prakosoee/indobert-price-cls

ABSA Model - Price

This directory contains a fine-tuned BERT model for Aspect-Based Sentiment Analysis (ABSA), specifically targeting the Price aspect in the Food and Beverage (F&B) industry.

Model Details

• Model Architecture: BERT for Sequence Classification
• Base Model: indobenchmark/indobert-base-p2
• Aspect: Price (harga, nilai uang, promo)
• Language: Indonesian
• Task: Multi-class Sentiment Classification

Label Mapping

The model classifies text into three sentiment categories:

Label ID	Sentiment	Description
0	Negative	Expensive, poor value for money, or complaints about pricing.
1	Neutral	General mentions of price without specific positive or negative sentiment.
2	Positive	Affordable, good value, or price praise.

Evaluation Metrics

Based on the fine-tuning results, the model achieves an overall accuracy of 80%.

Classification Report

Category	Precision	Recall	F1-Score	Support
Negative	0.7742	0.8276	0.8000	29
Neutral	0.9091	0.6452	0.7547	31
Positive	0.7632	0.9355	0.8406	31
Accuracy			0.8022	91
Macro Avg	0.8155	0.8027	0.7984	91
Weighted Avg	0.8164	0.8022	0.7984	91

Usage

This model is loaded and used by the ABSA API service. All aspect models share a single tokenizer for optimal performance.

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model_path = "./absa-fnb-model/model_price"
model = AutoModelForSequenceClassification.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained("indobenchmark/indobert-base-p2")