Hybrid Knowledge-Learner Joint Recommendation for Learning Partner Matching: Cold-Start Handling, Popularity Bias, and the Limits of Hard Contextual Filtering
1 Usmanu danfodiyo university sokoto
2 Usmanu Danfodiyo University sokoto
3 Federal University Birnin Kebbi
* Corresponding author: elmustty@gmail.com
2 Usmanu Danfodiyo University sokoto
3 Federal University Birnin Kebbi
* Corresponding author: elmustty@gmail.com
Abstract
Learning partner recommendation systems help online learners find compatible study partners, but graph-based approaches such as the Knowledge-Learner Joint Recommendation (KLJRec) system struggle with cold-start learners who have no interaction history and with popularity bias toward well-connected learners. This study evaluates a hybrid extension of a graph-based recommender that combines a GraphSAGE-style Graph Neural Network (GNN), a Profile Encoder that provides embeddings for learners without graph edges, contextual filtering on semester, major and preferred study time, and Maximum Marginal Relevance (MMR) re-ranking. Evaluation used a dataset of 206 learner records and 400 partnership edges privately collected from a distance learning institution, with ground truth defined as held-out test edges over five random 80/20 edge splits (mean ± standard deviation reported). Two flaws in the original evaluation were identified and corrected: a circular ground truth that guaranteed near-perfect precision, and an embedding objective without negative sampling that collapsed all embeddings. Under the corrected protocol, the original hard contextual filter reduced candidate lists to under two learners on average and lowered Recall@10 from about 0.04 to 0.003, a failure that was resolved by replacing the filter with a continuous contextual-compatibility score. With this change, the hybrid pipeline with MMR was statistically indistinguishable from a GNN-only baseline on Precision, Recall, F1 and NDCG (for example, Recall@10 of 0.040 versus 0.041) while lowering the Popularity Bias Index (2.98 versus 3.30). However, none of the learned models reliably outperformed random or popularity-only recommendation on this dataset, indicating that the available attributes and 320 training edges carry little learnable partnership signal. The study contributes a corrected evaluation protocol, evidence that exact-match contextual filtering is harmful, and an honest negative result that motivates validation on larger partnership datasets.
Keywords
learning partner recommendation
graph neural network
cold-start problem
popularity bias
maximum marginal relevance
References
- Bencito, G. R. L., Dabu, F. N., & Santiago, C. S., Jr. (2020). Development and evaluation of study-buddy pairing application for Android. Journal of Environmental Science, Computer Science and Engineering & Technology, 9(2), 230–237. https://doi.org/10.24214/jecet.B.9.2.23037
- Bouchet, F., Labarthe, H., Bachelet, R., & Yacef, K. (2017). Who wants to chat on a MOOC? Lessons from a peer recommender system. In Digital education: Out to the world and back to the campus (pp. 150–159). Springer. https://doi.org/10.1007/978-3-319-59044-8_17
- Bourkoukou, O., El Bachari, E., & El Adnani, M. (2017). A recommender model in e-learning environment. Arabian Journal for Science and Engineering, 42(2), 607–617. https://doi.org/10.1007/s13369-016-2292-2
- Carbonell, J., & Goldstein, J. (1998). The use of MMR, diversity-based reranking for reordering documents and producing summaries. In Proceedings of the 21st Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 335–336). ACM.
- Chounta, I.-A. (2019). Using learning analytics to provide personalized recommendations for finding peers. arXiv. https://arxiv.org/abs/1910.07381
- Elghomary, K., & Bouzidi, D. (2019). Dynamic peer recommendation system based on trust model for sustainable social tutoring in MOOCs. In 2019 International Conference on Smart Systems and Data Science (ICSSD). IEEE. https://doi.org/10.1109/ICSSD47982.2019.9003154
- Garg, M., & Goel, A. (2021). A data-driven approach for peer recommendation to reduce dropouts in MOOC. In Advances in computing and network communications: Proceedings of CoCoNet 2020 (Lecture Notes in Electrical Engineering, Vol. 735, pp. 217–229). Springer. https://doi.org/10.1007/978-981-33-6977-1_18
- Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Inductive representation learning on large graphs. Advances in Neural Information Processing Systems, 30.
- Hassan, M., & Hamada, M. (2016). Recommending learning peers for collaborative learning through social network sites. In 2016 7th International Conference on Intelligent Systems, Modelling and Simulation (ISMS) (pp. 60–63). IEEE. https://doi.org/10.1109/ISMS.2016.22
- Ma’ady, M. N. P. (2024). Strengthening research partner collaboration in higher education for searching innovation through machine learning-based recommender system. International Journal of Systematic Innovation, 8(1), 27–35. https://doi.org/10.6977/IJoSI.202403_8(1).0003
- Nguyen, T. T. A. M. (2020). Learning Partner Recommender System (LPRS): Promoting informal online learning communities in higher education [Doctoral dissertation, Monash University]. Monash University Bridges. https://doi.org/10.26180/12486329
- Oktavia, T., & Sujarwo, S. (2021). Interactive recommender system for identifying learning partner. International Journal of Emerging Technology and Advanced Engineering, 11(6), 78–86. https://doi.org/10.46338/ijetae0621_09
- Potts, B. A., Khosravi, H., Reidsema, C., Bakharia, A., Belonogoff, M., & Fleming, M. (2018). Reciprocal peer recommendation for learning purposes. In Proceedings of the 8th International Conference on Learning Analytics and Knowledge (pp. 226–235). ACM. https://doi.org/10.1145/3170358.3170400
- Riad, M., Qbadou, M., Aoula, E.-S., & Gouraguine, S. (2023). The new e-learning adaptation technique based on learner’s learning style and motivation. Journal of Education and Learning (EduLearn), 17(3), 472–482. https://doi.org/10.11591/edulearn.v17i3.20826
- Shao, M.-M., Jiang, W.-J., Wu, J., Shi, Y.-Q., Yum, T.-S., & Zhang, J. (2022). Improving friend recommendation for online learning with fine-grained evolving interest. Journal of Computer Science and Technology, 37(6), 1444–1463. https://doi.org/10.1007/s11390-021-2124-z
- Shou, Z., Chen, Y., Wen, H., Liu, J., Mo, J., & Zhang, H. (2024). Research on joint recommendation algorithm for knowledge concepts and learning partners based on improved multi-gate mixture-of-experts. Electronics, 13(7), Article 1272. https://doi.org/10.3390/electronics13071272
- Shou, Z., Shi, Z., Wen, H., Liu, J., & Zhang, H. (2023). Learning peer recommendation based on weighted heterogeneous information networks on online learning platforms. Electronics, 12(9), Article 2051. https://doi.org/10.3390/electronics12092051
How to Cite
Muhammad, M. M., Almu, A., Sani, H. M., Ahmed, A., & Gulumbe, A. U. (2026). Hybrid Knowledge-Learner Joint Recommendation for Learning Partner Matching: Cold-Start Handling, Popularity Bias, and the Limits of Hard Contextual Filtering. Journal of Studies in Science and Mathematics Education, 6(2), 373–384. https://doi.org/10.67203/jossme.2026.h4x5ddy3
M. M. Muhammad, A. Almu, H. M. Sani, A. Ahmed, and A. U. Gulumbe, "Hybrid Knowledge-Learner Joint Recommendation for Learning Partner Matching: Cold-Start Handling, Popularity Bias, and the Limits of Hard Contextual Filtering," Journal of Studies in Science and Mathematics Education, vol. 6, no. 2, pp. 373–384, November 2026. doi: 10.67203/jossme.2026.h4x5ddy3