Personalised Tag Recommendation for Financial & Education News Search
3,000 paragraphs annotated across core, keyword, location and sentiment tags — production-ready training data for semantic search
A knowledge-management startup partnered with Tictag under the Data Voucher program to build a curated sentence-to-tag dataset spanning Singapore financial and education news, unlocking semantic search and personalised tag recommendation.
Problem
The volume of unstructured text in finance and education has exploded alongside the rise of Large Language Models. Traditional keyword-based search fails to capture semantic meaning, leaving valuable content underutilised. Our client needed a way to help machines interpret meaning and extract standardised keywords, topics, entities, and sentiments from news and document data.
Solution
Tictag executed an end-to-end data pipeline under the Data Voucher program. We collected 3,000 paragraphs (≥300 characters each) from English-language financial and educational news focused on Singapore, then annotated each across four tag categories: Core Tags (overall topic), Keyword Tags (people, institutions, brands), Location Tags (geographic references), and Sentiment Tags (mood and trend signals). Every annotation was reviewed for consistency, converted to standardised metadata, and validated by domain experts.
Results
- A curated, production-ready sentence-to-tag dataset from Singapore's news corpus
- Structured knowledge graphs that transform unstructured text into metadata networks
- Improved semantic understanding and search efficiency for the client's downstream AI models