User signs up and signs in with a metamask wallet.
- Login Signup with wallet (or via the traditional wallet)
- Gets the user consent to push the data(browsing history titles) which is converted into embeddings onto the local Web database named orbitDB in order to match with other users’ embeddings.
it will be user consent-based only being local encrypted data storage
Example data for the analysis (via user permission):
- youtube hyperlinks:
- Spotify
- Medium
- Different different personal websites most visited
- Quora question
- Reddit posts
User first gets browsing data(in embeddings) stored in the local instance of the gunDB, which then is being used by the ML model in order to find the similarity with the current onchain detail parameters, and then based on its functionality,there will be the score of the profile predicted by the model/algorthim.
- browsing history embeddings(using google sentence encoder) data is being matched with other user’s data(history) for matching with other user’s history and finding a likely friend/connection match using cosine similarity and other algorithms.
- collaborative recommendation for getting feeds(content embedded youtube videos and open sea art collection images and embedded quora questions, embedded tweets) in other users based on their match with other user browsing history(content).
Big problem/blocker: finding similarities around every embedding with other on-chain user’s embeddings, is a highly compute-intensive process, so been thinking and searching for better approach.
And getting the right embeddings for the whole sentence and [individual word by word] in a title of the browsing history of the user.