EURO SEC 6, Sep. 7-11, 2026, Turin, Itala
Safety|구충완
A Retrieval Strategy for Supporting Reliable Judgments in Construction Arbitration
2026International Conference
Research AreaSafety
Professor구충완
AuthorsChoi, D., Seo, S., Choi, Y., Ahn, S., and Koo, C. (Corresponding Author)
Publication Year2026
Journal (Volume, Issue, Pages)EURO SEC 6, Sep. 7-11, 2026, Turin, Itala
AbstractIn construction arbitration, arbitral outcomes may vary depending on the arbitrator’s
experience and interpretation, making consistent reference to similar precedents for reference during case review. However, identifying and analyzing relevant past cases is time-consuming and cognitively demanding. This study proposes a systematic case retrieval strategy to support arbitration practice by examining how text unit design and retrieval techniques affect similarity search performance. Two types of text units— award summaries and full arbitral awards segmented into multiple chunk sizes (512, 1024, and 4096 tokens)—were combined with three retrieval methods: lexical, semantic, and hybrid search. A vector database was constructed using 80 construction arbitration cases from the Korean Commercial Arbitration Board, and a design-change dispute was used as the target query. Retrieval performance was evaluated using normalized Discounted Cumulative Gain (nDCG). Results show that semantic search applied to award summaries achieves the highest retrieval accuracy, while lexical and hybrid methods perform more robustly for chunked full awards. These findings demonstrate that retrieval effectiveness strongly depends on the interaction between text structure and search strategy. The proposed framework provides practical guidance for building reliable, efficient, and explainable case-retrieval systems for precedent identification and review in construction arbitration.
experience and interpretation, making consistent reference to similar precedents for reference during case review. However, identifying and analyzing relevant past cases is time-consuming and cognitively demanding. This study proposes a systematic case retrieval strategy to support arbitration practice by examining how text unit design and retrieval techniques affect similarity search performance. Two types of text units— award summaries and full arbitral awards segmented into multiple chunk sizes (512, 1024, and 4096 tokens)—were combined with three retrieval methods: lexical, semantic, and hybrid search. A vector database was constructed using 80 construction arbitration cases from the Korean Commercial Arbitration Board, and a design-change dispute was used as the target query. Retrieval performance was evaluated using normalized Discounted Cumulative Gain (nDCG). Results show that semantic search applied to award summaries achieves the highest retrieval accuracy, while lexical and hybrid methods perform more robustly for chunked full awards. These findings demonstrate that retrieval effectiveness strongly depends on the interaction between text structure and search strategy. The proposed framework provides practical guidance for building reliable, efficient, and explainable case-retrieval systems for precedent identification and review in construction arbitration.
Construction ArbitrationCase Similarity SearchAward SummaryLegal Information RetrievalSemantic SearchLexical SearchDecision Support System
PrevClustering-based model for identifying individual differences in vulnerability to heat strain – Using a means difference analysis of personal biometric characteristicsNextA Prediction Model for Greenhouse Gas Emissions and Patterns by Boundary Scope in the Construction Phase – Utilizing Explainable Machine Learning (XML)