DESERE: The 1st Workshop on Decentralised Search and Recommendation

Mohamed Ragab, Yury Savateev, Wenjie Wang, Reza Moosaei, Thanassis Tiropanis, Alexandra Poulovassilis, Adriane Chapman, Helen Oliver, George Roussos

Introduction

The DESERE workshop represents a foundational effort to transition search and recommendation services from centralized silos toward a decentralized Web architecture. By leveraging personal online datastores (pods), the workshop explores how to reconcile high-performance discovery tools with the increasing demand for individual data sovereignty and privacy.

The Challenge of Centralization

Currently, the majority of digital interactions are indexed and mediated by a few large, centralized entities. While this model offers efficiency, it introduces significant privacy risks and deprives individuals of control over their personal data.

Emerging frameworks such as Solid and Dataswyft’s Hub-of-All-Things (HAT) propose an alternative: Personal Online Datastores (pods). In this paradigm, users maintain ownership of their data, and applications must explicitly request permission for access. However, this shift creates a technical "gap." Traditional search and recommendation algorithms are designed for massive, unified data lakes. When data is fragmented across millions of private pods, performing efficient searches or providing relevant recommendations without compromising privacy becomes a primary research challenge.

Research Pillars and Methodology

The workshop identified three strategic pillars essential for bridging the gap between decentralization and functional discovery services:

  1. Societal & Regulatory Analysis: Investigating how decentralization impacts data governance, privacy legislation, and overall policy frameworks.
  2. Algorithmic Innovation: Moving beyond centralized indexing to explore Federated Learning (FL) and Structured Query Execution across distributed environments. This includes researching how Large Language Models (LLMs) can be integrated into federated recommendation systems.
  3. Infrastructure Development: Designing the underlying "plumbing" for a decentralized web, including peer-to-peer (P2P) overlay networks, innovative routing protocols, and information retrieval techniques that function without a central server.

Key Insights: The Rise of the Pod

A central theme of the workshop is the emergence of the "pod" as the fundamental unit of the next-generation web. This shifts the developer's focus from managing centralized databases to managing permissions and distributed queries.

However, decentralization introduces a "Performance Penalty." Overcoming these technical hurdles is not merely an optimization but a prerequisite for these systems to compete with centralized incumbents. Success requires a "confluence" of disciplines, including semantic web technologies, large-scale knowledge graphs, and legal expertise. Furthermore, the trend toward "Federated Everything"—specifically Federated Learning on Solid Servers—suggests a future where AI models are brought to the data, rather than harvesting data for a central model.

Workshop Program and Thematic Focus

The workshop brought together experts to discuss the roadmap for decentralized services. The sessions focused on secure search techniques, complex query execution across independent stores, and the robustness of distributed model training.

The DESERE workshop schedule highlights key sessions on federated learning, structured queries, and decentralized search.

As shown in the program above, the workshop featured diverse perspectives, ranging from technical deep-dives into structured query execution by Dr. Ruben Taelman to keynote talks on trustworthy federated learning by Prof. Han Yu.

Future Directions

The field of decentralized search and recommendation is currently in its early stages. The workshop identified several critical areas for future research:

  • Scalability: Developing more efficient routing and metadata propagation in P2P systems to match the scale of the modern web.
  • Algorithmic Parity: Addressing hurdles to ensure decentralized applications maintain the accuracy of their centralized counterparts.
  • Evaluation Standards: Creating robust benchmarks and performance evaluation tools specifically for decentralized architectures.
  • LLM Integration: Exploring the intersection of Large Language Models and decentralized pods to provide personalized experiences without mass data collection.

Summary updated

BibTeX
@inproceedings{ragab2024desere,
  author = {Mohamed Ragab and Yury Savateev and Wenjie Wang and Reza Moosaei and Thanassis Tiropanis and Alexandra Poulovassilis and Adriane Chapman and Helen Oliver and George Roussos},
  title = {{DESERE: The 1st Workshop on Decentralised Search and Recommendation}},
  booktitle = {The Web Conference},
  year = {2024},
  doi = {10.1145/3589335.3641302},
}
Figure