You can deploy this model and start earning money today!

farelnickholas/
RoBERTa-gojek-aspect-id-2026

Name: RoBERTa-gojek-aspect-id-2026
Availability: InStock

text-classification

roberta

aspect-based-sentiment-analysis

gojek

mit

Views

No views yet

RoBERTa Aspect Classification for Gojek Reviews

Model Description

This model is a fine-tuned RoBERTa model designed for Aspect Classification on Indonesian Gojek application reviews. It is part of an Aspect-Based Sentiment Analysis (ABSA) pipeline. The model classifies reviews into 5 distinct aspects related to the Gojek app experience.

Developed by: Wayan Farel Nickholas Sadewa
Model type: RoBERTa for Sequence Classification (using Keras/TensorFlow)
Language(s) (NLP): Indonesian (id)
License: MIT

Intended Uses & Limitations

This model is intended to be used on user reviews of the Gojek application to identify which aspect of the service the user is discussing. It is specifically trained on Gojek-related terms and Indonesian slang (which requires preprocessing).

Limitations

The model might not perform as well on domains outside of Gojek or ride-hailing/food delivery apps.
Requires preprocessing such as slang normalization to achieve optimal results.

Model Architecture

The model is built using TFRobertaModel from Hugging Face's transformers library, coupled with custom classification layers using Keras. It expects input sequences of maximum length 128.

Labels

The model classifies text into one of the following 5 aspect categories:

0: Kenaikan Biaya & Penurunan Manfaat Promo GoFood
1: Masalah Driver & Layanan (Ketersediaan, Profesionalitas, Keselamatan)
2: Masalah Akun, GoPaylater & Transaksi
3: Pujian & Kepuasan Pengguna
4: Masalah Performa & Stabilitas Aplikasi

Usage

The model is saved in HDF5 (.h5) format. It can be loaded using Keras:

python

import tensorflow as tf
from tensorflow import keras
from transformers import TFRobertaModel, RobertaTokenizerFast

# Load tokenizer
tokenizer = RobertaTokenizerFast.from_pretrained("roberta_tokenizer_aspect")

# Load model
model = keras.models.load_model(
    "roberta_aspect_model.h5",
    custom_objects={"TFRobertaModel": TFRobertaModel},
    compile=False
)

# Example usage
text = "drivernya ramah banget dan cepat sampai"
encoded = tokenizer(
    text,
    padding="max_length",
    truncation=True,
    max_length=128,
    return_tensors="tf"
)

predictions = model({
    "input_1": encoded["input_ids"],
    "input_2": encoded["attention_mask"]
})

aspect_idx = int(tf.argmax(predictions, axis=1)[0])
print(f"Predicted Aspect: {aspect_idx}")

farelnickholas/RoBERTa-gojek-aspect-id-2026