Arabic-Book-Review-Sentiment-Assessment

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

Arabic-Book-Review-Sentiment-Assessment

This model is a fine-tuned version of google-bert/bert-base-multilingual-uncased on labr dataset. It achieves the following results on the evaluation set:

• Loss: 1.5290

Model description

The purpose of this model is to analyze Arabic review texts and predict the appropriate rating for them, based on the sentiment and content of the review. This can be particularly useful in tasks such as sentiment analysis, customer feedback analysis, or any application where understanding the sentiment conveyed in an Arabic textual review is important.

Intended uses & limitations

While the model performs well with formal Arabic text (Examples 1, 3, and 4), it may struggle with slang or informal language, occasionally assigning higher ratings than expected (Example 2). Additionally, the model is not capable of interpreting verbally given ratings (Example 5). Users should be aware of these limitations and provide context-appropriate input for optimal performance.

Training and evaluation data

More information needed

Training procedure

import torch
from datasets import load_dataset
from transformers import (
  AutoModelForSequenceClassification,
  AutoTokenizer,
  DataCollatorWithPadding,
  TrainingArguments,
  Trainer
)


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

labr = load_dataset("labr")

labels = {0,1,2,3,4}
target_names = [
    "Poor",
    "Fair",
    "Good",
    "Very Good",
    "Excellent"
]

id2label = {idx: label for idx, label in enumerate(target_names)}
label2id = {label: idx for idx, label in enumerate(target_names)}


BERT_MODEL = "google-bert/bert-base-multilingual-uncased"

model = AutoModelForSequenceClassification.from_pretrained(BERT_MODEL, num_labels = len(id2label))
tokenizer = AutoTokenizer.from_pretrained(BERT_MODEL)

model.to(device)


def preprocess_function(examples):
    return tokenizer(examples["text"], truncation=True)

tokenized_labr = labr.map(preprocess_function, batched=True)


data_collator = DataCollatorWithPadding(tokenizer=tokenizer)

training_args = TrainingArguments(
    output_dir="Arabic-Book-Review-Sentiment-Assessment",
    learning_rate=2e-5,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=1,
    weight_decay=0.01,
    push_to_hub=True
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_labr["train"],
    eval_dataset=tokenized_labr["test"],
    tokenizer=tokenizer,
    data_collator=data_collator,
)

trainer.train()

trainer.evaluate(tokenized_labr["test"])

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Framework versions

• Transformers 4.39.1
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2