gelantu
commented
4 weeks ago I think this is a cool idea and I also agree with what you said about youtube not recommending what I always want to see and that could be better. I can contribute to this project because I have been a long time youtube user and I also make videos on Youtube. So I think I would have insight on analytics that others who don't create on Youtube would have. I also would like to see how we can go about scraping youtube so I could possibly use that in future projects I build.
Project Abstract This document proposes a solution to a problem that YouTube has not solved. If a user dislikes the videos being recommended to them, wants to see specific kinds of videos, or if they want to see videos from a larger variety of creators, YouTube’s home page is likely to not solve any of these problems. YouTube tends to specifically recommend creators on the home screen that the user has watched previously. By utilizing ChatGPT to provide the user with an array of searches that are relevant to the user’s watch history, a user can click on one of the suggested searches to find unique videos related to their history without the need to think about what they want to watch, with a larger variety of creators to choose from. Additionally, a setting may be enabled to tell ChatGPT to instead recommend random topics, along with another setting that can tell ChatGPT to only recommend topics related to a keyword (or key phrase) that the user inputs (these settings are subject to change and may not be implemented). Essentially, this is a Chrome extension that provides YouTube with a clickable GUI near its search bar with multiple options to choose from, which would automatically insert the clicked option into the search bar (insertion subject to change).
Conceptual Design This program will utilize a generative AI tool (like ChatGPT) to generate three to five unique clickable search prompts based on the most recent titles from the history of the user’s YouTube watch history. The user will first need to allow the Chrome extension and be logged in on YouTube, which then would prompt the user to navigate to the YouTube watch history page (this is necessary for privacy reasons). After the navigation, a JavaScript file would scrape the most recent video titles and make an AI API request. The output from the AI would also be received by this file. Another JavaScript file would need to handle the logic and user interactions with the GUI and receive the output from the previously mentioned file. There would also need to be an HTML file and CSS stylesheet for the GUI. Also, a “.json” file would need to exist for the project’s metadata and configuration, and possibly an icon folder too for the icons of the extension.
Proof of Concept https://github.com/cis3296s22/youtubestats?tab=readme-ov-file This repository inspired my project. I can fork it and modify it. The instructions to run are listed and it runs.
Background YouTube’s recommended video algorithm never seems to give me quality videos, but the search algorithm does give me interesting videos. I hope to make the search algorithm more convenient to use on YouTube by making it easier to think of search prompts. The product does not have any competitors that I can find, but the most similar competitor is this one, called Explore AI. (https://topai.tools/t/explore-ai). This one does not parse through one’s YouTube history, it only shows videos relating to specified topics. I will not be using any open-source products because there is no need for that. To my knowledge, this software is unique and there are no closed source/proprietary products similar to mine.
Required Resources I may need to pay for the AI API. I’m not sure which AI API would be easiest to work with yet, or which one would be the most cost-effective.
Slide link: https://docs.google.com/presentation/d/19HtBSnVGnIIdEPYfABYv2gtfNQ5LvbszLy3iDwQrC38/edit?usp=sharing