TheRockXu / youtube-title-scoring

Score YouTube Titles to Maximize views using Recurrent Neural Net

Score YouTube Titles

This is a RNN model that ranks a list of YouTube video titles for the potential of maximizing views.

Dependencies

tensorflow 2.1.0
pandas 1.04
selenium 3.1 for extracting trainning data

Model Architecture

Training Data

data.csv contains 40k raw data collected from YouTube. Feel free to use yt-data-extraction.py to collect more training data.

The data is collected by searching through a list of keywords on YouTube and scrape the top 25 results. You can add more search terms to the search_terms file. It will repeatedly run a headless selenium driver process to search YouTube.

Usage

To train the model, simply run python train.py

To test some titles simply run python test.py "What Your Car Says About You" "J.R. Smith Beats Up Alleged Car Vandalizer" "14 Ingenious Car Inventions & Technologies" "I DESTROYED RYANS CAR!!"

Here is the result of running the test script.

The result is a rank of the titles by their potential for views.