Xuefeng Fu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Xuefeng Fu has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
Guilin Qi
2 shared publications- Guilin Qi
Southeast University
2 shared publications - Jun Chen
Nanchang Institute of Technology, China
2 shared publications - Kailiang Wang
Nanchang Institute of Technology, China
2 shared publications - Songmao Zhang
Chinese Academy of Sciences
2 shared publications - Weikun Chen
Nanchang Institute of Technology, China
2 shared publications
Latest Publications
- Open Access
- 26 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 430-435
Fusing meta schema and attribute information's heterogeneous graph model for recipe recommendation
Recipe recommendation aims to provide some individual recipe to users. However, recipes in the real world have a wealth of diverse information, and how to effectively process and combine this information remains a challenge. To solve this problem, we propose a fusing meta schema and attribute information's heterogeneous graph model (MAHGM) to handle recipe recommendation tasks. we design two modules for modeling diverse information in heterogeneous graph. Firstly, we defined a heterogeneous graph aggregator of fusing meta schema information to mine the topological information between heterogeneous nodes. It contains a basic module to process heterogeneous neighbors’ information. On this basis, we introduce modeling of meta patterns to enhance the interaction between neighbors between nodes, so that each node fully learns the node information that interacts with it. In addition, since the results of recipe recommendations in the real world are often closely related to the characteristics and composition of the recipe itself. Therefore, we also model the attributes of recipes and integrate them into the final node embedding. Finally, we conducted many experiments to verify the validity of our model, and the experiment showed that MAHGM has superior performance compared to the baseline.
- 26 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 430-435
Open AccessFusing meta schema and attribute information's heterogeneous graph model for recipe recommendation
Recipe recommendation aims to provide some individual recipe to users. However, recipes in the real world have a wealth of diverse information, and how to effectively process and combine this information remains a challenge. To solve this problem, we propose a fusing meta schema and attribute information's heterogeneous graph model (MAHGM) to handle recipe recommendation tasks. we design two modules for modeling diverse information in heterogeneous graph. Firstly, we defined a heterogeneous graph aggregator of fusing meta schema information to mine the topological information between heterogeneous nodes. It contains a basic module to process heterogeneous neighbors’ information. On this basis, we introduce modeling of meta patterns to enhance the interaction between neighbors between nodes, so that each node fully learns the node information that interacts with it. In addition, since the results of recipe recommendations in the real world are often closely related to the characteristics and composition of the recipe itself. Therefore, we also model the attributes of recipes and integrate them into the final node embedding. Finally, we conducted many experiments to verify the validity of our model, and the experiment showed that MAHGM has superior performance compared to the baseline.
- Open Access
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 258-264
Fusion of Capsule Networks and Graph Convolution for Dual Channel Aspect-Level Sentiment Analysis
Most aspect-based sentiment analyses employ dependency trees for extracting text's syntactic structure. Nonetheless, this approach can encounter information loss and data sparsity issues. Therefore, this paper proposes BGCM model for solving these problems. The Bi-LSTM and Global Graph Network are utilized by the model to extract and integrate temporal and structural information from the original text. A Bi-GCN is also employed to fuse multiple features and combine contextual and syntactic structures. To extract deep sentiment features, the model utilizes the Capsule Network. The proposed method in this paper was put to the test using five types of publicly available data, including Twitter, Laptop14, Rest14, Rest15, and Rest16. The results of the experiments indicate that the BGCM model yielded significant improvements in performance compared to the baseline model.
- 19 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 258-264
Open AccessFusion of Capsule Networks and Graph Convolution for Dual Channel Aspect-Level Sentiment Analysis
Most aspect-based sentiment analyses employ dependency trees for extracting text's syntactic structure. Nonetheless, this approach can encounter information loss and data sparsity issues. Therefore, this paper proposes BGCM model for solving these problems. The Bi-LSTM and Global Graph Network are utilized by the model to extract and integrate temporal and structural information from the original text. A Bi-GCN is also employed to fuse multiple features and combine contextual and syntactic structures. To extract deep sentiment features, the model utilizes the Capsule Network. The proposed method in this paper was put to the test using five types of publicly available data, including Twitter, Laptop14, Rest14, Rest15, and Rest16. The results of the experiments indicate that the BGCM model yielded significant improvements in performance compared to the baseline model.
- Open Access
- 16 December 2022
- Published by Association for Computing Machinery (ACM)
- p. 553-558
A Chinese named entity approach for character semantic enhancement
Named entity recognition is a task in information extraction. In Chinese NER, it is more common to use characters as input. However, the character vector in this model carries less information and cannot represent the word information and polysemy problem. To solve these problems, BERT model, dictionary embedding and graph convolution neural network model are introduced, and a semantic enhancement method for Chinese NER is proposed. In the new method, the character vectors with prior knowledge obtained by BERT are inserted into the dictionary and fused with word-level and word-level features. Contextual semantic information is then obtained by encoding in LSTM, and input into the graph convolutional neural network to obtain the global semantic features of characters. Next, the fused semantic features are input to conditional random field for sequential label decoding. In this paper, model comparison experiments are carried out in four classical Chinese data sets. In the Resume dataset experiment, compared with the comparison model, the three indexes of the new method are optimal. In the experiment of ne and nm data sets in WeiBo data set, the index F1 of the new method is optimal. In the OntoNotes dataset experiment, P and F1 of the new method are optimal.
