https://github.com/luluw8071/text-sentiment-analysis

Fine-Tuning Distil BERT and LSTM for Comparative Analysis
https://github.com/luluw8071/text-sentiment-analysis

bert bert-fine-tuning lstm-neural-networks nlp pytorch sentiment-classification text-classification

Last synced: 2 days ago
JSON representation

Fine-Tuning Distil BERT and LSTM for Comparative Analysis

• Host: GitHub
• URL: https://github.com/luluw8071/text-sentiment-analysis
• Owner: LuluW8071
• License: mit
• Created: 2023-12-27T12:22:36.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2024-11-05T08:06:38.000Z (9 days ago)
• Last Synced: 2024-11-05T09:19:17.355Z (9 days ago)
• Topics: bert, bert-fine-tuning, lstm-neural-networks, nlp, pytorch, sentiment-classification, text-classification
• Language: Jupyter Notebook
• Homepage:
• Size: 883 KB
• Stars: 7
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Text Sentiment Analysis using BERT & LSTM



**Sentiment analysis** is an NLP technique to classify the sentiment (_positive_, _negative_, or _neutral_) of text data. It involves processing textual input, classifying sentiment using techniques like lexicon-based, machine learning, or deep learning models, and has applications in business, social media monitoring, finance, and healthcare. It helps understand public sentiment, customer satisfaction, and market trends.

### [Download Dataset](https://drive.usercontent.google.com/download?id=1Jp3D5gdxGrwa5dHbr4p-pECrD8wi7vik&authuser=0)

### Text Sentiment Analysis using BERT



**BERT** is pre-trained on a large text corpus using tasks like masked language modeling and next sentence prediction. Fine-tuning on specific tasks involves adjusting the final layers of the pre-trained BERT model which allows us to adapt it to specific NLP tasks such as **text classification**, **named entity recognition**, **sentiment analysis**, and **question answering**.

### Text Sentiment Analysis using LSTM

For LSTM Model, the **Embedding Layer** converts tokens into **dense vectors** to capture word **similarities**. The **LSTM Layer** processes **sequences**, maintaining **hidden states** for **long-term dependencies**. A **Dropout Layer** prevents **overfitting** by randomly setting input units to **zero**. The **Fully Connected Layer** transforms LSTM output, and a **Sigmoid Activation Function** squashes values to predict **sentiment probabilities**. This architecture effectively analyzes text for **sentiment classification**.



>Note: Above image is not correct representation of LSTM model defined in my notebook. It is just a representation of how LSTM works on text sentiment analysis. 

---

Feel free to send issues if you face any problem.

Don't forget to star the repo :star: