracy in similar environments. Therefore, this paper proposes an indoor localization method based on semi-hard sample calibration and bayesian comprehensive learning particle swarm optimization (BCLPSO). This method includes offline training and online localization. In the offline training, channel state information (CSI) undergoes phase extraction and preprocessing to enhance angle of arrival (AoA) estimation accuracy. CSI samples are constructed using AoA and processed through a designed LocNet network. Then, the directional mining strategy is introduced to screen semi-hard samples, and the training triplet is dynamically constructed accordingly. Finally, LocNet is embedded as a weight-shared sub-network within the triplet architecture, and an adaptive discriminant boundary loss function is designed for network calibration. High-discriminability feature vectors extracted by the calibrated network are used to build the fingerprint database. In the online localization, features from target data are mapped into the embedding space using the calibrated triplet network. Position estimation is performed via BCLPSO. Extensive experiments in meeting room and laboratory environments show that the MAE and STD of the SHSLoc are 1.59 and 1.02 m (meeting room) and 2.02 and 1.50 m (laboratory), respectively, which are superior to the baseline method in terms of positioning accuracy and stability.
- 30 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. PP (00189545) , 1-16
SHSLoc: Semi-Hard Sample Calibration and BCLPSO for Indoor Localization
Indoor localization is critical for applications in internet of thing, artificial intelligence, and location-based services. However, existing methods lack dynamic optimization mechanisms to address ambiguous boundary sample, resulting in insufficient fingerprint discrimination, high confusion, and limited accuracy in similar environments. Therefore, this paper proposes an indoor localization method based on semi-hard sample calibration and bayesian comprehensive learning particle swarm optimization (BCLPSO). This method includes offline training and online localization. In the offline training, channel state information (CSI) undergoes phase extraction and preprocessing to enhance angle of arrival (AoA) estimation accuracy. CSI samples are constructed using AoA and processed through a designed LocNet network. Then, the directional mining strategy is introduced to screen semi-hard samples, and the training triplet is dynamically constructed accordingly. Finally, LocNet is embedded as a weight-shared sub-network within the triplet architecture, and an adaptive discriminant boundary loss function is designed for network calibration. High-discriminability feature vectors extracted by the calibrated network are used to build the fingerprint database. In the online localization, features from target data are mapped into the embedding space using the calibrated triplet network. Position estimation is performed via BCLPSO. Extensive experiments in meeting room and laboratory environments show that the MAE and STD of the SHSLoc are 1.59 and 1.02 m (meeting room) and 2.02 and 1.50 m (laboratory), respectively, which are superior to the baseline method in terms of positioning accuracy and stability.
- 31 December 2025
- Published by Elsevier in Computers & Security
- Vol. 163
A survey on network flow watermarking: A problem-oriented perspective
- 31 December 2025
- Published by Elsevier in Computers & Security
- Vol. 163
A survey on network flow watermarking: A problem-oriented perspective
- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 365-381
An Efficient Website Fingerprinting for New Websites Emerging Based on Incremental Learning
Website fingerprinting attacks leverage encrypted traffic features to identify specific services accessed by users within anonymity networks such as Tor. Although existing WF methods achieve high accuracy on static datasets using deep learning techniques, they struggle in dynamic environments where anonymous websites continually evolve. These methods typically require full retraining on composite datasets, resulting in substantial computational and storage burdens, and are particularly vulnerable to classification bias caused by data imbalance and concept drift. To address these challenges, we propose EIL-WF, a dynamic WF framework based on incremental learning that enables efficient adaptation to newly emerging websites without the need for full retraining. EIL-WF incrementally trains lightweight, independent classifiers for new website classes and integrates them through classifier normalization and energy alignment strategies grounded in energy-based model theory, thereby constructing a unified and robust classification model. Comprehensive experiments on two public Tor traffic datasets demonstrate that EIL-WF outperforms existing incremental learning methods by 6.2%–20.2% in identifying new websites and reduces catastrophic forgetting by 5.4%–20%. Notably, EIL-WF exhibits strong resilience against data imbalance and concept drift, maintaining stable classification performance across evolving distributions. Furthermore,EIL-WF decreases training time during model updates by 2–3 orders of magnitude, demonstrating substantial advantages over conventional full retraining paradigms.
- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 365-381
An Efficient Website Fingerprinting for New Websites Emerging Based on Incremental Learning
Website fingerprinting attacks leverage encrypted traffic features to identify specific services accessed by users within anonymity networks such as Tor. Although existing WF methods achieve high accuracy on static datasets using deep learning techniques, they struggle in dynamic environments where anonymous websites continually evolve. These methods typically require full retraining on composite datasets, resulting in substantial computational and storage burdens, and are particularly vulnerable to classification bias caused by data imbalance and concept drift. To address these challenges, we propose EIL-WF, a dynamic WF framework based on incremental learning that enables efficient adaptation to newly emerging websites without the need for full retraining. EIL-WF incrementally trains lightweight, independent classifiers for new website classes and integrates them through classifier normalization and energy alignment strategies grounded in energy-based model theory, thereby constructing a unified and robust classification model. Comprehensive experiments on two public Tor traffic datasets demonstrate that EIL-WF outperforms existing incremental learning methods by 6.2%–20.2% in identifying new websites and reduces catastrophic forgetting by 5.4%–20%. Notably, EIL-WF exhibits strong resilience against data imbalance and concept drift, maintaining stable classification performance across evolving distributions. Furthermore,EIL-WF decreases training time during model updates by 2–3 orders of magnitude, demonstrating substantial advantages over conventional full retraining paradigms.
- 22 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23 (2) , 2100-2116
An Empirical Analysis of Information Leakage of File Operations on Android External Storage
Current Android apps rely heavily on external storage. When using the external storage, apps apply different security strategies (e.g., randomizing file name, encrypting file content) to prevent privacy risks. Even so, we find that privacy risks still exist, i.e., information leakage of apps' file operations. Users follow their habits to run apps and some of the apps conduct file operations on external storage, which potentially expose users' regular activities. In this paper, we conduct the first empirical study on this problem and implement a file-operation-based pipeline, OP-PERUSE. Besides a dataset of 5,359,339 file operation events collected from the volunteers, we crawl 22,484 app records from the third-party app statistics websites. By combining these data, we get some timely and fine-grained information about the users, e.g., current affiliation, position, habits, etc. To further understand this problem's severity, we conduct a static code analysis on 15,098 apps. We find that 1,305 (8.64%) apps tend to collect file operation events, and more than half of these apps adopt the third-party SDKs which gather file operation events, indicating this problem could have persisted over a long period of time. To prevent this problem, we provide some security recommendations for different stakeholders.
- 22 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 23 (2) , 2100-2116
An Empirical Analysis of Information Leakage of File Operations on Android External Storage
Current Android apps rely heavily on external storage. When using the external storage, apps apply different security strategies (e.g., randomizing file name, encrypting file content) to prevent privacy risks. Even so, we find that privacy risks still exist, i.e., information leakage of apps' file operations. Users follow their habits to run apps and some of the apps conduct file operations on external storage, which potentially expose users' regular activities. In this paper, we conduct the first empirical study on this problem and implement a file-operation-based pipeline, OP-PERUSE. Besides a dataset of 5,359,339 file operation events collected from the volunteers, we crawl 22,484 app records from the third-party app statistics websites. By combining these data, we get some timely and fine-grained information about the users, e.g., current affiliation, position, habits, etc. To further understand this problem's severity, we conduct a static code analysis on 15,098 apps. We find that 1,305 (8.64%) apps tend to collect file operation events, and more than half of these apps adopt the third-party SDKs which gather file operation events, indicating this problem could have persisted over a long period of time. To prevent this problem, we provide some security recommendations for different stakeholders.
