This is a prototype version of the Innovation Map based on Tech News. Tech News retrieved from ChatGPT is summarized using ChatGPT, Embedding Vectors are obtained from the text, and the SOM is ordered by reducing it to 10 dimensions using PCA.
The fine print label is the title of each article. If there is a large amount of data, there is a limit to displaying it as labels on the map, but in Viscovery SOMine you can display it in the data recodes window. You can see that each cluster corresponds to a different field. Bold labels indicate each field.
You can see in the attribute picture corresponding to PC that each principal component strongly responds to a specific cluster.
For reference, we also show an image of the map ordered by the Embedding Vector without dimension reduction.
As we accumulate data and create maps with more articles, we can expect to see interactions between different technical areas.
However, there are currently limitations to accessing news articles via ChatGPT, and the problem is that the freshness of news differs depending on the field. In the future, we plan to experiment with different AIs to find the best way to map news articles.
Please refer the information about membeship service of Mindwaer Innovation Maps.