using knowledge graph
“Generation and utilization of large-scale language models specific to individual companies” to the practical stage
Full-scale rollout of new AI solution “Knowledge Hub”
~ Compatible with field-specific expertise such as insurance assessment and manufacturing safety work ~

Our company has moved from practical research to the practical stage regarding ``generation and utilization of large-scale language models tailored to individual companies,'' and has created a new AI solution ``Knowledge Hub'' that combines existing AI models. )" will be fully developed.

Generative AI, which applies large-scale language models such as ChatGPT (OpenAI, USA), is currently making a worldwide breakthrough, and is attracting attention for its introduction into companies and the development of new services. Masu. At Cinnamon AI, we have been working on research and development of natural language processing that uses knowledge graphs* to achieve "higher explainability" even in highly specialized fields in AI solutions in the same field. We have now decided to move to the practical stage.

*Technology that systematically links various types of knowledge and generates a knowledge network in a graph structure

Image of “Knowledge Hub”

"Knowledge Hub" is an abbreviation for an orchestration AI model that consists of Cinnamon AI's proprietary technology: automatic knowledge extraction technology (IDP) from unstructured data, a large-scale language model, and a knowledge graph based on human-in-the-loop architecture. In the future, we will provide Cinnamon AI to a wide range of industries, including industries that require deep specialized knowledge (insurance, finance, manufacturing), such as insurance assessment and manufacturing safety and maintenance operations, where Cinnamon AI has particularly accumulated knowledge and know-how.

◇ Technologies used in Cinnamon AI's "Knowledge Hub"

●AIEfficient [knowledge graph] semi-automatic generation technology using
AI technology that systematically links the vast amount of various knowledge held by companies and generates a knowledge network using a graph structure. By creating a database of natural language such as documents and conversations, and searching this database using AI, it is possible to perform advanced searches and provide highly descriptive responses. Generating such a knowledge graph requires the analysis and analysis of a large amount of data, which requires a huge amount of human effort, but Cinnamon AI uses AI to solve the problem. We have succeeded in significantly streamlining the knowledge graph generation process.

●Technology to incorporate the generated knowledge graph into a [large-scale language model]
A technology that integrates knowledge graphs into complex large-scale language model databases. By incorporating the original knowledge graph generated for each individual company, a large-scale language model that complements high explainability is realized.

●Automatic knowledge extraction technology from unstructured data [IDP = Intelligent Document Processing ]
A technology that structures unstructured data such as text, audio, and images, and extracts and organizes meaningful information (knowledge) from it. By utilizing the knowledge of each company, we will realize functions that are even more unique and individualized.
*Cinnamon AI has particular strengths in this technology, and is a member of the global conference ICDAR2020 (International Conference on Document). Received the Best Paper Award at the Conference on Document Analysis and Recognition. The paper is here:https://arxiv.org/pdf/2106.00952.pdf

●【HITL = Human in the Loop ]Reinforcement learning of knowledge graphs with architecture
A design process that embeds humans within the framework of AI feedback loops to continuously improve performance. By having humans respond to data feedback, which AI is not good at at any given time, it becomes possible to efficiently and continuously incorporate new learning data into AI.

*Cinnamon AI HITL paper: Conference of the North American Association for Computational Linguistics:
NAACL 2022, July, 2022
https://aclanthology.org/2022.findings-naacl.147.pdf (Obtained Rank A)