Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedDTRE: Federated Dialogue Generation Models Powered by Trustworthiness Evaluation

Shule Lu, Lingxiang Wang, Sijia Wen, Ziwei Wang, Hainan Zhang

TL;DR

This paper tackles privacy-preserving open-domain dialogue generation in federated settings by introducing FedDTRE, a trustworthiness-guided adaptive aggregation strategy. A federated, BERT-based trustworthiness evaluator is trained to produce scores that balance semantic relevance with privacy considerations, and these scores dynamically regulate the global model’s contribution during local updates via an adaptive fusion coefficient $\alpha$. Empirical results on Synthetic-Persona-Chat, CMU_DoG, and WoW show that FedDTRE improves lexical and contextual quality on two datasets while maintaining competitive semantic fidelity, with some trade-offs on strictly knowledge-grounded tasks. Overall, FedDTRE reduces overfitting to small local datasets and preserves global knowledge with modest computational overhead, making it well-suited for privacy-sensitive, heterogeneous dialogue applications.

Abstract

With the rapid development of artificial intelligence, dialogue systems have become a prominent form of human-computer interaction. However, traditional centralized or fully local training approaches face challenges in balancing privacy preservation and personalization due to data privacy concerns and heterogeneous device capabilities. Federated learning, as a representative distributed paradigm, offers a promising solution. However, existing methods often suffer from overfitting under limited client data and tend to forget global information after multiple training rounds, leading to poor generalization. To address these issues, we propose FedDTRE, a Federated adaptive aggregation strategy for Dialogue generation based on Trustworthiness Evaluation. Instead of directly replacing local models with the global model, FedDTRE leverages trustworthiness scores of both global and local models on a fairness-oriented evaluation dataset to dynamically regulate the global model's contribution during local updates. Experimental results demonstrate that FedDTRE can improve dialogue model performance and enhance the quality of dialogue generation.

FedDTRE: Federated Dialogue Generation Models Powered by Trustworthiness Evaluation

TL;DR

This paper tackles privacy-preserving open-domain dialogue generation in federated settings by introducing FedDTRE, a trustworthiness-guided adaptive aggregation strategy. A federated, BERT-based trustworthiness evaluator is trained to produce scores that balance semantic relevance with privacy considerations, and these scores dynamically regulate the global model’s contribution during local updates via an adaptive fusion coefficient . Empirical results on Synthetic-Persona-Chat, CMU_DoG, and WoW show that FedDTRE improves lexical and contextual quality on two datasets while maintaining competitive semantic fidelity, with some trade-offs on strictly knowledge-grounded tasks. Overall, FedDTRE reduces overfitting to small local datasets and preserves global knowledge with modest computational overhead, making it well-suited for privacy-sensitive, heterogeneous dialogue applications.

Abstract

With the rapid development of artificial intelligence, dialogue systems have become a prominent form of human-computer interaction. However, traditional centralized or fully local training approaches face challenges in balancing privacy preservation and personalization due to data privacy concerns and heterogeneous device capabilities. Federated learning, as a representative distributed paradigm, offers a promising solution. However, existing methods often suffer from overfitting under limited client data and tend to forget global information after multiple training rounds, leading to poor generalization. To address these issues, we propose FedDTRE, a Federated adaptive aggregation strategy for Dialogue generation based on Trustworthiness Evaluation. Instead of directly replacing local models with the global model, FedDTRE leverages trustworthiness scores of both global and local models on a fairness-oriented evaluation dataset to dynamically regulate the global model's contribution during local updates. Experimental results demonstrate that FedDTRE can improve dialogue model performance and enhance the quality of dialogue generation.

Paper Structure

This paper contains 15 sections, 12 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model
  4. Trustworthiness Evaluation
  5. Federated Dialogue Generation
  6. Experiments
  7. Experimental Settings
  8. Evaluation Datasets
  9. Trustworthiness Dataset
  10. Training Details
  11. Evaluation Metrics
  12. Baselines
  13. Main Results
  14. Ablation Study
  15. Conclusion

Figures (2)

  • Figure 1: The overall framework of FedDTRE. It evaluates the trustworthiness of global and local models on a fairness-oriented dataset, and dynamically regulates the global model’s contribution during local updates.
  • Figure 2: Federated training process of the BERT-based trustworthiness model. Each client trains the model on its local private data and uploads only model parameters to the server. The server performs global aggregation to obtain an optimized Global Trustworthiness Evaluator.