FLASH: Federated Learning-Based LLMs for Advanced Query Processing in Social Networks through RAG

TL;DR

The paper tackles the challenge of extracting trustworthy, real-time social-media insights from vast, privacy-sensitive data. It proposes a federated learning–driven GPT framework that combines role-based prompting, PEFT (LoRA and P-Tuning-v2), and context-aware vector embeddings with OpenSearch and SVM-based retrieval to deliver personalized, timely information without exposing raw data. The architecture is modular: data ingestion and metadata enrichment, context generation with a context-GPT trained via FedAvg across multiple clients, and an interactive UI for professionals, with quantization and checkpointing to improve efficiency. The results indicate improved retrieval relevance and real-time performance, along with transparency about data origins and user-controlled contexts, highlighting practical benefits for privacy-preserving social media analytics and engagement.

Abstract

Our paper introduces a novel approach to social network information retrieval and user engagement through a personalized chatbot system empowered by Federated Learning GPT. The system is designed to seamlessly aggregate and curate diverse social media data sources, including user posts, multimedia content, and trending news. Leveraging Federated Learning techniques, the GPT model is trained on decentralized data sources to ensure privacy and security while providing personalized insights and recommendations. Users interact with the chatbot through an intuitive interface, accessing tailored information and real-time updates on social media trends and user-generated content. The system's innovative architecture enables efficient processing of input files, parsing and enriching text data with metadata, and generating relevant questions and answers using advanced language models. By facilitating interactive access to a wealth of social network information, this personalized chatbot system represents a significant advancement in social media communication and knowledge dissemination.

Figures (4)

• Figure 1: Federated learning architecture explaining the workflow of creating social media context-based GPT.
• Figure 2: The system architecture illustrates various components. On the leftmost side, the initial module depicts the flow of the federated learning system for generating a context-based GPT. This module outlines the key steps involved in its implementation, from initiation to completion of the workflow. Within the figure, we demonstrate the process of enhancing user queries by supplementing them with pertinent information blocks. These are provided alongside the user's question to the GPT model, prompting it to expand the query based on role-specific cues.
• Figure 3: The user interface of the chatbot designed for social media professionals is thoughtfully structured to enhance user experience and streamline interactions. The central feature of the interface is the 'Ask AI' search window, which serves as the primary access point for users to engage with the chatbot. This search window allows users to input their queries or requests, leveraging the chatbot's capabilities to retrieve and generate relevant information or responses.
• Figure 4: The four figures above depict varying accuracy levels for FedGPT across different metric scales. Blue represents p-Tuning, while orange represents LoRA. The y-axis indicates accuracies, and the x-axis represents the number of training rounds conducted.