Papers by Jiayu Song
An Encoding Strategy Based Word-Character LSTM for Chinese NER (N19-1)
Copied to clipboard
| Challenge: | Existing word-based model can not be trained in batches due to its DAG structure. |
| Approach: | They propose a lattice model that integrates word information into the start or end characters of a word and integrates it into a fixed-sized representation for efficient batch training. |
| Outcome: | The proposed model outperforms other state-of-the-art models on benchmark datasets and shows that it can be trained in batches without a shortcut path. |
Temporal reasoning for timeline summarisation in social media (2025.acl-long)
Copied to clipboard
| Challenge: | Existing temporal reasoning datasets focus on pair-wise event relationships. |
| Approach: | They propose a temporal reasoning dataset focused on temporal relationships among sequential events within narratives that combines temporal thinking with timeline summarisation through a knowledge distillation framework. |
| Outcome: | The proposed model achieves superior performance on mental health-related timeline summarisation tasks, highlighting the importance and generalisability of leveraging temporal reasoning to improve timeline summaries. |
MSG-LLM: A Multi-scale Interactive Framework for Graph-enhanced Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Existing graph-enhanced large language models (LLMs) focus on matching subgraphs between subgraph and candidate subgraph at the same scale, neglecting that subgraph with different scales may also share similar semantics or structures. |
| Approach: | They propose to use graph kernel search to discover subgraphs from the entire graph to bridge the graph and LLMs, helping with graph retrieval and LRM generation. |
| Outcome: | The proposed method achieves state-of-the-art on two graph-based tasks and the results are published in the journal Nature. |
Retrieving to Recover: Towards Incomplete Audio-Visual Question Answering via Semantic-consistent Purification (2026.acl-long)
Copied to clipboard
| Challenge: | Recent audio-visual question answering methods lack effective mechanisms for handling missing modalities, leading to performance degradation in real-world scenarios with data interruptions. |
| Approach: | They propose a framework that shifts the paradigm of missing modality handling to retrieval-based recovery . they leverage cross-modal retrieval via unified semantic embeddings to acquire missing domain-specific knowledge. |
| Outcome: | The proposed framework improves AVQA and enhances robustness in modal-incomplete scenarios. |
Identifying Moments of Change from Longitudinal User Text (2022.acl-long)
Copied to clipboard
| Challenge: | Identifying changes in individuals’ behaviour and mood via shared content is gaining importance given the global increase in mental health disorders and the limited access to support services. |
| Approach: | They propose a task of identifying moments of change in individuals on the basis of their shared content online. |
| Outcome: | The proposed task is based on 500 manually annotated user timelines and shows that it performs best through context aware sequential modelling. |
Revisiting Epistemic Markers in Confidence Estimation: Can Markers Accurately Reflect Large Language Models’ Uncertainty? (2025.acl-short)
Copied to clipboard
| Challenge: | Large language models (LLMs) are increasingly used in high-stakes domains, but their confidence is inconsistent in out-of-distribution scenarios. |
| Approach: | They define "marker confidence" as the observed accuracy when a model employs an epistemic marker. |
| Outcome: | The proposed model generalizes well within the same distribution, but its confidence is inconsistent in out-of-distribution scenarios. |
Unsupervised Opinion Summarisation in the Wasserstein Space (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent work on opinion summarisation of social media posts has focused on reviews . however, it is important to capture user opinions in online discussions over specific topics . |
| Approach: | They propose an unsupervised opinion summarisation model which uses the Wasserstein distance to generate a single summary from a group of documents. |
| Outcome: | The proposed model outperforms the state-of-the-art on ROUGE metrics and produces the best summaries with respect to meaning preservation according to human evaluations. |
Combining Hierachical VAEs with LLMs for clinically meaningful timeline summarisation in social media (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that social media users' posts can help identify depression, bipolar disorder or self-harm. |
| Approach: | They propose a hybrid abstractive summarisation approach combining hierarchical VAEs with LLMs to produce clinically meaningful summaries from social media timelines. |
| Outcome: | The proposed approach produces clinically meaningful summaries from social media user timelines, suitable for mental health monitoring. |
DIXITWORLD: Evaluating Multimodal Abductive Reasoning in Vision-Language Models with Multi-Agent Dixit Gameplay (2026.acl-short)
Copied to clipboard
Yunxiang MO, Tianshi Zheng, Qing Zong, Jiayu Liu, Baixuan Xu, Yauwai Yim, Chunkit Chan, Jiaxin Bai, Yangqiu Song
| Challenge: | Existing evaluations of multimodal abductive reasoning are limited to static, single-agent tasks. |
| Approach: | They propose a multiagent evaluation suite that deconstructs the current evaluations of multimodal abductive reasoning in vision–language models. |
| Outcome: | The evaluation suite is based on two core components: DixitArena and DixitsBench. |