Traditional methods for processing classical Chinese typically segment language understanding into discrete tasks, which overlook crucial background information and reduce user engagement.Large language models (LLMs) provide integrated solutions, yet they entail high computational costs and risks of generating inaccurate historical information.To tackle these challenges, we propose a novel framework, TEACH (conTrastive knowlEdge Adaptive distillation with enhanCed Historical interpretability), which focuses on classical Chinese understanding by integrating word sense disambiguation with sentence translation.This integration leverages a confidenceannotated knowledge base and a step-by-step Chain-of-Thought prompting mechanism to minimize hallucinations and improve semantic analysis.Moreover, TEACH employs contrastive distillation learning to efficiently transfer capabilities from larger models to smaller ones (e.g., Qwen2-1.5B),addressing overly liberal translations.Additionally, we introduce an innovative generation evaluation metric using iterative word alignment, enhancing LLM performance assessments by distinguishing additional information and addressing excessive translation issues.Experiments conducted on real-world datasets validate TEACH's efficacy in classical Chinese educational scenarios.

Graph contrastive learning plays a crucial role in many fields, such as social network analysis and recommendation, and has become a hot research area in recent years. The key to improving the quality of contrastive learning is to obtain high-quality sample pairs that reflect the structural features of the data more effectively. Existing methods typically sample based on node similarity and can only select negative samples of a fixed difficulty level, leading to issues like false positives and false negatives. This study proposes a novel paradigm called Hardness-aware Negative Sampling Graph Contrastive Learning with LLMs. We outline the criteria that the adaptive hardness sampling method should follow and provide concrete instances. By adaptively selecting negative samples of appropriate hardness during the training process, this method effectively mitigates false negative and false positive problems. Additionally, we leverage large language models to integrate data with supplementary textual attributes. Extensive experiments and analyses demonstrate the superiority and effectiveness of our proposed method.

Nowadays, with political ideologies becoming increasingly polarized, detecting political perspectives has become more and more crucial. Previous studies mainly focused on reasoning with real-world entities as background knowledge, while they fail to effectively model hierarchical relationships in the text structure and leave out the document-level knowledge such as topics in a news article. To overcome these limitations, we propose KMGN, a novel Knowledge-enhanced Multi-Granularity Interaction Network for political perspective prediction which consists of two key components: (1) a knowledge infusion method for learning both the topic representations and political entities of a news article (2) a multi-granularity interaction network to learn semantic representations for words, sentences, and document and integrate the infused knowledge across different granularities. Extensive experiments shows that our proposed approach is effective in predicting the political stance in a news article. An ablation study demonstrates the contribution of each component in our model.

Video Question Answering (VideoQA), as a challenging multimodal reasoning task, aims to answer the given question based on multimodal modeling, reasoning and interacting. Most existing studies usually extract visual and linguistic embeddings separately, then perform graph-based reasoning in a single modality at a specific level (e.g., object-level, frame-level). However, an overlooked observation is that, while video is presented in frame sequence, the visual elements are not sequential but rather hierarchical. In terms of cross-modal interaction, most of the methods also hardly perform the inter- and intra-modality reasoning simultaneously. To tackle the challenges above, we propose a Bottom-Up Hierarchical Propagation Networks with Heterogeneous Graph Modeling framework named HPN, to hierarchically learn the semantic correlations among different modalities. We first regard these multilevel features as nodes, constructing a uniform heterogeneous graph at each level to perform cross-modal interaction, these updated visual and linguistic features are then bottom-up propagated to the upper level to adjust the hierarchical nature of the video element. The experiments conducted on two VideoQA datasets show that HPN outperforms the state-of-the-art methods, achieving a significant improvement.

The surge in video and social media content underscores the need for a deeper understanding of multimedia data. Most of the existing mature video understanding techniques perform well with short formats and content that requires only shallow understanding, but do not perform well with long format videos that require deep understanding and reasoning. Deep Video Understanding (DVU) Challenge aims to push the boundaries of multimodal extraction, fusion, and analytics to address the problem of holistically analyzing long videos and extract useful knowledge to solve different types of queries. This paper introduces a query-aware method for long video localization and relation discrimination, leveraging an image-language pretrained model. This model adeptly selects frames pertinent to queries, obviating the need for a complete movie-level knowledge graph. Our approach achieved first and fourth positions for two groups of movie-level queries. Sufficient experiments and final rankings demonstrate its effectiveness and robustness.