- Placement and Design
Unsupervised connection data learning technology
Unsupervised learning of universal characteristics hidden in the data that expresses connections between people and things.
- Developed a learning method that uses the characteristic of similarities in adjacent items in connection data (graph), such as social media.
- Using this method with data on academic papers showing the relationship between academic papers and citations from those papers, or data related to concurrent selling showing products that are easy to purchase together, we confirmed that the accuracy of data categorization increases.
Applications
- Categorization of products using purchasing data
- Predicting companies, transaction types, and alternative companies using supply chain data
- Categorization of drawings using reference relationships that link electrical circuit drawings referenced during design
Benchmarks, strengths, and track record
- Using three open data sets for which linking information was academic paper citation relationships (Wiki CS) and product concurrent sales relationships (Amazon-Photo/Computers), the accuracy of academic paper and product categorization improved in comparison to conventional methods.
| WikiCS (%) | Amazon-Photo (%) | Amazon-Computers (%) | |
|---|---|---|---|
| Conventional method | 78.06 | 92.65 | 89.07 |
| Proposed method | 79.15 | 93.07 | 89.16 |
Inquiries
Contact the Toshiba Corporate Research & Development Center
Please include the title “Toshiba AI Technology Catalog: Unsupervised connection data learning technology” 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.
