RecBole 2.0: Towards a More Up-to-Date Recommendation Library

Abstract: In order to support the study of recent advances in recommender systems, this paper presents an extended recommendation library consisting of eight packages for up-to-date topics and architectures. First of all, from a data perspective, we consider three important topics related to data issues (i.e., sparsity, bias and distribution shift), and develop five packages accordingly: meta-learning, data augmentation, debiasing, fairness and cross-domain recommendation. Furthermore, from a model perspective, we develop two benchmarking packages for Transformer-based and graph neural network (GNN)-based models, respectively. All the packages (consisting of 65 new models) are developed based on a popular recommendation framework RecBole, ensuring that both the implementation and interface are unified. For each package, we provide complete implementations from data loading, experimental setup, evaluation and algorithm implementation. This library provides a valuable resource to facilitate the up-to-date research in recommender systems. The project is released at the link: https://github.com/RUCAIBox/RecBole2.0.

• The paper introduces RecBole 2.0, which integrates eight comprehensive packages and 65 new models to tackle issues like data sparsity, bias, and distribution shifts.
• It employs advanced techniques such as data augmentation, meta-learning, and fairness metrics to enhance the performance and reliability of recommendation systems.
• The library leverages modern architectures including Graph Neural Networks and Transformers, standardizing experimental protocols to foster reproducible research.

RecBole 2.0: Enhancements in Recommender Systems Research

The paper "RecBole 2.0: Towards a More Up-to-Date Recommendation Library" presents an extended version of the RecBole recommendation library, incorporating the latest advances in recommender systems research. The primary goal of this extension is to support emerging research topics and modern architectural models, which have gained significant traction in the field. RecBole 2.0 is designed to address both data-centric challenges and evolving model architectures in recommender systems, enhancing the reproducibility and comparability of experimental results. The library offers eight comprehensive packages, including 65 new models, addressing pressing issues such as data sparsity, bias, and distributional shifts.

Data-Oriented Packages

The extended library addresses three primary data-centric challenges in recommender systems: sparsity, bias, and distribution shift.

1. Data Augmentation (RecBole-DA): This package aims to tackle data sparsity by generating additional interaction samples. It includes models utilizing heuristic methods, model-based approaches, and hybrid strategies to densify user-item interactions effectively.
2. Meta Recommendation (RecBole-MetaRec): Meta-learning techniques are employed here to solve cold-start issues by learning from minimal data. The package includes models for prediction, parameterization, and embedding tasks, showcasing the meta-learning framework's flexibility in addressing these challenges.
3. Debiased Recommendation (RecBole-Debias): This package focuses on correcting biases in user-item interactions, with models targeting selection bias, popularity bias, and exposure bias. It emphasizes improving the accuracy of recommendations by reducing skewed data influence.
4. Fairness Recommendation (RecBole-FairRec): Different from debiased recommendation, this package concentrates on fairness from the user's perspective, ensuring equitable treatment across diverse user groups. It provides models and metrics tailored to measure and enforce fairness in recommendations.
5. Cross-Domain Recommendation (RecBole-CDR): This package addresses distribution shifts by supporting cross-domain recommendation scenarios. It includes models leveraging collective matrix factorization, representation sharing, and knowledge mapping, reflecting the diversity of strategies used to manage such shifts.

Model-Oriented Packages

RecBole 2.0 also advances model-centric developments by focusing on recent architectural trends:

1. Graph Neural Network Based Recommendation (RecBole-GNN): The package implements models utilizing graph structures to capture complex user-item interactions and sequential patterns. It covers collaborative filtering, sequential recommendation, and social recommendation tasks, demonstrating GNNs' versatility.
2. Transformer Based Recommendation (RecBole-TRM): Recognizing the widespread adoption of Transformer architectures, this package includes models tailored for sequential and news recommendation tasks. It capitalizes on the natural capacity of Transformers for handling sequence modeling and semantic extraction.

Specialized Application: Person-Job Fit

In addition to domain-agnostic packages, RecBole 2.0 introduces a specialized application package for the task of person-job fit, demonstrating the library's adaptability to specific industrial contexts. This package includes models designed for collaborative filtering, content-based filtering, and hybrid approaches, addressing the intricate nature of job recommendation systems.

Implications and Future Directions

The extension of RecBole into RecBole 2.0 holds significant implications for both practical applications and theoretical developments in recommender systems. By unifying the implementation of contemporary models and addressing foundational data challenges, RecBole 2.0 facilitates consistency and reliability in research protocols. This ongoing standardization effort is critical for the future landscape of AI research, where robust benchmarking is key to scientific progress. Furthermore, as the library continues to evolve, it is likely to incorporate emerging models and applications, expanding its impact across various domains beyond traditional recommendation scenarios.

RecBole 2.0 represents a substantial step forward in providing a comprehensive tool for the research community, enabling the exploration of modern challenges and solutions in recommender systems with efficiency and rigor. The library's ongoing development promises to keep it at the forefront of AI research, fostering innovation and collaboration.