Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,134 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# π¬ Sentiment-Analysis-for-Product-Release-Sentiment
|
2 |
+
|
3 |
+
A **BERT-based sentiment analysis model** fine-tuned on a product review dataset. It predicts the sentiment of a text as **Positive**, **Neutral**, or **Negative** with a confidence score. This model is ideal for analyzing customer feedback, reviews, or user comments.
|
4 |
+
|
5 |
+
---
|
6 |
+
|
7 |
+
## β¨ Model Highlights
|
8 |
+
|
9 |
+
- π **Architecture**: Based on [`bert-base-uncased`](https://huggingface.co/bert-base-uncased) by Google
|
10 |
+
- π§ **Fine-tuned** on labeled product review data
|
11 |
+
- π **3-way sentiment classification**: `Negative (0)`, `Neutral (1)`, `Positive (2)`
|
12 |
+
- πΎ **Quantized version available** for faster inference
|
13 |
+
|
14 |
+
---
|
15 |
+
|
16 |
+
## π§ Intended Uses
|
17 |
+
|
18 |
+
- β
Classifying product feedback and user reviews
|
19 |
+
- β
Sentiment analysis for e-commerce platforms
|
20 |
+
- β
Social media monitoring and customer opinion mining
|
21 |
+
|
22 |
+
---
|
23 |
+
|
24 |
+
## π« Limitations
|
25 |
+
|
26 |
+
- β Designed for English texts only
|
27 |
+
- β May not perform well on sarcastic or ironic inputs
|
28 |
+
- β May struggle with domains very different from product reviews
|
29 |
+
- β Input texts longer than 128 tokens are truncated
|
30 |
+
|
31 |
+
---
|
32 |
+
|
33 |
+
## ποΈββοΈ Training Details
|
34 |
+
|
35 |
+
- **Base Model**: `bert-base-uncased`
|
36 |
+
- **Dataset**: Custom-labeled product review dataset
|
37 |
+
- **Epochs**: 5
|
38 |
+
- **Batch Size**: 8
|
39 |
+
- **Max Length**: 128 tokens
|
40 |
+
- **Optimizer**: AdamW
|
41 |
+
- **Loss Function**: CrossEntropyLoss (with class balancing)
|
42 |
+
- **Hardware**: Trained on NVIDIA GPU (CUDA-enabled)
|
43 |
+
|
44 |
+
---
|
45 |
+
|
46 |
+
## π Evaluation Metrics
|
47 |
+
|
48 |
+
| Metric | Score |
|
49 |
+
|------------|-------|
|
50 |
+
| Accuracy | 0.90 |
|
51 |
+
| F1 | 0.90 |
|
52 |
+
| Precision | 0.90 |
|
53 |
+
| Recall | 0.90 |
|
54 |
+
|
55 |
+
---
|
56 |
+
|
57 |
+
## π Label Mapping
|
58 |
+
|
59 |
+
| Label ID | Sentiment |
|
60 |
+
|----------|-----------|
|
61 |
+
| 0 | Negative |
|
62 |
+
| 1 | Neutral |
|
63 |
+
| 2 | Positive |
|
64 |
+
|
65 |
+
---
|
66 |
+
|
67 |
+
## π Usage Example
|
68 |
+
|
69 |
+
```python
|
70 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
|
71 |
+
from transformers import DataCollatorWithPadding
|
72 |
+
import torch
|
73 |
+
import torch.nn.functional as F
|
74 |
+
|
75 |
+
# Load model and tokenizer
|
76 |
+
model_name = "AventIQ-AI/Sentiment-Analysis-for-Product-Release-Sentiment"
|
77 |
+
tokenizer = BertTokenizer.from_pretrained(model_name)
|
78 |
+
model = BertForSequenceClassification.from_pretrained(model_name)
|
79 |
+
model.eval()
|
80 |
+
|
81 |
+
# Inference
|
82 |
+
def predict_sentiment(text):
|
83 |
+
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
84 |
+
inputs = {k: v.to(quantized_model.device) for k, v in inputs.items()}
|
85 |
+
with torch.no_grad():
|
86 |
+
logits = quantized_model(**inputs).logits
|
87 |
+
probs = F.softmax(logits, dim=1)
|
88 |
+
|
89 |
+
predicted_class_id = torch.argmax(probs, dim=1).item()
|
90 |
+
confidence = probs[0][predicted_class_id].item()
|
91 |
+
|
92 |
+
label_map = {0: "Negative", 1: "Positive"}
|
93 |
+
label = label_map[predicted_class_id]
|
94 |
+
confidence_str = f"confidence : {confidence * 100:.1f}%"
|
95 |
+
|
96 |
+
return label, confidence_str
|
97 |
+
|
98 |
+
# Example
|
99 |
+
print(predict_sentiment(
|
100 |
+
"The service was excellent and the staff was friendly.")
|
101 |
+
)
|
102 |
+
```
|
103 |
+
|
104 |
+
---
|
105 |
+
|
106 |
+
## π§ͺ Quantization
|
107 |
+
|
108 |
+
- Applied **post-training dynamic quantization** using PyTorch to reduce model size and speed up inference.
|
109 |
+
- Quantized model supports CPU-based deployments.
|
110 |
+
|
111 |
+
---
|
112 |
+
|
113 |
+
## π Repository Structure
|
114 |
+
|
115 |
+
```
|
116 |
+
.
|
117 |
+
βββ model/ # Quantized model files
|
118 |
+
βββ tokenizer/ # Tokenizer config and vocabulary
|
119 |
+
βββ model.safetensors/ # Fine-tuned full-precision model
|
120 |
+
βββ README.md # Model documentation
|
121 |
+
```
|
122 |
+
|
123 |
+
---
|
124 |
+
|
125 |
+
## π Limitations
|
126 |
+
|
127 |
+
- May not generalize to completely different domains (e.g., medical, legal)
|
128 |
+
- Quantized version may show slight drop in accuracy compared to full-precision model
|
129 |
+
|
130 |
+
---
|
131 |
+
|
132 |
+
## π€ Contributing
|
133 |
+
|
134 |
+
We welcome contributions! Please feel free to raise an issue or submit a pull request if you find a bug or have a suggestion.
|