Author name cluster

Lijie Wen

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

1 author row

EAAI Journal 2026 Journal Article

AGRENET: Attributed graph rotation embedding network for clustering

Ying Xie
Junnan Shen
Zhiqiang Xu
Jixiang Wang
Lijie Wen
Rongbin Xu
Yun Yang

Recent studies leverage graph neural networks (GNNs) to learn node embeddings, subsequently employing conventional clustering techniques to identify clusters. However, these methods face several challenges. Firstly, the use of the original graph structure is not ideal for clustering as it is often plagued by noise and sparsity issues. Additionally, non-clustering driven losses are typically used, which might fail to represent the overall cluster structure. Consequently, the generated embeddings are often insufficient for the subsequent clustering task. To tackle these challenges, we introduce an innovative framework, the attributed graph rotation embedding network for clustering (AGRENET), which leverages full graph information based on the proposed dynamic global attention to improve graph structure. The proposed framework learns node embeddings through a spectral rotation embedding loss that integrates both feature and structure information into a kernel order learning via a higher-order graph convolution. Combining graph structure improvement, embedding network construction, and kernel order learning, AGRENET allows the improved graph structure to encode third-order proximities, thus reducing noise and sparsity issues. Experiments on four real-world and two synthetic attributed graph datasets demonstrate that AGRENET consistently outperforms state-of-the-art methods. Specifically, it achieves an average improvement of 11. 5% in Accuracy, 17. 3% in Normalized Mutual Information, 17. 3% in Adjusted Rand Index, and 13. 8% in F1-score over the best baseline. The results of experiments conducted on six benchmark graphs show that AGRENET surpasses cutting-edge methods with respect to clustering performance. The code is available at: https: //github. com/YingXie/AGRENET.

Details DOI

AAAI Conference 2026 Conference Paper

OmniDPO: A Preference Optimization Framework to Address Omni-Modal Hallucination

Junzhe Chen
TIANSHU ZHANG
Shiyu Huang
Yuwei Niu
Chao Sun
Rongzhou Zhang
Guanyu Zhou
Lijie Wen

Recently, Omni-modal large language models (OLLMs) have sparked a new wave of research, achieving impressive results in tasks such as audio-video understanding and real-time environment perception. However, hallucination issues still persist. Similar to the bimodal setting, the priors from the text modality tend to dominate, leading OLLMs to rely more heavily on textual cues while neglecting visual and audio information. In addition, fully multimodal scenarios introduce new challenges. Most existing models align visual or auditory modalities with text independently during training, while ignoring the intrinsic correlations between video and its corresponding audio. This oversight results in hallucinations when reasoning requires interpreting hidden audio cues embedded in video content. To address these challenges, we propose OmniDPO, a preference-alignment framework designed to mitigate hallucinations in OLLMs. Specifically, OmniDPO incorporates two strategies: (1) constructing text-preference sample pairs to enhance the model’s understanding of audio-video interactions; and (2) constructing multimodal-preference sample pairs to strengthen the model’s attention to visual and auditory information. By tackling both challenges, OmniDPO effectively improves multimodal grounding and reduces hallucination. Experiments conducted on two OLLMs demonstrate that OmniDPO not only effectively mitigates multimodal hallucinations but also significantly enhances the models' reasoning capabilities across modalities.

PDF Details DOI

AAAI Conference 2021 Conference Paper

Conceptualized and Contextualized Gaussian Embedding

Chen Qian
Fuli Feng
Lijie Wen
Tat-Seng Chua

Word embedding can represent a word as a point vector or a Gaussian distribution in high-dimensional spaces. Gaussian distribution is innately more expressive than point vector owing to the ability to additionally capture semantic uncertainties of words, and thus can express asymmetric relations among words more naturally (e. g. , animal entails cat but not the reverse). However, previous Gaussian embedders neglect inner-word conceptual knowledge and lack tailored Gaussian contextualizer, leading to inferior performance on both intrinsic (context-agnostic) and extrinsic (context-sensitive) tasks. In this paper, we first propose a novel Gaussian embedder which explicitly accounts for innerword conceptual units (sememes) to represent word semantics more precisely; during learning, we propose Gaussian Distribution Attention over Gaussian representations to adaptively aggregate multiple sememe distributions into a word distribution, which guarantees the Gaussian linear combination property. Additionally, we propose a Gaussian contextualizer to utilize outer-word contexts in a sentence, producing contextualized Gaussian representations for contextsensitive tasks. Extensive experiments on intrinsic and extrinsic tasks demonstrate the effectiveness of the proposed approach, achieving state-of-the-art performance with near 5. 00% relative improvement.

PDF Details

AAAI Conference 2020 Conference Paper

Solving Sequential Text Classification as Board-Game Playing

Chen Qian
Fuli Feng
Lijie Wen
Zhenpeng Chen
Li Lin
Yanan Zheng
Tat-Seng Chua

Sequential Text Classiﬁcation (STC) aims to classify a sequence of text fragments (e. g. , words in a sentence or sentences in a document) into a sequence of labels. In addition to the intra-fragment text contents, considering the interfragment context dependencies is also important for STC. Previous sequence labeling approaches largely generate a sequence of labels in left-to-right reading order. However, the need for context information in making decisions varies across different fragments and is not strictly organized in a left-to-right order. Therefore, it is appealing to label the fragments that need less consideration of context information ﬁrst before labeling the fragments that need more. In this paper, we propose a novel model that labels a sequence of fragments in jumping order. Speciﬁcally, we devise a dedicated boardgame to develop a correspondence between solving STC and board-game playing. By deﬁning proper game rules and devising a game state evaluator in which context clues are injected, at each round, each player is effectively pushed to ﬁnd the optimal move without position restrictions via considering the current game state, which corresponds to producing a label for an unlabeled fragment jumpily with the consideration of the contexts clues. The ﬁnal game-end state is viewed as the optimal label sequence. Extensive results on three representative datasets show that the proposed approach outperforms the state-of-the-art methods with statistical signiﬁcance.

PDF Details