Students achieve outstanding results in Web Intelligence Classification Challenge
Team wins prizes for accuracy and innovation for their novel framework blending hybrid lexical‑semantic retrieval and LLM‑based ensembling to classify multilingual EU job ads, a task crucial for producing meaningful statistics on the labour market.
Two UCD CS PhD candidates from UCD SFI ML-Labs, UCD ASEADOS Research group, Hong-Hanh Nguyen-Le (under the supervision of Assoc. Prof. Nhien-An Le-Khac) and Quang-Tien Tran (under the supervision of Prof. Michela Bertolotto), collaborated with other PhD candidates from Trinity College Dublin and Dublin City University to achieve outstanding results in the prestigious (opens in a new window)European Statistics Awards, winning two prizes in the Web Intelligence Classification Challenge.
The team, named FVNWL, earned an impressive 2nd place in the Accuracy Award (€5,000) and 3rd place in the Innovativity Award (€1,000). Their success came in a highly competitive field, with participants from companies, research institutions, and public bodies across Europe.
The challenge, organized by Eurostat (the statistical office of the European Union), challenged participants to develop groundbreaking automated systems for classifying multilingual online job advertisements (OJAs) from across the EU. The goal was to assign standardized occupational labels to a vast dataset, a task crucial for producing meaningful statistics on the labour market.
The FVNWL team’s winning submission was a novel framework titled "Generation-Assisted Retrieval for Job Classification". Their innovative method synergistically combined lexical-semantic retrieval models with the power of Large Language Models (LLMs). The approach introduced a unique "Hybrid Lexical-Semantic Retrieval" module to effectively capture the relationships between job descriptions and occupational categories. A key innovation was their "Digit-wise Hierarchical Ensembling Technique," which intelligently aggregated predictions from multiple predictions to enhance accuracy, tailored specifically for the hierarchical structure of the job codes. This sophisticated system was designed not only for precision but also for interpretability, allowing human operators to understand the reasoning behind each classification.
This accomplishment highlights the cutting-edge AI research and collaborative spirit fostered at UCD's School of Computer Science. Congratulations to Hong-Hanh, Quang-Tien, and the entire FVNWL team on their fantastic achievement!
ASEADOS (opens in a new window)website
8 September 2025