Mingwei Tang
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
Mingwei Tang has a total of 81 co-authors. Here are the top co-authors by number of shared publications:
- Jie Hu
School of Computing and Artificial Intelligence, Southwest Jiaotong University, Chengdu, China
10 shared publications Mingfeng Zhao
10 shared publicationsJie Hu
School of Computing and Artificial Intelligence, Southwest Jiaotong University, Chengdu, China
8 shared publicationsLiansong Zong
Xihua University
4 shared publicationsLinxi Li
4 shared publicationsYanxi Zheng
School of Computer and Software Engineering, Xihua University, Chengdu, China
4 shared publications
Latest Publications
- Open Access
- 13 October 2026
- Published by Elsevier in SSRN Electronic Journal
Reliability-Modulated Text Anchoring with Prompt Memory for Multimodal Sentiment Analysis with Missing Modalities
Missing modalities remove affective evidence and make modality reliability sample-dependent in multimodal sentiment analysis (MSA). Reconstruction, distillation, prompt-based, and state-space approaches improve robustness, but evidence assessment, prior selection, and fusion remain insufficiently coordinated. We propose ACP-Mamba, a reliability-first retrieval-and-fusion framework connecting these decisions. First, text-conditioned interaction prepares aligned evidence, and modality-wise reliability estimation and competitive routing determine each modality's contribution to retrieval. Second, an asymmetric prompt memory matches the incomplete-view query through semantic and reliability-derived missingness-profile compatibility. Retrieved complete-view modality values support a sample-specific semantic anchor. Third, coupled prompt Mamba fusion conditions state propagation on the anchor, prompts, modality reliability, retrieval confidence, and token-level observation status, followed by hierarchical multimodal fusion and adaptive readout. Experiments cover random missing rates from 0 to 0.9. On MOSI, ACP-Mamba achieves Acc-7 and Acc-5 of 34.14% and 39.01%, respectively. On MOSEI, ACP-Mamba leads the reported classification and regression metrics, reaching an Acc-7 of 47.22% and an Acc-5 of 48.59%, together with an MAE of 0.658 and a correlation of 0.601. On SIMS, it obtains the best Acc-5, Acc-3, and correlation.
- 13 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessReliability-Modulated Text Anchoring with Prompt Memory for Multimodal Sentiment Analysis with Missing Modalities
Missing modalities remove affective evidence and make modality reliability sample-dependent in multimodal sentiment analysis (MSA). Reconstruction, distillation, prompt-based, and state-space approaches improve robustness, but evidence assessment, prior selection, and fusion remain insufficiently coordinated. We propose ACP-Mamba, a reliability-first retrieval-and-fusion framework connecting these decisions. First, text-conditioned interaction prepares aligned evidence, and modality-wise reliability estimation and competitive routing determine each modality's contribution to retrieval. Second, an asymmetric prompt memory matches the incomplete-view query through semantic and reliability-derived missingness-profile compatibility. Retrieved complete-view modality values support a sample-specific semantic anchor. Third, coupled prompt Mamba fusion conditions state propagation on the anchor, prompts, modality reliability, retrieval confidence, and token-level observation status, followed by hierarchical multimodal fusion and adaptive readout. Experiments cover random missing rates from 0 to 0.9. On MOSI, ACP-Mamba achieves Acc-7 and Acc-5 of 34.14% and 39.01%, respectively. On MOSEI, ACP-Mamba leads the reported classification and regression metrics, reaching an Acc-7 of 47.22% and an Acc-5 of 48.59%, together with an MAE of 0.658 and a correlation of 0.601. On SIMS, it obtains the best Acc-5, Acc-3, and correlation.
