A5101705596- Topic Modeling
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Data Quality and Management
- Intelligent Tutoring Systems and Adaptive Learning
- Software Engineering Research
- Dam Engineering and Safety
- Sentiment Analysis and Opinion Mining
- Bioinformatics and Genomic Networks
- Expert finding and Q&A systems
- Text and Document Classification Technologies
- Gas Sensing Nanomaterials and Sensors
- Service-Oriented Architecture and Web Services
- Imbalanced Data Classification Techniques
- Groundwater flow and contamination studies
- ZnO doping and properties
- Advanced Text Analysis Techniques
- Semantic Web and Ontologies
- Catalytic Processes in Materials Science
- Higher Education and Teaching Methods
- Advanced Photocatalysis Techniques
- Rough Sets and Fuzzy Logic
- Machine Learning and Algorithms
- Software System Performance and Reliability
- Mathematical Dynamics and Fractals
Hohai University
2022-2025
Central China Normal University
2016-2024
Shanghai Jiao Tong University
2023-2024
Wuhan University
1987-2024
Shanghai First People's Hospital
2022-2024
East China University of Technology
2023-2024
University of Science and Technology Liaoning
2022
China Southern Power Grid (China)
2022
York University
2020
The task of machine reading comprehension (MRC) is to enable read and understand a piece text then answer the corresponding question correctly. This requires not only be able perform semantic understanding but also possess logical reasoning capabilities. Just like human reading, it involves thinking about from two interacting perspectives semantics logic. However, previous methods based on either consider structure or cannot simultaneously balance reasoning. single form make fully meaning...
Sentiment classification aims to automatically predict sentiment polarity (e.g., positive or negative) of user generated data reviews, blogs).Due the mismatch among different domains, a classifier trained in one domain may not work well when directly applied other domains.Thus, adaptation for algorithms are highly desirable reduce discrepancy and manual labeling costs.To address above challenge, we propose novel method, called Bi-Transferring Deep Neural Networks (BTDNNs).The proposed BTDNNs...
The goal of Knowledge graph embedding (KGE) is to learn how represent the low dimensional vectors for entities and relations based on observed triples. conventional shallow models are limited their expressiveness. ConvE (Dettmers et al., 2018) takes advantage CNN improves expressive power with parameter efficient operators by increasing interactions between head relation embeddings. However, there no structural information in space ConvE, performance still number interactions. recent KBGAT...
Abstract Background The aim of this study is to build a prognostic model for cutaneous melanoma (CM) using fatty acid-related genes and evaluate its capacity predicting prognosis, identifying the tumor immune microenvironment (TIME) composition, assessing drug sensitivity. Methods Through analysis transcriptional data from TCGA-SKCM GTEx datasets, we screened differentially expressed acids-related (DEFAGs). Additionally, employed clinical GSE65904 identify associated with prognosis....
With the development of blockchain technologies, many Ponzi schemes disguise themselves under veil smart contracts. The scheme contracts cause serious financial losses, which has a bad effect on blockchain. Existing contract detection studies have mainly focused extracting hand-crafted features and training machine learning classifier to detect However, cannot capture structural semantic feature source code. Therefore, in this study, we propose method called MTCformer (Multi-channel Text...
Temporal knowledge graph embedding (TKGE) aims to learn the of entities and relations in a temporal (TKG). Although previous neural networks (GNN) based models have achieved promising results, they cannot directly capture interactions multi-facts at different timestamps. To address above limitation, we propose time-aware relational attention model (TARGAT), which takes timestamps as unified graph. First, develop generator dynamically generate series message transformation matrices, jointly...
Complex knowledge graph question answering (KGQA) aims at natural language questions by entities retrieving from a (KG). Recently, the relation path-based models have shown unique advantage for complex KGQA. However, these existing ignore dependency between different paths, which leads to aimless reasoning over KG. To resolve this issue, we propose question-directed with relation-aware attention network (QRGAT) that encodes process as graph. The GAT can recognize neighbor along corresponding...
Math Word Problem (MWP) solving is a critical task in natural language processing, has garnered significant research interest recent years. Various studies heavily rely on Seq2Seq models and their extensions (e.g., Seq2Tree Graph2Tree) to generate mathematical equations. While effective, these struggle diverse but counterpart solution equations, limiting generalization across various math problem scenarios. In this paper, we introduce novel Diversity-enhanced Knowledge Distillation (DivKD)...
Just-in-time (JIT) bug prediction is an effective quality assurance activity that identifies whether a code commit will introduce bugs into the mobile app, aiming to provide prompt feedback practitioners for priority review. Since collecting sufficient labeled data not always feasible some apps, one possible approach leverage cross-app models. In this work, we propose new cross-triplet deep feature embedding method, called CDFE, JIT task. The CDFE method incorporates state-of-the-art loss...
Cross-lingual entity alignment has attracted considerable attention in recent years. Past studies using conventional approaches to match entities share the common problem of missing important structural information beyond modeling process. This allows graph neural network models step in. Most existing model individual knowledge graphs (KGs) separately with a small amount pre-aligned served as anchors connect different KG embedding spaces. However, this characteristic can cause several major...
The graph attention network (GAT) [1] has started to become a mainstream neural architecture since 2018, yielding remarkable performance gains in various natural language processing (NLP) tasks. Although GAT reached the state-of-the-art (SOTA) as recent success relation prediction knowledge graph, current model is still limited by following two aspects: (1) existing only considers neighbors from inbound-direction of given entity, but ignores rich neighborhood information outbound-directions;...