My name is Arthur, and I’m a PhD in Computer Science from TU Delft, in The Netherlands, working specifically with Natural Language Processing and Information Retrieval.
I feel that something that would improve Obsidian a LOT (or any other PKM system) is a more robust “Unliked Mentions” tab. By that, I mean some type of Machine Learning system for predicting links between notes (this is an active research area, called, you guessed it, Link prediction, but mainly focused on large graphs of entities or friends, for instance).
So, I propose developing a plugin that would, given the current open note, improve upon the current “Unlinked Mention” tab by proposing other notes to be linked to the current one. The user would still have to actively accept the suggestion and add the context, of course.
To do so, the first thing we need is a large enough dataset of linked notes, where experts (i.e. Obsidian users) have a large number of linked notes in context, so that an algorithm (probably a Transformer, BERT-Like model, for those who like that) could process the semantics of the notes and learn if two notes should be linked or not.
For that, we need data donators. That means people who already have a reasonably developed (or maybe not so much developed) vault, willing to donate their vault data for this cause.
So, my question is: Anyone willing to do so? Or someone has any idea if such a dataset is already available?
Naturally, we will need a few people to help with that. So, feel free to jump in!