Twitch Chat Sentiment Analysis

This repository provides a Sentiment Analysis System for Twitch chat messages, leveraging modern technologies for real-time data processing, analysis, and visualization.

Overview

The project is divided into three services:

1. Chat Reader
   A Go-based service that connects to Twitch IRC, reads chat messages, and produces them to a Kafka topic for further processing.

2. Message Analyzer
   A Python-based service that consumes messages from Kafka, performs sentiment analysis using a Hugging Face neural network model, and stores results in a PostgreSQL database.

   • Model Used: lxyuan/distilbert-base-multilingual-cased-sentiments-student.

3. Website
   A Go + Vue-based web service that displays the sentiment analysis results and recent messages in real-time via WebSocket updates.

Key Features

• Real-Time Data Processing: Stream and analyze Twitch chat messages instantly.
• Scalable Architecture: Decoupled services using Kafka.
• Insightful Visualization: Interactive dashboard for visualizing results.
• Containerized Deployment: Docker and Docker Compose for streamlined setup.
• Metrics Integration: Prometheus and Grafana for monitoring system performance.

---

System Architecture

flowchart LR
    TwitchChat -- "chat<br>messages" --> MessagesTopic
    subgraph ChatReader["Chat Reader"]
        direction TB

        TwitchChat(Twitch Chat<br>IRC)
    end

    subgraph KafkaBroker["Kafka Broker"]
    MessagesTopic@{ shape: das, label: "messages" }
    end

    MessagesTopic --> Model
    Model -- "Classified<br>messages" --> ResultsTable
    subgraph Analyzer["Message Analyzer"]
        direction TB

        Model[Sentiment<br>Analysis<br>Model]
    end
    
    subgraph PostgresDatabase["PostgreSQL"]
    ResultsTable@{ shape: cyl, label: "results" }
    end

    ResultsTable --> Backend
    subgraph Website["Website"]
        direction TB

        Backend["Backend"]
        Frontend["Frontend"]

        Backend -- Channel analysis --> Frontend
        Backend -- Last messages --> Frontend
    end

Loading

---

Demo

demo.mp4

---

Tech Stack

Chat Reader

• Language: Go
• Libraries:
  • franz-go: Kafka client
  • go-twitch-irc: Twitch IRC client
  • Testcontainers: For integration tests

Message Analyzer

• Language: Python
• Libraries:
  • confluent-kafka-python: Kafka client
  • transformers: Neural network API
  • psycopg2: PostgreSQL client

Website

• Backend: Go
  • gorilla/websocket: WebSocket support
  • pgx: PostgreSQL driver
  • Testcontainers: For integration tests
• Frontend: Vue.js
  • chart.js + vue-chartjs: Data visualization
  • shadcn-vue: UI components
  • tailwindcss: CSS framework

---

Metrics and Monitoring

Prometheus is used to collect metrics, and Grafana visualizes them via a dashboard.

Available Metrics

Chat Reader

• kafka_messages_processed_total: Total produced messages
• twitch_messages_read_total: Total messages read and filtered by the client

Website

• broadcast_hub_clients_total: Total active WebSocket clients
• broadcast_hub_messages_total: Total messages broadcasted to frontend clients

Grafana Dashboard

---

How to Use

1. Clone the Repository

2. Setup Dependencies:

   • Install Docker Engine

3. Run the Project:

   docker compose up

4. Access the Dashboard:

   • Website: http://localhost:8080
   • Grafana: http://localhost:3000
     • Default user and password: admin

---

Limitations and Bias Disclaimer

This project uses the lxyuan/distilbert-base-multilingual-cased-sentiments-student model for sentiment analysis. While the model provides robust multilingual sentiment analysis, it is important to note:

1. Potential Bias: The model may reflect biases present in its training data, potentially impacting the accuracy and fairness of sentiment classification.
2. Domain Limitations: The model's performance may vary depending on the language and context of the chat messages.

Users should interpret the results with caution and consider these limitations when analyzing the data.

---

License

This project is licensed under the MIT License.