- Open Access
- 26 May 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on the Web
- Vol. 19 (2) , 1-24
Twitter User Geolocation Based on Location Feature Enhancement
User location discovery from social media is crucial for location-based services such as emergency awareness and event monitoring. Existing approaches generally integrate user-generated text features and social relationships but insufficiently explore location-specific features and geographically proximate relationships, leading to suboptimal accuracy. In this article, we propose a Twitter user geolocation method based on location feature enhancement to better capture the location characteristics in users’ tweets and social relationships. Specifically, a user tweet representation algorithm based on location feature separation (TwLS) is designed. By leveraging words’ location-aware weight matrix and pre-trained embeddings, TwLS calculates a tweet representation for each user in every location, explicitly indicating the relevance between users and various locations. Additionally, we develop the local celebrity discovery method (LocCel) to construct social networks by identifying and preserving geographically concentrated high-degree nodes while filtering noise. Thereby, LocCel enhances local relationships and strengthens location-proximate connections within the user social network. Experiments on two real-world datasets show that our method outperforms seven baselines, improving user geolocation accuracy by 3.1% ∼ 8.1% and 1.8% ∼ 8.8%, while reducing median error by 22.2% ∼ 52.8% and 19.4% ∼ 50.7%, respectively.
- 26 May 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on the Web
- Vol. 19 (2) , 1-24
Open AccessTwitter User Geolocation Based on Location Feature Enhancement
User location discovery from social media is crucial for location-based services such as emergency awareness and event monitoring. Existing approaches generally integrate user-generated text features and social relationships but insufficiently explore location-specific features and geographically proximate relationships, leading to suboptimal accuracy. In this article, we propose a Twitter user geolocation method based on location feature enhancement to better capture the location characteristics in users’ tweets and social relationships. Specifically, a user tweet representation algorithm based on location feature separation (TwLS) is designed. By leveraging words’ location-aware weight matrix and pre-trained embeddings, TwLS calculates a tweet representation for each user in every location, explicitly indicating the relevance between users and various locations. Additionally, we develop the local celebrity discovery method (LocCel) to construct social networks by identifying and preserving geographically concentrated high-degree nodes while filtering noise. Thereby, LocCel enhances local relationships and strengthens location-proximate connections within the user social network. Experiments on two real-world datasets show that our method outperforms seven baselines, improving user geolocation accuracy by 3.1% ∼ 8.1% and 1.8% ∼ 8.8%, while reducing median error by 22.2% ∼ 52.8% and 19.4% ∼ 50.7%, respectively.
- 16 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 30366-30378
Localization Algorithm Based on the Relationship Between Trapezoidal Trajectory and Energy Consumption of Mobile Anchor Nodes
Node localization technology is increasingly receiving extensive attention from academia and industry due to its strong concealment and high fault tolerance in wireless sensor networks (WSNs). Mobile anchor nodes (MANs) assisted localization is often used in existing WSNs. However, assisted localization based on MANs is still a challenging problem. On one hand, it is difficult to determine the number of anchor nodes (ANs) to support the energy required for the entire movement trajectory. On the other hand, unknown nodes (UNs) at the boundary region are difficult to obtain sufficient beacon information for localization. A localization algorithm based on the relationship between trapezoidal trajectory and energy consumption of MANs is proposed to solve this challenging problem in the current research. In the proposed algorithm, we design a trapezoidal trajectory based localization algorithm for MANs (TTLMA) to optimize the movement trajectory of ANs. At the same time, determine the number of ANs by analyzing the relationship between the initial energy of ANs and the energy required by the localization algorithm. Select an appropriate algorithm to locate UNs according to the number of beacon information received by them. We conducted multiple simulations to evaluate the proposed algorithm’s performance. The experimental results indicate that compared with five existing typical localization algorithms, the proposed algorithms have positive advantages in terms of localization error and coverage, with average localization error reduced by 0.15 m–1.16 m, average localization coverage improved by 8%–38%. Moreover, the energy consumption of the proposed algorithm is relatively low, requiring only one AN to traverse the designed trapezoidal trajectory.