- Open Access
- 24 September 2026
- Published by Elsevier in SSRN Electronic Journal
Position-Sensitive Evidence Retrieval and Bounded Residual Updating for Locally Refined Multimodal Sentiment Analysis
Multimodal sentiment analysis relies on complementary linguistic, acoustic, and visual cues. However, acquisition and transmission errors can cause observations to be missing at arbitrary positions. Existing methods typically reconstruct representations of the entire modality, temporal segment, or global representation, which not only ignore the semantic differences between missing positions but may also overwrite important features already preserved in the base model. To address these issues, we propose MiARC, a locally refined multimodal sentiment analysis framework built on Position-Sensitive Evidence Retrieval and Dynamic Bounded Residual Updating. First, Position-Sensitive Evidence Retrieval maps the exact observation masks into latent space and retrieves current-sample evidence for each corrupted position, conditioned on the ordinal affective context estimated from visible modalities. Second, Dynamic Bounded Residual Updating leverages missing strength and a global gain cap to refine corrupted positions while preserving unaffected semantic features. Third, during model training, Same-Mask Cross-Modal Evidence Contrast constructs paired views that share identical language inputs, sentiment labels, and missing masks, encouraging reliance on sample-matched non-textual evidence.Extensive experiments on three benchmark datasets demonstrate that MiARC achieves state-of-the-art results in nearly all cases.In particular, across ten missing rates on CMU-MOSI, CMU-MOSEI, and CH-SIMS, MiARC improves 21 of 22 metrics over RECAP under the same protocol.
- 24 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPosition-Sensitive Evidence Retrieval and Bounded Residual Updating for Locally Refined Multimodal Sentiment Analysis
Multimodal sentiment analysis relies on complementary linguistic, acoustic, and visual cues. However, acquisition and transmission errors can cause observations to be missing at arbitrary positions. Existing methods typically reconstruct representations of the entire modality, temporal segment, or global representation, which not only ignore the semantic differences between missing positions but may also overwrite important features already preserved in the base model. To address these issues, we propose MiARC, a locally refined multimodal sentiment analysis framework built on Position-Sensitive Evidence Retrieval and Dynamic Bounded Residual Updating. First, Position-Sensitive Evidence Retrieval maps the exact observation masks into latent space and retrieves current-sample evidence for each corrupted position, conditioned on the ordinal affective context estimated from visible modalities. Second, Dynamic Bounded Residual Updating leverages missing strength and a global gain cap to refine corrupted positions while preserving unaffected semantic features. Third, during model training, Same-Mask Cross-Modal Evidence Contrast constructs paired views that share identical language inputs, sentiment labels, and missing masks, encouraging reliance on sample-matched non-textual evidence.Extensive experiments on three benchmark datasets demonstrate that MiARC achieves state-of-the-art results in nearly all cases.In particular, across ten missing rates on CMU-MOSI, CMU-MOSEI, and CH-SIMS, MiARC improves 21 of 22 metrics over RECAP under the same protocol.
- 20 August 2026
- Published by Elsevier in Neural Networks
- Vol. 205 (Pt B) , 109535
Improving cross-architecture black-box adversarial transferability via hybrid structure-aware feature perturbations
- 20 August 2026
- Published by Elsevier in Neural Networks
- Vol. 205 (Pt B) , 109535
Improving cross-architecture black-box adversarial transferability via hybrid structure-aware feature perturbations
- Open Access
- 17 July 2026
- Published by Elsevier in SSRN Electronic Journal
Reliability-Calibrated Text Anchoring with Prompt Memory for Robust Multimodal Sentiment Analysis
Missing-modality multimodal sentiment analysis (MSA) is challenging because affective evidence is removed and modality reliability becomes sample-dependent. Text often carries explicit sentiment semantics, but corrupted transcripts should not remain fixed controllers when audio and vision cues are more reliable. Existing reconstruction, distillation, prompt-compensation, and state-space methods improve robustness, yet they often rely on current incomplete observations or retain fixed language dominance. In this paper, we propose ACP‑Mamba, a robust fusion framework that transforms static text dominance into reliability‑calibrated text anchoring. First, a reconstruction‑regularized text‑conditioned encoder is proposed to estimate dynamic reliability scores for text, vision, and audio on a per‑sample basis. Second, a reliability‑guided prompt anchoring module is designed to retrieve clean cross‑modal prompt priors from a memory bank according to both semantic similarity and missing‑pattern compatibility. Third, a coupled prompt Mamba fusion mechanism is designed to conditionalize selective state updates based on the calibrated text anchor, reliability scores, and observation masks, enabling placeholder‑aware state propagation. Under random missing rates from 0 to 0.9, ACP-Mamba improves MOSI Acc-7 and Acc-5 over TF-Mamba by 2.39 and 3.41 percentage points, improves MOSEI MAE and Corr from 0.689 and 0.562 to 0.658 and 0.602, and obtains the best SIMS Acc-5, Acc-3, and correlation.
