Unique identifier for notebooks

[hectorperez]

Hi, thanks for your efforts with Livebook!

Could we introduce a unique notebook id or token? Maybe as notebook metadata.

At notes.club, currently, a change in a notebook’s title or path prompts us to delete and recreate it, losing interactions like likes and, in the future, comments. A consistent identifier would allow us to maintain these likes and maybe comments by updating the existing record instead of recreating it.

Also, this would also allow us to identify when people re-upload the content of a repo (without a normal fork). For example, this happens with DockYard Academy's notebooks and we get duplicated notebooks.

Let me know if that would be possible.
Thanks!

[jonatanklosko]

Hey @hectorperez! We could technically generate a uuid and put it in notebook metadata comment, although this would imply that all notebooks come with a comment by default, which ideally I would like to avoid. That said though, there's a more serious concern with authenticity, anyone could put an existing id in their notebook, and in turn "intercept" comments, likes etc (you can imagine a malicious notebook making it to the top of the list, because it used an id of another popular notebook). This could be an issue even in non-malicious cases, such as someone downloading tutorial notebooks, working on them, and pushing to their GitHub account.

Unfortunately, I can't think of a reliable way to achieve the linking, given the distributed nature :<

[hectorperez]

Hi @jonatanklosko! I see your concern about notebooks being duplicated or misused when copied and modified. Perhaps we could address this by not storing new notebooks that carry an existing uuid in our database, or by linking these copies to the original (oldest) without sharing likes and comments. This way, we can manage duplications without necessarily extending likes or comments to copies.

I understand that including a metadata comment in all notebooks might be inconvenient. If it's too cumbersome or if you foresee further complications, we can skip it. Thanks for considering this!

[josevalim]

I don't think this could be a unique id to identify notebooks, rather a "lineage id", to identify related things. They can be forks, clones, etc. However, I haven't considered Jonatan concerns about security and this could be used indeed for malicious purposes: for example, if there is a popular notebook, I could attach to it as related content, even if we are ultimately completely disconnected.

However, if you want to identify related things, techniques such as cosine similarity of document vectors will be much more reliable. If you want to identify renames, perhaps text distance algorithms, such as Levenshtein Distance, will be good enough (you can limit them on a per repository search/comparison).

[hectorperez]

Thanks for your comment José!

The main issue is to identify renames and copies — forks are less of an issue since GitHub clearly marks them.

Regarding text distance algorithms, are you suggestion we compare the file path and name or the content of the notebook?
For example, the file [1] was moved and renamed from nx/guides/tensor-aggregation-101.livemd to nx/nx/guides/advanced/aggregation.livemd. As the distance Levenshtein Distance is 15, not sure whether it would work well in general. Perhaps considering the content could provide a more reliable solution?

Thank you!

[1] https://notes.club/elixir-nx/nx/nx/guides/advanced/aggregation

[josevalim]

Yes, exactly, you would need to define some heuristic to compare content.

Btw, by fork, I mean an informal fork. Someone gets the aggregation notebook above, makes some changes, and then publishes it elsewhere. :)

[hectorperez]

Aha, by copies I meant informal forks too. For example, when we search metaprogramming [1] we get many informal forks from @BrooklinJazz's DockYard Academy repository.

[1] https://notes.club/search?q=metaprogramming