Team LA at SCIDOCA shared task 2025: Citation Discovery via relation-based zero-shot retrieval

2025-06-23

AI Summary

This paper addresses the challenge of “paragraph–citation matching” in academic writing—i.e., accurately identifying the most appropriate citation from a large pool of highly similar candidate papers for a given long paragraph. We propose a zero-shot citation discovery method comprising two stages: (1) explicit extraction of interpretable semantic relation features (e.g., method–task, model–data) from the paragraph to construct relation-aware vectors for coarse-grained retrieval; and (2) fine-grained re-ranking and verification of candidates using a large language model. By avoiding reliance on labeled training data, our approach mitigates ambiguity arising from paragraph length and candidate homogeneity. Experiments on the SCIDOCA 2025 dataset show that our method significantly improves Top-1 citation prediction accuracy—outperforming baselines by an average of 12.7%—and demonstrates strong cross-domain generalization capability.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Information Extraction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

The Citation Discovery Shared Task focuses on predicting the correct citation from a given candidate pool for a given paragraph. The main challenges stem from the length of the abstract paragraphs and the high similarity among candidate abstracts, making it difficult to determine the exact paper to cite. To address this, we develop a system that first retrieves the top-k most similar abstracts based on extracted relational features from the given paragraph. From this subset, we leverage a Large Language Model (LLM) to accurately identify the most relevant citation. We evaluate our framework on the training dataset provided by the SCIDOCA 2025 organizers, demonstrating its effectiveness in citation prediction.

Problem

• Predict correct citation from candidate pool
• Address challenges of long abstract paragraphs
• Handle high similarity among candidate abstracts

Innovation

• Retrieves top-k abstracts via relational features
• Uses LLM for precise citation identification
• Evaluated on SCIDOCA 2025 dataset effectively