- Language media analysis
- Knowledge organization
Document understanding AI regarding infrastructure that leverages the knowledge of experts in advanced maintenance
Advanced understanding of specialized documents in the infrastructure field contributes to rapid and efficient infrastructure maintenance.
- While efficiently learning general-purpose language from a large-scale general-purpose language model, also learn specialized language from fields such as infrastructure, through a separate curriculum using minimal specialized documents.
- Passing on knowledge from a supervised model enables learning from minimal text and adjustments to a model with a more compact scale, so the system can be used in environments with minimal computation resources.
Applications
- Similar trouble search services
- Trouble trend analysis services
Benchmarks, strengths, and track record
- Through natural language processing tasks that extract trouble expressions from trouble reports, confirmed that the system demonstrates higher performance than language models (Japanese BERT) learned from large-scale Japanese documents.
Tag extraction performance (%) | Learning time (h) | No. of parameters (M) | |
---|---|---|---|
Conventional method | 61.9 | 21.4 | 111.2 |
Proposed method | 67.1 | 5 | 70.4 |
Inquiries
Contact the Toshiba Corporate Research & Development Center
Please include the title “Toshiba AI Technology Catalog: Document understanding AI regarding infrastructure that leverages the knowledge of experts in advanced maintenance” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
References:
- Association for Natural Language Processing 29th Annual Meeting: ExDistilBERT: Domain-specific language model based on model distillation enabling dictionary expansion (in Japanese) (PDF)
- Press release: Development of infrastructure document understanding AI that leverages the knowledge of experts in advanced maintenance