TalkPole

Overview

TalkPole is an innovative sentiment analysis model designed to accurately predict the sentiment of French sentences. Leveraging a range of neural network architectures, TalkPole excels in discerning both positive and negative sentiments with high precision. Trained on a comprehensive French dataset, TalkPole is finely tuned to understand and interpret the nuances of the French language.

As a forward-thinking solution, TalkPole seamlessly integrates into microservices architectures and offers flexible deployment through REST APIs, making it a versatile choice for modern applications.

Features

Multiple Models

• CNN
• LSTM
• CNN-BiLSTM
• DistilBERT (still under development)

Here is a comparative analysis of the models tested for classifying text as positive or negative. The CNN model demonstrated the best performance in accurately labeling the sentiment.

However, when it comes to precision in determining the sentiment polarity, the CNN model fell short. In contrast, the CNN-BiLSTM model excelled in this aspect. The improved precision of the CNN-BiLSTM model is largely due to the context-sensitive capabilities of the BiLSTM layer, which enhances its ability to capture and interpret subtle nuances in sentiment more effectively.

Here's when to use each model:

CNN

• Perfomance Requirements
• Precision in Positivity or Negativity
• No need to determine Neutrality
• No context awareness needed
• Large Amount of Data

CNN-BiLSTM & LSTM

• Context awareness needed
• Need to determine Neutrality
• Polarity Precision
• Small Amount of Data

REST API Version

Exposes a REST API for interaction with the sentiment analysis service.

Kafka Version

Integrates with Kafka to process and analyze data in real-time.

This version could be used in Microservices architecture.

UI Version

In Case, You wanted to experiment with different models and check documetations, you can use the UI Version

Tools

This project wouldn't be possible without these tools:

Installation

Prerequisites

• Python 3.x
• Tensorflow 2.17.0
• Keras 3.x
• Flask 3.x
• Kafka
• Docker

Steps

For Kafka Versions

1. Clone Repository or use docker image in dockerhub

git clone https://github.com/Redtius/TalkPole.git -b Kafka-CNN-BiLSTM

OR

docker pull redtius/talkpole:kafka-cnn-bilstm

2. Set Up Kafka for the container
   make sure you setup these Required Environment variables:

• KAFKA_PORT : set it to the port of the kafka server.
• KAFKA_HOST : set it to the host of the kafka server.

3. Run the Application

you can use docker compose to run the application

docker-compose up -d

OR

run the container depending on your own use

docker run -e KAFKA_PORT=9092 -e KAFKA_HOST=localhost -p 5000:5000 talkpole

Usage

Request Body:

{
 #"ref":"reference-id",
 "content":"c'est le meilleur produit"
}

Response Body:

{
 #"ref":"reference-id",
 #"content":"c'est le meilleur produit",
 "result":"0.97864554"
}

Kafka Integration

• Producer: Sends Messages to talkpole_out topic.
• Consumer: Listens to talkpole_in topic.

Testing

To run Unit Tests for the project

pytest

Troubleshooting

• Common Issues
  • Kafka connection problems: Ensure Kafka is running and accessible.
  • Model loading issues: Verify model files are present and correctly referenced.
  • Versions incompatibility: Make sure all dependencies have the right versions and python is 3.12+.

License

This Project is under the Apache 2.0 License.