NER for Restaurant Reviews

Set Up Virtual Environment

$ python3 -m venv ner-reviews
$ source ner-reviews/bin/activate
(ner-reviews) $ pip3 install -r requirements.txt

If you prefer conda:

$ conda create -n ner-reviews python=3.9
$ conda activate ner-reviews
(ner-reviews) $ pip3 install -r requirements.txt

Data

html_reviews

This directory contains the HTML files for New York Times food reviews. It includes url_list.txt listing the corresponding URLs.

raw_data

This directory contains txt files for a processed dataset of New York Times food reviews. It includes cleaned_reviews.json and edit_cleaned_reviews.json, which contain all of the processed food review data. It includes unprocessed_URLs.txt, listing any unprocessed URLs.

tags.json

This file contains entity types to import into an annotation tool.

small_batch_annotation

This directory contains the JSON files for the annotated data from the initial small batch annotation.

annotated_data

This directory contains the JSON files for the annotated data, along with corresponding files that were corrected to remove encoding issues.

all_annotations.txt

This file contains all annotations from all annotators in CONLL format.

adjudicated_annotations.txt

This file contains adjudicated annotations in CONLL format.

counts.txt

This file contains counts of entities in adjudicated_annotations.txt.

final_data_splits

This directory contains the train, dev, and test sets for the adjudicated annotations.

Scripts

review_fetcher.py

This script downloads food reviews from New York Times and stores them in html_reviews.

clean_data.py

This script processes and cleans html_reviews, and outputs raw_data.

prepare_data.py

This script prepares raw_data for annotation.

fix_encoding_issue.py

This script fixes encoding issues in annotated_data.

data_process_pipeline.py

This script accepts annotated_data and processes the data to output all_annotations.txt.

data_process_utils.py

This script includes additional functions to support data_process_pipeline.py.

inter_annotator_agreement.py

This script calculates inter-annotator agreement with the processed data in all_annotations.txt.

adjudication.py

This script runs adjudication to produce gold standard annotations and outputs adjudicated_annotations.txt.

create_train_dev_test.py

This script accepts adjudicated_annotations.txt and splits the data into train, dev, and test sets.

NER_restaurant_reviews.ipynb

This script runs the model and experiments using final_data_splits.