ABSA Beer dataset and code

The datasets are results of the paper 

Unsupervised Aspect-Based Sentiment Analysis through LLM: A Case Study of an Unlabeled Portuguese Beer Database

Data sets:

Reviews Main (step_3_reviews_main.csv):

Step 3 produced this dataset, which supports the analysis of review comments, quantitative beer attributes, and review-related information, as summarized in Table 2. From the 67,083 reviews obtained in the previous step, 7,100 reviews (10.58%) were classified as non-relevant, while 59,982 reviews (89.42%) were deemed relevant. Consequently, the final dataset comprises 59,982 records.

Reviews Sample (step_4_1_reviews_sample.csv):

Step 4 generated this dataset as a subset of the Reviews Main, with 108 reviews, particularly suitable for benchmarking and validating new ABSA methodologies along side the ABSA Gold base.

ABSA Gold (step_4_ABSA_Gold.csv):

This manually annotated dataset, derived from Reviews Sample, serves as a gold standard for comparative evaluation of prompt-based approaches applied to the Reviews Sample dataset. It contains 1710 annotated BC from 108 reviews (columns: index, aspect, category, sentiment). 

ABSA Main (step_4_ABSA_main.csv):

The final dataset produced in Step 4 is designed to enable large-scale analysis and knowledge discovery within the beer industry. It comprises 880,373 records and supports the extraction of novel insights from consumer reviews (columns: index, aspect, category, sentiment).

ABSA Final (step_5_ABSA-FINAL.csv):

The dataset produced in Step 5 comprises only essential information for extracting the results from this step. Columns: index, review_comment, review_datetime, beer_style,  review_general_rate, review_aroma, review_visual, review_flavor, review_sensation, review_general_set, aspect, category, sentiment, year.