A5061277006- Sentiment Analysis and Opinion Mining
- Topic Modeling
- Advanced Text Analysis Techniques
- Text and Document Classification Technologies
- Emotion and Mood Recognition
- Computational Fluid Dynamics and Aerodynamics
- Natural Language Processing Techniques
- Speech and dialogue systems
- Fluid Dynamics and Turbulent Flows
- Domain Adaptation and Few-Shot Learning
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Social Robot Interaction and HRI
- Aerospace Engineering and Energy Systems
- Mechanical stress and fatigue analysis
- Robotics and Automated Systems
- Vibration and Dynamic Analysis
- Particle Dynamics in Fluid Flows
- Meteorological Phenomena and Simulations
- Biomimetic flight and propulsion mechanisms
- Generative Adversarial Networks and Image Synthesis
- Speech and Audio Processing
- Aerospace and Aviation Technology
- HVDC Systems and Fault Protection
Harbin Institute of Technology
2017-2024
Tencent (China)
2023
National University of Defense Technology
2009-2020
Anhui University of Technology
2017
Multimodal fusion is a core problem for multimodal sentiment analysis.Previous works usually treat all three modal features equally and implicitly explore the interactions between different modalities.In this paper, we break kind of methods in two ways.Firstly, observe that textual modality plays most important role analysis, can be seen from previous works.Secondly, comparing to modality, other kinds nontextual modalities (visual acoustic) provide semantics, shared private semantics.The...
Sarcasm is a sophisticated linguistic phenomenon and commonly manifests on social media platforms, which poses great challenge for opinion mining systems. Therefore, multimodal sarcasm detection, aims to understand the implied sentiment in video, has gained more attention. However, previous works mostly focus feature fusion without explicitly modeling incongruity between modalities, such as expressing verbal compliments while rolling eyes, an obvious cue detecting sarcasm. In this article,...
Emotion recognition in conversations (ERC) has received much attention recently the natural language processing community. Considering that emotions of utterances are interactive, previous works usually implicitly model emotion interaction between by modeling dialogue context, but misleading information from context often interferes with interaction. We noticed gold labels can provide explicit and accurate interaction, it is impossible to input at inference time. To address this problem, we...
Multimodal sentiment analysis has attracted increasing attention and lots of models have been proposed. However, the performance state-of-the-art decreases sharply when they are deployed in real world. We find that main reason is real-world applications can only access text outputs by automatic speech recognition (ASR) models, which may be with errors because limitation model capacity. Through further ASR outputs, we some cases words, key elements textual modality, recognized as other makes...
Multi-modal emotion recognition aims to recognize states from multi-modal inputs. Existing end-to-end models typically fuse the uni-modal representations in last layers without leveraging interactions among intermediate representations. In this paper, we propose Recurrent Intermediate-Layer Aggregation (RILA) model explore effectiveness of deep pre-trained transformers for recognition. At heart our is Intermediate-Representation Fusion Module (IRFM), which consists aggregation gating module...
As a part of the power system, stability enclosed isolated-phase bus bar (EIPB) plays an important role in working safety system. Usually, dynamic-thermal EIPB is analyzed by theoretical analysis or finite-element method (FEM). The dynamic can be calculated 2-D model. For complex structure and huge size EIPB, large synchronous generator-side system rarely analyzed. In this paper, new co-simulation studied. subsegments model based on FEM established, temperature distribution joint surfaces...
The event recognition is an important foundation in the extraction task. This paper uses Chinese Emergencies Corpus (CEC) as raw corpus, and vector representation of text obtained by embedding a pre-trained language model ALBERT(A Lite Bidirectional Encoder Representation from Transformers). Then BiLSTM-CRF (Bidirectional Long Short Term Memory-Conditional Random Field) used to extract features label-dependent features, finally outputs predicted label sequence realize boundary type trigger...
Addressing domain shifts for complex perception tasks in autonomous driving has long been a challenging problem. In this paper, we show that existing adaptation methods pay little attention to the content mismatch issue between source and target domains, thus weakening per-formance decoupling of domain-invariant domain-specific representations. To solve aforementioned problems, propose an image-level framework aims at adapting source-domain images with content-aligned source-target image...
Half-Rotating Wing (HRW) is a new power wing which had been developed by our work team using rotating-type flapping instead of oscillating-type flapping. Air Vehicle (HRWAV) similar as Bionic Flapping (BFWAV). It necessary to guarantee pitching stability HRWAV maintain flight stability. The working principle HRW was firstly introduced in this paper. rule motion indicated that the fuselage without empennage would overturn forward it generated increased movement. Therefore, added on tail...