- 16 December 2022
- Published by Association for Computing Machinery (ACM)
- p. 553-558
Open AccessA Chinese named entity approach for character semantic enhancement
Named entity recognition is a task in information extraction. In Chinese NER, it is more common to use characters as input. However, the character vector in this model carries less information and cannot represent the word information and polysemy problem. To solve these problems, BERT model, dictionary embedding and graph convolution neural network model are introduced, and a semantic enhancement method for Chinese NER is proposed. In the new method, the character vectors with prior knowledge obtained by BERT are inserted into the dictionary and fused with word-level and word-level features. Contextual semantic information is then obtained by encoding in LSTM, and input into the graph convolutional neural network to obtain the global semantic features of characters. Next, the fused semantic features are input to conditional random field for sequential label decoding. In this paper, model comparison experiments are carried out in four classical Chinese data sets. In the Resume dataset experiment, compared with the comparison model, the three indexes of the new method are optimal. In the experiment of ne and nm data sets in WeiBo data set, the index F1 of the new method is optimal. In the OntoNotes dataset experiment, P and F1 of the new method are optimal.
- 23 August 2022
- Published by Elsevier in Expert Systems with Applications
- Vol. 211
A graph-based method for interactive mapping revision in DL-Lite
- 23 August 2022
- Published by Elsevier in Expert Systems with Applications
- Vol. 211
A graph-based method for interactive mapping revision in DL-Lite
- Open Access
- 11 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 153360-153370
Emotion Wheel Attention-Based Emotion Distribution Learning
Emotion distribution learning is an effective multi-emotion analysis model proposed in recent years. Its core idea is to record the expression degree of examples on each emotion through emotion distribution, which is suitable for handling emotion analysis tasks with emotional ambiguity. To solve the problem that the prior knowledge of emotion psychology is seldom considered in the existing emotion distribution learning methods, we propose an Emotion Wheel Attention based Emotion Distribution Learning (EWA-EDL) model. EWA-EDL generates a prior emotion distribution describing the relevance of emotional psychology for each basic emotion, and then directly integrates the prior knowledge based on the emotion wheel into the deep neural network through the attention mechanism. The deep network of EWA-EDL is trained using an end-to-end approach to learn both emotion distribution prediction and emotion classification tasks. The EWA-EDL architecture includes five main parts: input layer, convolutional layer, pooling layer, attention layer and multi-task loss layer. Extensive comparative experiments on 8 commonly used textual emotion datasets show that EWA-EDL outperforms the comparison emotion distribution learning methods on both emotion distribution prediction and emotion classification task.
- 11 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 153360-153370
Open AccessEmotion Wheel Attention-Based Emotion Distribution Learning
Emotion distribution learning is an effective multi-emotion analysis model proposed in recent years. Its core idea is to record the expression degree of examples on each emotion through emotion distribution, which is suitable for handling emotion analysis tasks with emotional ambiguity. To solve the problem that the prior knowledge of emotion psychology is seldom considered in the existing emotion distribution learning methods, we propose an Emotion Wheel Attention based Emotion Distribution Learning (EWA-EDL) model. EWA-EDL generates a prior emotion distribution describing the relevance of emotional psychology for each basic emotion, and then directly integrates the prior knowledge based on the emotion wheel into the deep neural network through the attention mechanism. The deep network of EWA-EDL is trained using an end-to-end approach to learn both emotion distribution prediction and emotion classification tasks. The EWA-EDL architecture includes five main parts: input layer, convolutional layer, pooling layer, attention layer and multi-task loss layer. Extensive comparative experiments on 8 commonly used textual emotion datasets show that EWA-EDL outperforms the comparison emotion distribution learning methods on both emotion distribution prediction and emotion classification task.
- 8 May 2019
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 11 (1) , 111-121
Incomplete label distribution learning based on supervised neighborhood information
- 8 May 2019
- Published by Springer Nature in International Journal of Machine Learning and Cybernetics
- Vol. 11 (1) , 111-121
Incomplete label distribution learning based on supervised neighborhood information
- 14 November 2018
- Published by Springer Nature
- p. 244-261
A Graph-Based Method for Interactive Mapping Revision
- 14 November 2018
- Published by Springer Nature
- p. 244-261
A Graph-Based Method for Interactive Mapping Revision
- 1 May 2016
- Published by Elsevier in Knowledge-Based Systems
- Vol. 100, 1-12
Graph-based approaches to debugging and revision of terminologies in DL-Lite
- 1 May 2016
- Published by Elsevier in Knowledge-Based Systems
- Vol. 100, 1-12
Graph-based approaches to debugging and revision of terminologies in DL-Lite
- 1 January 2013
- Published by Springer Nature in Communications in Computer and Information Science
- p. 239-242
GORS: A Graph-Based System for Ontology Revision
- 1 January 2013
- Published by Springer Nature in Communications in Computer and Information Science
- p. 239-242
GORS: A Graph-Based System for Ontology Revision