- 16 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (15) , 30366-30378
Localization Algorithm Based on the Relationship Between Trapezoidal Trajectory and Energy Consumption of Mobile Anchor Nodes
Node localization technology is increasingly receiving extensive attention from academia and industry due to its strong concealment and high fault tolerance in wireless sensor networks (WSNs). Mobile anchor nodes (MANs) assisted localization is often used in existing WSNs. However, assisted localization based on MANs is still a challenging problem. On one hand, it is difficult to determine the number of anchor nodes (ANs) to support the energy required for the entire movement trajectory. On the other hand, unknown nodes (UNs) at the boundary region are difficult to obtain sufficient beacon information for localization. A localization algorithm based on the relationship between trapezoidal trajectory and energy consumption of MANs is proposed to solve this challenging problem in the current research. In the proposed algorithm, we design a trapezoidal trajectory based localization algorithm for MANs (TTLMA) to optimize the movement trajectory of ANs. At the same time, determine the number of ANs by analyzing the relationship between the initial energy of ANs and the energy required by the localization algorithm. Select an appropriate algorithm to locate UNs according to the number of beacon information received by them. We conducted multiple simulations to evaluate the proposed algorithm’s performance. The experimental results indicate that compared with five existing typical localization algorithms, the proposed algorithms have positive advantages in terms of localization error and coverage, with average localization error reduced by 0.15 m–1.16 m, average localization coverage improved by 8%–38%. Moreover, the energy consumption of the proposed algorithm is relatively low, requiring only one AN to traverse the designed trapezoidal trajectory.
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 12 (2) , 1280-1296
IPv6Landmarker: Enhancing IPv6 Street-Level Geolocation Through Network Landmark Mining and Targeted Updates
IP geolocation accuracy heavily relies on the availability of numerous high-quality network landmarks. However, IPv6 geolocation faces challenges due to its vast address space and rotating prefixes. Existing landmark mining methods struggle to meet the stringent demands of IPv6 street-level geolocation. We introduce IPv6Landmarker, a novel approach that enhances IPv6 geolocation precision through landmark mining and targeted updates. By associating WAN IPv6 addresses with WiFi BSSIDs in wireless routers, we employ a multi-association coordinate filtering algorithm to select reliable IPv6 street-level landmarks. We also implement targeted updates based on IPv6 prefix rotation patterns. Using real-world data, we demonstrate significant improvements, including a range increase of 16.75% to 46.68% in candidate landmarks acquired globally and of 10.06% to 126.39% in landmarks acquired specifically within target cities. In particular, there is a range of 16.67% to 66.67% enhancement in the geolocation success of ground truth landmarks, coupled with a range of 6.09% to 40.34% reduction in geolocation error. Additionally, it shows a remarkable 82.36% improvement in landmark set stability.
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network Science and Engineering
- Vol. 12 (2) , 1280-1296
IPv6Landmarker: Enhancing IPv6 Street-Level Geolocation Through Network Landmark Mining and Targeted Updates
IP geolocation accuracy heavily relies on the availability of numerous high-quality network landmarks. However, IPv6 geolocation faces challenges due to its vast address space and rotating prefixes. Existing landmark mining methods struggle to meet the stringent demands of IPv6 street-level geolocation. We introduce IPv6Landmarker, a novel approach that enhances IPv6 geolocation precision through landmark mining and targeted updates. By associating WAN IPv6 addresses with WiFi BSSIDs in wireless routers, we employ a multi-association coordinate filtering algorithm to select reliable IPv6 street-level landmarks. We also implement targeted updates based on IPv6 prefix rotation patterns. Using real-world data, we demonstrate significant improvements, including a range increase of 16.75% to 46.68% in candidate landmarks acquired globally and of 10.06% to 126.39% in landmarks acquired specifically within target cities. In particular, there is a range of 16.67% to 66.67% enhancement in the geolocation success of ground truth landmarks, coupled with a range of 6.09% to 40.34% reduction in geolocation error. Additionally, it shows a remarkable 82.36% improvement in landmark set stability.
