How to parse and add video's link to DialogFlow

[Shiko]

Dear all and @xavidop 

I would really need for your help please.

I'm currently using Google's "Vertex AI Search & Conversational agent"(DialogFlow CX) to create an agent for chatting with my data. The challenge I'm facing is that my data consists of videos in MP4 format. I've extracted all the transcripts and saved them in TXT files with the following structure:

• Video link: [link]
• Video Title: [Title]
• Video Transcription: [Text]

I've uploaded all of this to Google Storage and created an app to train on this storage. The agent successfully generates answers from the text content and includes citations. However, the problem is that the citation currently refers to the link of the file on the storage instead of the video link, which is already present in the file.

I'm unsure about how to resolve this issue or even is this the correct approach (Let me know if there is a better solution), and I find myself stuck at this point. Any guidance or suggestions you could provide would be greatly appreciated.

[xavidop]

yes, that is correct, it will link you to the file name.

What I would suggest is to name the files with an ID so then you can retrieve all the information you need from youtube.

Best,

Xavi

[Shiko]

Thanks @xavidop for your quick answer.

Could I ask what did you mean by "name the files with an ID so then you can retrieve all the information you need"?

How this will help to cite the original video link and not the file storge link?

Sorry, but I am new to Vertex AI Search & Conversational and feel lost there.

[xavidop]

Each video has an ID on youtube right?

So just name your files on your Google Cloud Storage with that ID, then you will be able to retrieve the information you are looking fro from Youtube since you have the id in the file name.

Best,

Xavi

[Shiko]

@xavidop 
the videos we have are not publicly accessible. We operate on a platform that features video thumbnails and titles. To view a video, users need to click on the thumbnail, create an account, and then be redirected to a completely different platform that hosts all the videos.

The primary goal is to assist users in finding videos that match their needs by guiding them to the video page link. From there, they can proceed with subsequent steps, such as creating an account.

In conclusion, the video link I intend to include in the citation is not a direct link to a page with a video ID. Instead, it is a standard webpage containing thumbnails and general information.

Is there a possibility of adjusting the agent logic to cite the file storage link and extract the video link from the file content to display it as a citation?

[xavidop]

gotcha,

The only thing that you can do with that is use a data store and then try to extract the information by sending a request to an LLM.

If the data is structured, it should be fine.

But again, if you have the transcript per file setting the id, you can build the URL your users need.

Xavi

[xavidop]

the file name should be the ID of the video on your platform. if your platform is private or public, it doesn't; matter. you just need the id to create the URL you want for your users