Fuzzy and semantic search for captioned YouTube videos

#1

Post by Aditya_Ramesh » Thu, Sep 05, 2019, 6:22 AM UTC

Fuzzy and semantic search for captioned YouTube videos
github.com

Re: Fuzzy and semantic search for captioned YouTube videos

#2

Post by est31 » Thu, Sep 05, 2019, 7:51 AM UTC

Very cool!

I wonder why Google, a search company, is not offering this feature themselves. At least it got much nicer to search now with them offering transcripts of every video (you can simply ctrl+f with the transcript open) but it's still not as nice as it could be.

Re: Fuzzy and semantic search for captioned YouTube videos

#3

Post by ryacko » Thu, Sep 05, 2019, 8:51 AM UTC

est31 wrote: post #2

Very cool! I wonder why Google, a search company, is not offering this feature themselves. At least it got much nicer to search now with them offering transcripts of every video (you can simply ctrl+f with the transcript open) but it's still not as nice as it could be.

Corporations don’t like advertising imperfection, partially because many people have the wrong expectations (except for the cases where you have already purchased the product).

Re: Fuzzy and semantic search for captioned YouTube videos

#4

Post by codetrotter » Thu, Sep 05, 2019, 9:16 AM UTC

This is nice work. Would you consider adding an open source license to the root of the repository, so that you release the code under the terms of said license?

If so then I suggest using the ISC license. It is short, simple and very permissive.

Re: Fuzzy and semantic search for captioned YouTube videos

#5

Post by solarkraft » Thu, Sep 05, 2019, 10:15 AM UTC

What, a good search feature for YouTube? Amazing! Long term I'd like to have an independent YouTube index which this could be a part of, because in general YouTube's search features are absolute garbage (again amazing for a search company).

Are there alternative front-ends that accomplish this?

Re: Fuzzy and semantic search for captioned YouTube videos

#6

Post by Aditya_Ramesh » Thu, Sep 05, 2019, 10:33 AM UTC

codetrotter wrote: post #4

This is nice work. Would you consider adding an open source license to the root of the repository, so that you release the code under the terms of said license? If so then I suggest using the ISC license. It is short, simple and very permissive.

That sounds good to me - I'll do that as soon as possible.

And thanks for the nice words!

Re: Fuzzy and semantic search for captioned YouTube videos

#7

Post by priansh » Thu, Sep 05, 2019, 10:46 AM UTC

I've been looking for something like this for a while! It's awesome to be able to search videos like this.

Any hope for a Chrome extension?

Re: Fuzzy and semantic search for captioned YouTube videos

#8

Post by BiasRegularizer » Thu, Sep 05, 2019, 12:53 PM UTC

Nice job OP. I noticed that you used word2vec for the embedding space, do you think more modern ones like seq2seq or BERT might give better performance?

Re: Fuzzy and semantic search for captioned YouTube videos

#9

Post by Aditya_Ramesh » Thu, Sep 05, 2019, 1:09 PM UTC

BiasRegularizer wrote: post #8

Nice job OP. I noticed that you used word2vec for the embedding space, do you think more modern ones like seq2seq or BERT might give better performance?

Yeah, I'm currently looking into them - I'm new to the field and exploring existing solutions to see what works best and what's also computationally feasible since the end user would mostly have limited access to powerful hardware.

Re: Fuzzy and semantic search for captioned YouTube videos

#10

Post by therein » Thu, Sep 05, 2019, 9:39 PM UTC

This is great. I wish YouTube indexed this officially.

Also another search problem I have is with GitHub. It is terrible, at the very least it needs to expose a flag to do fork deduplication.