- Open Access
- 3 May 2023
- Published by EDP Sciences in Security and Safety
- Vol. 2
An accurate identification method for network devices based on spatial attention mechanism
With the metaverse being the development direction of the next generation Internet, the popularity of intelligent devices, and the maturity of various emerging technologies, more and more intelligent devices try to connect to the Internet, which poses a major threat to the management and security protection of network equipment. At present, the mainstream method of network equipment identification in the metaverse is to obtain the network traffic data generated in the process of device communication, extract the device features through analysis and processing, and identify the device based on a variety of learning algorithms. Such methods often require manual participation, and it is difficult to capture the small differences between similar devices, leading to identification errors. Therefore, we propose a deep learning device recognition method based on a spatial attention mechanism. Firstly, we extract the required feature fields from the acquired network traffic data. Then, we normalize the data and convert it into grayscale images. After that, we add a spatial attention mechanism to CNN and MLP respectively to increase the difference between similar network devices and further improve the recognition accuracy. Finally, we identify devices based on the deep learning model. A large number of experiments were carried out on 31 types of network devices such as web cameras, wireless routers, and smartwatches. The results show that the accuracy of the proposed recognition method based on the spatial attention mechanism is increased by 0.8% and 2.0%, respectively, compared with the recognition method based only on the deep learning model under the CNN and MLP models. The method proposed in this paper is significantly superior to the existing method of device-type recognition based only on a deep learning model.
- 3 May 2023
- Published by EDP Sciences in Security and Safety
- Vol. 2
Open AccessAn accurate identification method for network devices based on spatial attention mechanism
With the metaverse being the development direction of the next generation Internet, the popularity of intelligent devices, and the maturity of various emerging technologies, more and more intelligent devices try to connect to the Internet, which poses a major threat to the management and security protection of network equipment. At present, the mainstream method of network equipment identification in the metaverse is to obtain the network traffic data generated in the process of device communication, extract the device features through analysis and processing, and identify the device based on a variety of learning algorithms. Such methods often require manual participation, and it is difficult to capture the small differences between similar devices, leading to identification errors. Therefore, we propose a deep learning device recognition method based on a spatial attention mechanism. Firstly, we extract the required feature fields from the acquired network traffic data. Then, we normalize the data and convert it into grayscale images. After that, we add a spatial attention mechanism to CNN and MLP respectively to increase the difference between similar network devices and further improve the recognition accuracy. Finally, we identify devices based on the deep learning model. A large number of experiments were carried out on 31 types of network devices such as web cameras, wireless routers, and smartwatches. The results show that the accuracy of the proposed recognition method based on the spatial attention mechanism is increased by 0.8% and 2.0%, respectively, compared with the recognition method based only on the deep learning model under the CNN and MLP models. The method proposed in this paper is significantly superior to the existing method of device-type recognition based only on a deep learning model.
- 7 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 9 (4) , 1128-1141
UGCC: Social Media User Geolocation via Cyclic Coupling
Social media user geolocation is to infer users’ resident locations based on social media data, including user texts and social relationships. Existing methods mainly rely on the textual feature propagation in the social graph to fuse users’ textual and social information. The geolocation accuracy is susceptible to insufficient data sources and inadequate fusion. In this paper, a social media user geolocation algorithm based on cyclic coupling (called UGCC) is proposed. We collapse the social graph based on the neighbor location proximity, which reduces noisy information while enriching social relationships. Unlike existing methods that ignore the social graph's structure, UGCC measures the probability of users being in the candidate locations according to users’ structural location in the social sub-graph. Finally, we design a cyclic coupling mechanism to fuse the users’ textual and social information, which enables the two kinds of information to enhance each other and geolocate users cooperatively. Compared with ten typical existing methods (such as RELP and HGNN), experimental results show UGCC's superior performance. On two public datasets, the city-level accuracies of UGCC reach 40.8% and 50.1%; the median errors are 35.1% and 23.4% lower than the state-of-the-art methods.
