NLP
GitHub RepositoryYouTube Transcript Pipeline
From selected YouTube links to clean datasets and repeatable NLP experiments.
Summary
The pipeline starts with a curated list of YouTube videos, retrieves their transcripts, cleans and structures the text, and produces datasets ready for analysis. It then supports repeatable classification experiments ranging from TF-IDF and logistic regression baselines to LSTM models, with reusable outputs for later work.
From selected videos to reusable NLP artifacts
Video list
Curated links define the source collection
Retrieval
Automation obtains available transcripts
Preparation
Cleaning creates analysis-ready datasets
Experiments
Classical and LSTM models run classifications
Artifacts
Data, metrics, and models remain reusable
What the system handled
Turns curated YouTube links into cleaned transcript datasets, baseline experiments, and reusable artifacts.
Technology
Stack
Areas
Conclusions
- Treating transcript collection as a repeatable pipeline gives later NLP experiments a reproducible source of text.
- Classical TF-IDF baselines provide an important reference before the added complexity of an LSTM model is evaluated.
- Transcript availability and quality remain external constraints, so failed retrievals and noisy captions need to stay visible in the resulting dataset.
Want to inspect the implementation?
The repository contains the source, setup notes, and development history.