- 17 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessReliability-Calibrated Text Anchoring with Prompt Memory for Robust Multimodal Sentiment Analysis
Missing-modality multimodal sentiment analysis (MSA) is challenging because affective evidence is removed and modality reliability becomes sample-dependent. Text often carries explicit sentiment semantics, but corrupted transcripts should not remain fixed controllers when audio and vision cues are more reliable. Existing reconstruction, distillation, prompt-compensation, and state-space methods improve robustness, yet they often rely on current incomplete observations or retain fixed language dominance. In this paper, we propose ACP‑Mamba, a robust fusion framework that transforms static text dominance into reliability‑calibrated text anchoring. First, a reconstruction‑regularized text‑conditioned encoder is proposed to estimate dynamic reliability scores for text, vision, and audio on a per‑sample basis. Second, a reliability‑guided prompt anchoring module is designed to retrieve clean cross‑modal prompt priors from a memory bank according to both semantic similarity and missing‑pattern compatibility. Third, a coupled prompt Mamba fusion mechanism is designed to conditionalize selective state updates based on the calibrated text anchor, reliability scores, and observation masks, enabling placeholder‑aware state propagation. Under random missing rates from 0 to 0.9, ACP-Mamba improves MOSI Acc-7 and Acc-5 over TF-Mamba by 2.39 and 3.41 percentage points, improves MOSEI MAE and Corr from 0.689 and 0.562 to 0.658 and 0.602, and obtains the best SIMS Acc-5, Acc-3, and correlation.
- 11 July 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (6) , 402
BiCTM: lightweight video recognition via convolutional tube masking and bidirectional motion features
- 11 July 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (6) , 402
BiCTM: lightweight video recognition via convolutional tube masking and bidirectional motion features
- 12 February 2026
- Published by Wiley in Expert Systems
- Vol. 43 (3)
An Enhanced Graph Attention Aggregation Network With Multi‐Level Syntactic and Semantic Prompts for Aspect‐Based Sentiment Triple Extraction
Aspect‐Based Sentiment Triplet Extraction (ASTE) is one of the hot topics in recent years. Relevant researchers have proposed many neural network models for aspect‐based sentiment triplet extraction. However, they fail to model the complex syntactic and semantic associations between tokens, which limits the synergy of features from different angles, which may lead to inaccurate encoding of the relationship between tokens and thus inaccurate extraction of triples. In response to the problems mentioned above, a graph attention aggregation network with multi‐level Syntactic and Semantic Enhanced Prompts for aspect‐Based sentiment triple extraction (SSEP) is proposed. First, a graph attention relation aggregation module is designed in the context encoding part. Specifically, the module first constructs a relation aggregation graph through the output of the pre‐trained language model, then designs a graph attention aggregator, and finally aggregates the multi‐level output of the pre‐trained language model according to the relation aggregation graph and the graph attention aggregator. Second, a syntactic and semantic enhanced prompt module is proposed. The module uses relation table attention to prompt the model, and then uses a dual‐channel graph neural network to further enhance syntactic and semantic information and prune unimportant information. In addition, a joint boundary detection module is designed. The module can directly extract sentiment triplets using joint boundary detection labels. Finally, experimental results on four public datasets show that SSEP achieves state‐of‐the‐art performance and outperforms other models.
- 12 February 2026
- Published by Wiley in Expert Systems
- Vol. 43 (3)
An Enhanced Graph Attention Aggregation Network With Multi‐Level Syntactic and Semantic Prompts for Aspect‐Based Sentiment Triple Extraction
Aspect‐Based Sentiment Triplet Extraction (ASTE) is one of the hot topics in recent years. Relevant researchers have proposed many neural network models for aspect‐based sentiment triplet extraction. However, they fail to model the complex syntactic and semantic associations between tokens, which limits the synergy of features from different angles, which may lead to inaccurate encoding of the relationship between tokens and thus inaccurate extraction of triples. In response to the problems mentioned above, a graph attention aggregation network with multi‐level Syntactic and Semantic Enhanced Prompts for aspect‐Based sentiment triple extraction (SSEP) is proposed. First, a graph attention relation aggregation module is designed in the context encoding part. Specifically, the module first constructs a relation aggregation graph through the output of the pre‐trained language model, then designs a graph attention aggregator, and finally aggregates the multi‐level output of the pre‐trained language model according to the relation aggregation graph and the graph attention aggregator. Second, a syntactic and semantic enhanced prompt module is proposed. The module uses relation table attention to prompt the model, and then uses a dual‐channel graph neural network to further enhance syntactic and semantic information and prune unimportant information. In addition, a joint boundary detection module is designed. The module can directly extract sentiment triplets using joint boundary detection labels. Finally, experimental results on four public datasets show that SSEP achieves state‐of‐the‐art performance and outperforms other models.
