[Paper Review] AFEL-REC: A Recommender System for Providing Learning Resource Recommendations in Social Learning Environments

Dominik Kowald, Emanuel Lacić+2 more|arXiv (Cornell University)|Aug 14, 2018

Recommender Systems and Techniques10 references5 citations

TL;DR

AFEL-REC is a scalable, real-time recommender system for social learning environments that leverages user interactions and social tags to improve learning resource recommendations. It demonstrates that tag-based collaborative filtering significantly enhances both recommendation accuracy and coverage compared to interaction-based methods.

ABSTRACT

In this paper, we present preliminary results of AFEL-REC, a recommender system for social learning environments. AFEL-REC is build upon a scalable software architecture to provide recommendations of learning resources in near real-time. Furthermore, AFEL-REC can cope with any kind of data that is present in social learning environments such as resource metadata, user interactions or social tags. We provide a preliminary evaluation of three recommendation use cases implemented in AFEL-REC and we find that utilizing social data in form of tags is helpful for not only improving recommendation accuracy but also coverage. This paper should be valuable for both researchers and practitioners interested in providing resource recommendations in social learning environments.

Motivation & Objective

• To develop a scalable, real-time recommender system tailored for social learning environments.
• To evaluate the effectiveness of social data—particularly user-generated tags—in enhancing recommendation accuracy and coverage.
• To demonstrate the feasibility of integrating diverse data types (metadata, interactions, tags) within a unified recommendation framework.
• To provide a flexible software architecture supporting multiple recommendation use cases in near real-time.

Proposed method

• AFEL-REC is built on a scalable software architecture using Apache Solr and Apache ZooKeeper for high-performance data storage and coordination.
• The system uses a REST-based service layer to expose recommendation APIs and ingest new data (e.g., interactions, tags) in real time.
• The Recommender Engine computes recommendations using similarity-based algorithms, including user-based collaborative filtering on interactions (CF_i) and tags (CF_t).
• The Recommender Customizer enables runtime tuning of algorithm parameters (e.g., neighborhood size) via configurable recommendation profiles.
• The system supports multiple recommendation use cases, including personalized recommendations based on user behavior and social tagging.
• Evaluation is conducted using data from the Didactalia social learning platform, with metrics including precision, MAP, nDCG, and coverage (C).

Experimental results

Research questions

• RQ1 Can a recommender system effectively leverage social tagging data to improve learning resource recommendations in social learning environments?
• RQ2 How does the inclusion of social tags affect the accuracy and coverage of recommendations compared to interaction-based collaborative filtering?
• RQ3 To what extent can a scalable, real-time architecture support diverse recommendation use cases in social learning platforms?
• RQ4 Can tag-based collaborative filtering outperform interaction-based filtering in terms of both accuracy and coverage?

Key findings

• The tag-based collaborative filtering approach (CF_t) achieved a precision@20 of 0.070, significantly outperforming the interaction-based method (CF_i) with 0.046.
• CF_t achieved a normalized Discounted Cumulative Gain (nDCG@20) of 0.044, compared to 0.032 for CF_i, indicating better ranking quality.
• CF_t improved coverage to 53%, surpassing CF_i’s 40%, demonstrating that social tags help reach more users, including those with sparse interaction histories.
• The unpersonalized MostPopular (MP) baseline achieved 100% coverage but only 0.007 precision@20, highlighting the trade-off between coverage and accuracy.
• The results confirm that social tagging data enhances both recommendation accuracy and coverage, making it valuable for improving personalization in social learning environments.
• The system’s scalable architecture enables real-time recommendations and supports dynamic configuration of recommendation algorithms through the Recommender Customizer.