TranscriptNLP

Project for comparing the caption of popular youtuber's apology videos. Scores for similarity.

Get Started

Structure this project:

PARENTFOLDER/
    TRANSCRIPTS/
        TEXT/
        JSON/
        GRAPHS/
        ratio_list.json
        youtube_vids.json
        
    TranscriptNLP/
        graph.py
        main.py

1. Clone the repo.
2. Run pip install -r requirements.txt
3. Go to main.py and change score_word to True.
4. Run main.py.
5. Profit!

Overview

main.py

Using the files in the TEXT/ folder, it will train a gensim model with "wiki-gigaword-100" and all the sentences from the transcripts. It will generate a score and save it to similarity_score.json.

Using the ratio_list.json generated by TranscriptCollect, it will create a .csv file and save it as ratios.csv.

graph.py

Using similarity_score.json it will create a heatmap png named similarity_graph.png. Requires /TEXTS/ to match the file.

Using ratios.csv, it will create a grouped bar graph named ratio_graph.png.

Need to get the captions?

Try this repo out!