dipawidia/xlnet-base-cased-product-review-sentiment-analysis

<!-- This model card has been generated automatically according to the information Keras had access to. You should probably proofread and complete it, then remove this comment. -->

dipawidia/xlnet-base-cased-product-review-sentiment-analysis

This model is a fine-tuned version of xlnet-base-cased on any type of product reviews dataset gathered from several e-commerce such as shopee, tokopedia, blibli, lazada, and zalora. The dataset can be found here It achieves the following results on the evaluation set:

• —Train Loss: 0.1085
• —Train Accuracy: 0.9617
• —Validation Loss: 0.1910
• —Validation Accuracy: 0.9414
• —Epoch: 4

Intended uses & limitations

This fine-tuned XLNet model is used for sentiment analysis with 2 labels text classification: 0 -> Negative; 1 -> Positive.

Example Pipeline

from transformers import pipeline
pipe = pipeline("text-classification", model="dipawidia/xlnet-base-cased-product-review-sentiment-analysis")
pipe("This shoes is awesome")

[{'label': 'Positive', 'score': 0.9995703101158142}]

Full classification example

from transformers import XLNetTokenizer, TFXLNetForSequenceClassification
import tensorflow as tf
import numpy as np

tokenizer = XLNetTokenizer.from_pretrained("dipawidia/xlnet-base-cased-product-review-sentiment-analysis")
model = TFXLNetForSequenceClassification.from_pretrained("dipawidia/xlnet-base-cased-product-review-sentiment-analysis")

def get_sentimen(text):
  tokenize_text = tokenizer(text, return_tensors = 'tf')
  preds = model.predict(dict(tokenize_text))['logits']
  class_preds = np.argmax(tf.keras.layers.Softmax()(preds))
  if class_preds == 1:
    label = 'Positive'
  else:
    label = 'Negative'
  return(label)

get_sentimen('i hate this product')

Output:

Negative

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —optimizer: {'name': 'AdamW', 'weightdecay': 0.004, 'clipnorm': None, 'globalclipnorm': None, 'clipvalue': None, 'useema': False, 'emamomentum': 0.99, 'emaoverwritefrequency': None, 'jitcompile': True, 'islegacyoptimizer': False, 'learningrate': 3e-05, 'beta1': 0.9, 'beta2': 0.999, 'epsilon': 1e-07, 'amsgrad': False}
• —training_precision: float32

Training results

Framework versions

• —Transformers 4.41.2
• —TensorFlow 2.15.0
• —Tokenizers 0.19.1