JapanPublic sector2025 – ongoingHiroshima Prefectural Government
AI extraction and clustering of public opinions
Hiroshima's AI pilot extracted 3,570 opinions from election-related posts in about 15 minutes
Problem
Public comments mixed multiple arguments with news reposts and promotional material. Staff needed to organize the text into topics for analysis.
Approach
Hiroshima AI Lab tested Kouchou AI on election-related X posts and public survey comments. It used Azure OpenAI GPT-4o for text processing and embeddings for clustering. Staff compared prompts to improve the exclusion of material that contained no opinion.
Results
Takeaway
Extraction speed did not settle classification quality: changing the prompts changed which material the system retained.
Categories
- Size
- Large enterprise
- Industry
- Public sector
- Business function
- Strategy and organizationInternal work and knowledge management
- Innovation type
- Process innovation
- AI technology
- Generative AIPredictive analytics and ML
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Company's own claim
Sources
Source summaries
The lab reported extracting 3,570 opinions from 4,924 X posts in about 15 minutes. It also tested survey comments and examined how prompt changes affected filtering and classification.
Full details are in the original sources.
