Opinion Classification with Kili Technology and HuggingFace AutoML

The project repository for this article.

Just clone the repository and download the models by using scripts/download_models.py

I haven't included the torch installation in the requirements.txt. That's because the installation procedure wouldn't be same for everyone.

I have used PROJECT_ID and KILI_API_KEY environment variables to pass the API key and the project ID. I was using dotenv to load the environment variables from .env file.

Scripts usage:

collect_reviews.py

Script to collect the reviews by using google-play-scraper

Arguments

lang (l): The language reviews written in, default: English

country (c): The country to pull the reviews from, default: US

name (n): Package name of the application (Google Play Store)

count (co): Number of reviews to collect, default:10, all to get all reviews

filter (f): Integer to filter the reviews by rating, None for all rating. default: None

help: Shows the argument descriptions

Usage

python scripts/collect_reviews.py -co 100 --name medium.com.reader

download_models.py

Downloads the pre-trained models (finetuned models for 40 and 20 runs) into the current working directory. Uses gdown to download the models from google drive. Folder ID can be found in the script file

Arguments

None

Usage

python scripts/download_models.py

move_diff_to_review.py

Moves the distinct elements of the given two datasets into To Review section in the project.

Arguments

first: first CSV dataset to load,

second: second CSV dataset to load

Usage

python scripts/move_diff_to_review.py --first data/processed/dataset-2.csv --second data/processed/dataset-3.csv

prepare_dataset.py

Pulls the labeled and reviewed data separately and saves it with the given file name in data/processed folder.

Arguments

output_name: File name for the pulled dataset. (it should include .csv extension too)

remove (optional): Label to remove from the dataset

Usage

python scripts/prepare_dataset.py --output_name final_dataset.csv