Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RecWizard: A Toolkit for Conversational Recommendation with Modular, Portable Models and Interactive User Interface

Zeyuan Zhang, Tanmay Laud, Zihang He, Xiaojie Chen, Xinshuang Liu, Zhouhang Xie, Julian McAuley, Zhankui He

TL;DR

RecWizard addresses the fragmented landscape of Conversational Recommender System tooling by提供 a Hugging Face–based, modular, and portable framework. It introduces a two-tier CRS abstraction (modules and pipelines) with text-focused data flow and optional tensor-based communication to maintain modularity, plus INFO and DEBUG UI modes for development and evaluation. The paper highlights open-source integration with pretrained modules (e.g., UniCRS, ReDIAL-based components), a composite tokenizer approach, and HF-compatible push_to_hub workflows to streamline reuse and deployment. The toolkit aims to lower barriers to CRS experimentation, accelerate pipeline-level evaluation, and foster community contributions through interactive interfaces and extensible templates. The work envisions expanding pipelines, benchmarking, and online services to broaden practical impact in CRS research and deployment.

Abstract

We present a new Python toolkit called RecWizard for Conversational Recommender Systems (CRS). RecWizard offers support for development of models and interactive user interface, drawing from the best practices of the Huggingface ecosystems. CRS with RecWizard are modular, portable, interactive and Large Language Models (LLMs)-friendly, to streamline the learning process and reduce the additional effort for CRS research. For more comprehensive information about RecWizard, please check our GitHub https://github.com/McAuley-Lab/RecWizard.

RecWizard: A Toolkit for Conversational Recommendation with Modular, Portable Models and Interactive User Interface

TL;DR

RecWizard addresses the fragmented landscape of Conversational Recommender System tooling by提供 a Hugging Face–based, modular, and portable framework. It introduces a two-tier CRS abstraction (modules and pipelines) with text-focused data flow and optional tensor-based communication to maintain modularity, plus INFO and DEBUG UI modes for development and evaluation. The paper highlights open-source integration with pretrained modules (e.g., UniCRS, ReDIAL-based components), a composite tokenizer approach, and HF-compatible push_to_hub workflows to streamline reuse and deployment. The toolkit aims to lower barriers to CRS experimentation, accelerate pipeline-level evaluation, and foster community contributions through interactive interfaces and extensible templates. The work envisions expanding pipelines, benchmarking, and online services to broaden practical impact in CRS research and deployment.

Abstract

We present a new Python toolkit called RecWizard for Conversational Recommender Systems (CRS). RecWizard offers support for development of models and interactive user interface, drawing from the best practices of the Huggingface ecosystems. CRS with RecWizard are modular, portable, interactive and Large Language Models (LLMs)-friendly, to streamline the learning process and reduce the additional effort for CRS research. For more comprehensive information about RecWizard, please check our GitHub https://github.com/McAuley-Lab/RecWizard.
Paper Structure (14 sections, 3 figures, 2 tables)

This paper contains 14 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Design Principles
  4. Why Use RecWizard
  5. Effortless Usage
  6. Flexible Development
  7. User-Friendly Interface
  8. Conclusion and Future Work
  9. Discussion on CRS Toolkits
  10. Current CRS Models in RecWizard
  11. More Details about Toolkit Use Cases
  12. An Example for Using Existing Models
  13. A Coding Template for New Models
  14. More Examples about User Interface

Figures (3)

  • Figure 1: (a) The RecWizard architecture comprises pipeline and module levels. Text data flows between modules and the pipeline, while tensor data flows within modules after being processed by RecWizard tokenizers, ensuring RecWizard modularity and portability. (b) In INFO mode, the RecWizard example ChatGPT-expansion includes the ReDIAL-Rec recommender module and ChatGPT with prompts as the generator module. Users can select models, set basic arguments, and chat with RecWizard. (c) In DEBUG mode, using the UniCRS-ReDIAL model as an example, module-level timeline visualization and intermediate messages are enabled for debugging or in-depth demonstrations in the user interface.
  • Figure 2: Model selection in user interface.
  • Figure 3: Monitoring example.