- 7 February 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 9 (4) , 1128-1141
UGCC: Social Media User Geolocation via Cyclic Coupling
Social media user geolocation is to infer users’ resident locations based on social media data, including user texts and social relationships. Existing methods mainly rely on the textual feature propagation in the social graph to fuse users’ textual and social information. The geolocation accuracy is susceptible to insufficient data sources and inadequate fusion. In this paper, a social media user geolocation algorithm based on cyclic coupling (called UGCC) is proposed. We collapse the social graph based on the neighbor location proximity, which reduces noisy information while enriching social relationships. Unlike existing methods that ignore the social graph's structure, UGCC measures the probability of users being in the candidate locations according to users’ structural location in the social sub-graph. Finally, we design a cyclic coupling mechanism to fuse the users’ textual and social information, which enables the two kinds of information to enhance each other and geolocate users cooperatively. Compared with ten typical existing methods (such as RELP and HGNN), experimental results show UGCC's superior performance. On two public datasets, the city-level accuracies of UGCC reach 40.8% and 50.1%; the median errors are 35.1% and 23.4% lower than the state-of-the-art methods.
- 6 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 20 (2) , 1125-1139
Neural Attention Networks for Recommendation With Auxiliary Data
With the rapid development of Internet technologies, an increasing amount of auxiliary data can be readily obtained through web services. To alleviate the data sparsity issue, auxiliary data based recommendation has emerged for better recommendation performance. However, existing auxiliary data based methods suffer from two problems. First, only the relation features related to the meta-paths are extracted from auxiliary data, which may lead to features useful for recommendation being lost irreversibly. Second, an assumption is made that an individual has the same preference over the identical characteristic of different items, which is often invalid and may lead to misleading recommendations. Actually, a user may place different importance on the same feature of different items, and an item may get different attention from the same feature of different users. In this paper, we propose a neural network framework, named Neural Attention Recommendation model (NARec), for auxiliary data based collaborative filtering. For the first problem, we characterize users and items from three aspects, namely latent features, attribute features, and meta-path based relation features, which can comprehensively extract the useful recommendation features from auxiliary data. Regarding the second problem, we integrate different user features and item features into an attention mechanism based rating prediction model for recommendation, which can adaptively characterize the personalized features of users and items. Extensive experiments on three real-world datasets demonstrate that NARec significantly outperforms the state-of-the-art recommendation methods in the rating prediction task.
- 6 December 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 20 (2) , 1125-1139
Neural Attention Networks for Recommendation With Auxiliary Data
With the rapid development of Internet technologies, an increasing amount of auxiliary data can be readily obtained through web services. To alleviate the data sparsity issue, auxiliary data based recommendation has emerged for better recommendation performance. However, existing auxiliary data based methods suffer from two problems. First, only the relation features related to the meta-paths are extracted from auxiliary data, which may lead to features useful for recommendation being lost irreversibly. Second, an assumption is made that an individual has the same preference over the identical characteristic of different items, which is often invalid and may lead to misleading recommendations. Actually, a user may place different importance on the same feature of different items, and an item may get different attention from the same feature of different users. In this paper, we propose a neural network framework, named Neural Attention Recommendation model (NARec), for auxiliary data based collaborative filtering. For the first problem, we characterize users and items from three aspects, namely latent features, attribute features, and meta-path based relation features, which can comprehensively extract the useful recommendation features from auxiliary data. Regarding the second problem, we integrate different user features and item features into an attention mechanism based rating prediction model for recommendation, which can adaptively characterize the personalized features of users and items. Extensive experiments on three real-world datasets demonstrate that NARec significantly outperforms the state-of-the-art recommendation methods in the rating prediction task.