Empathica

Project Overview

This project aims to develop a humanised chatbot targeting B2B interactions, leveraging Large Pretrained Language Models (LPU based LLMs) to enhance user experience. The chatbot integrates sentiment analysis, retrieval augmented generation, and a feedback mechanism, all orchestrated through a federal learning approach for decentralized global model generation.

Contributors

• Vinayak Goyal
• Krishna Rathore
• Hemant Chaurasia
• Anvaya Sharma
• Rishabh Raunak

Features

Sentiment Analysis

• Methodology: Integrating sentiment analysis into the chatbot pipeline to determine the sentiment of user messages.

• Interpretation: Assigning sentiment scores (positive, negative, or neutral) to messages.

• Response Strategies:

  • Positive Sentiment: Respond enthusiastically and affirmatively.
  • Negative Sentiment: Express empathy and offer support or solutions.
  • Neutral Sentiment: Respond neutrally or informatively.

• Contextual Understanding: Analyzing sentiment in the context of the conversation for better accuracy.

  

  

Retrieval Augmented Generation

• Methodology: Utilizing retrieval augmented generation for context-based query responses. This technique combines the advantages of retrieval-based and generative-based approaches, allowing the chatbot to retrieve relevant information from a knowledge base and generate responses accordingly.

• Advantages: Enables the chatbot to provide more accurate and contextually relevant responses to user queries.

• Implementation: Incorporating a knowledge base and retrieval mechanism to support the chatbot in retrieving and generating responses based on user queries.

  

  

Enhancing User Experience

• Anthropomorphic Visual Cues: Implementing visual cues for a human-like effect to enhance social presence and perceived homophily.
• Benefits:
  • Greater Social Presence: Leading to favorable attitudes and increased behavioral intentions to return to the website.
  • Perceived Homophily: Contributing to positive attitudes and increased return intentions.
• Implementation: Designing the chatbot interface with anthropomorphic visual cues such as avatars or emojis to foster a sense of human-like interaction.

Feedback Mechanism

• Real-Time Feedback Integration: Allowing users to provide feedback on response relevance and quality in real-time.
• Feedback Interpretation: Analyzing feedback sentiment and categorizing it based on specific response aspects.
• Dynamic Response Adjustment: Using feedback to dynamically adjust response generation.
• Relevance Ranking Optimization: Incorporating feedback into the relevance ranking mechanism to prioritize relevant information retrieval.
• Learning from User Interactions: Employing machine learning algorithms to analyze feedback patterns and iteratively improve chatbot performance.
• Implementation: Integrating feedback buttons or text input within the chatbot interface for users to provide real-time feedback.

Federal Learning

• Approach: Implementing federal learning for decentralized global model generation. This approach allows multiple distributed nodes to collaboratively train a global model while preserving data privacy and security.
• Advantages:
  • Data Privacy: Ensures that sensitive user data remains localized and is not shared across the network.
  • Scalability: Facilitates scalable model training by distributing computation across multiple nodes.
  • Robustness: Increases model robustness by aggregating knowledge from diverse data sources.
• Implementation: Configuring the chatbot architecture to incorporate federal learning methodologies for model training and updates.

Usage

• Clone the repository.
• Install dependencies.
• Run the chatbot application.
• Monitor feedback and performance metrics.
• Iterate on improvements based on user interactions.

Dependencies

• LPU based Large Language Models
• Sentiment Analysis Model
• CRM System
• Visual Cues Implementation
• Machine Learning Algorithms

Future Improvements

• Multilingual Support: Enhance chatbot capabilities to support multiple languages for broader user reach.
• Integration with Voice Assistants: Enable voice interaction for users who prefer spoken communication.
• Advanced Contextual Understanding: Implement more sophisticated algorithms for understanding and responding to complex user queries.
• Personalization Capabilities: Incorporate user preferences and history to tailor responses and recommendations.
• Integration with Business Systems: Connect the chatbot with enterprise systems for seamless workflow assistance and data integration.

License

This project is licensed under the MIT License.