- 15 January 2026
- Published by Springer Nature in Applied Intelligence
- Vol. 56 (1) , 47
BSAN: bilateral synergistic aggregation network for aspect-based sentiment analysis
- 15 January 2026
- Published by Springer Nature in Applied Intelligence
- Vol. 56 (1) , 47
BSAN: bilateral synergistic aggregation network for aspect-based sentiment analysis
- 12 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Knowledge graph-guided and dependency distance enhanced multi-graph networks for aspect-based sentiment analysis
- 12 November 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Knowledge graph-guided and dependency distance enhanced multi-graph networks for aspect-based sentiment analysis
- Open Access
- 29 October 2025
- Published by MDPI AG in Electronics
- Vol. 14 (21) , 4229
Progressive Prompt Generative Graph Convolutional Network for Aspect-Based Sentiment Quadruple Prediction
Aspect-based sentiment quadruple prediction has important application value in the current information age. There are often implicit expressions and multi-level semantic relationships in sentences, making accurate prediction for existing methods still a complex and challenging task. To address the above problems, this paper proposes the Progressive Prompt-Driven Generative Graph Convolutional Network for Aspect-Based Sentiment Quadruple Prediction (ProPGCN). Firstly, a progressive prompt module is proposed. The module uses progressive prompt templates to generate paradigm expressions of corresponding orders and introduces third-order element prompt templates to associate high-order semantics in sentences, providing a bridge for modeling the final global semantics. Secondly, a graph convolutional relation-enhanced reasoning module is designed, which can make full use of contextual dependency information to enhance the recognition of implicit aspects and implicit opinions. In addition, a graph convolutional aggregation strategy is constructed. The strategy uses graph convolutional networks to aggregate adjacent node information and correct conflicting implicit logical relationships. Finally, experimental results show that the ProPGCN model can achieve state-of-the-art performance. Specifically, our ProPGCN model achieves overall F1 scores of 65.04% and 47.89% on the Restaurant and Laptop datasets, respectively, which represent improvements of +0.83% and +0.61% over the previous strongest generative baseline.
- 29 October 2025
- Published by MDPI AG in Electronics
- Vol. 14 (21) , 4229
Open AccessProgressive Prompt Generative Graph Convolutional Network for Aspect-Based Sentiment Quadruple Prediction
Aspect-based sentiment quadruple prediction has important application value in the current information age. There are often implicit expressions and multi-level semantic relationships in sentences, making accurate prediction for existing methods still a complex and challenging task. To address the above problems, this paper proposes the Progressive Prompt-Driven Generative Graph Convolutional Network for Aspect-Based Sentiment Quadruple Prediction (ProPGCN). Firstly, a progressive prompt module is proposed. The module uses progressive prompt templates to generate paradigm expressions of corresponding orders and introduces third-order element prompt templates to associate high-order semantics in sentences, providing a bridge for modeling the final global semantics. Secondly, a graph convolutional relation-enhanced reasoning module is designed, which can make full use of contextual dependency information to enhance the recognition of implicit aspects and implicit opinions. In addition, a graph convolutional aggregation strategy is constructed. The strategy uses graph convolutional networks to aggregate adjacent node information and correct conflicting implicit logical relationships. Finally, experimental results show that the ProPGCN model can achieve state-of-the-art performance. Specifically, our ProPGCN model achieves overall F1 scores of 65.04% and 47.89% on the Restaurant and Laptop datasets, respectively, which represent improvements of +0.83% and +0.61% over the previous strongest generative baseline.