Session-Based Recommendation Method Using Popularity-Stratified Preference Modeling
Abstract
:1. Introduction
2. Related Work
2.1. Session-Based Recommender Systems
2.2. KNN-Based Models
2.3. Neural Models
2.4. Objective Assessment of Recommender Systems
2.5. Runtime Analysis and Complexity Considerations
3. Methods
3.1. Propensity Score Calculation
- The propensity score is assumed to be independent of the user. This assumption, initially proposed by Yang et al., addresses the absence of extra user information in datasets of user–item interactions [12]. This assumption is retained in our approach, given that user information is typically absent in session-based contexts. Under this assumption, the propensity score for a specific item can be expressed as
- . This implies that a user’s preference for an item remains unaffected by whether it is recommended to them. Consequently, the likelihood of a user interacting with an item, given that it is recommended, is equal to the true probability of the user interacting with that item. As this probability does not depend on the user, the item’s true popularity directly determines it, denoted as :
- is proportional to , where is directly proportional to the item’s true popularity. This originates from a widely recognized framework of popularity bias introduced by Steck and is supported by empirical observations [13]. Additionally, considering that is from binomial distribution sampling with parameters defined by , it can express the item’s estimated propensity score as
3.2. SKNN
3.3. GRU4REC
4. Experiments
4.1. Datasets
4.2. Stratified Dataset Evaluation
- Stratification via the target item’s propensity: In this method, the propensity of the target item is directly adopted. However, in practice, our algorithm must not have access to the target item of an action until after the assessment of that action is finished. Therefore, while our approach is ideal, it somewhat violates the assessment procedure by revealing future data.
- Stratification according to the inclinations of historical items within the session: Instead of depending on information from the target item, we utilize data from all items that have been encountered earlier. A simple method of aggregating this information is by calculating the mean value of the propensity scores of the historical items within the session. Suppose this aggregated metric is correlated with the target item’s propensity. To illustrate this, Figure 2 demonstrates the empirical correlation between these metrics for each dataset.
4.3. Evaluation Metrics
- HitRate (HR) assesses the fraction of instances where the model’s predicted item aligns with the target item.
- The relevance of recommendations is assessed by Mean Reciprocal Rank (MRR) by computing the inverse rank of each recommended item and subsequently averaging this value across all actions.
4.4. Training and Evaluation
- For SKNN, set sample size = 500, cosine similarity, k = 100.
- For GRU4REC, Adagrad was employed as the learning algorithm; set the bpr-max loss function, momentum = 0.1, learning rate = 0.03, and dropout rate = 0.3 [6].
5. Results
5.1. Item Propensity Distribution
5.2. Evaluation Based on Propensity
6. Discussion
6.1. Ensemble Methods Based on Fixed Weight
6.2. Ensemble Methods Based on Dynamic Weight
7. Conclusions
Author Contributions
Funding
Data Availability Statement
Conflicts of Interest
References
- Jadidinejad, A.H.; Macdonald, C.; Ounis, I. The Simpson’s paradox in the offline evaluation of recommendation systems. ACM Trans. Inf. Syst. (TOIS) 2021, 40, 1–22. [Google Scholar] [CrossRef]
- Ludewig, M.; Mauro, N.; Latifi, S.; Jannach, D. Empirical analysis of session-based recommendation algorithms. User Model. User-Adapt. Interact. 2020, 31, 149–181. [Google Scholar] [CrossRef]
- Maurício, J.; Domingues, I.; Bernardino, J. Comparing Vision Transformers and Convolutional Neural Networks for Image Classification: A Literature Review. Appl. Sci. 2023, 13, 5521. [Google Scholar] [CrossRef]
- Saadatfar, H.; Khosravi, S.; Joloudari, J.H.; Mosavi, A.; Shamshirband, S. A New K-Nearest Neighbors Classifier for Big Data Based on Efficient Data Pruning. Mathematics 2020, 8, 286. [Google Scholar] [CrossRef]
- Jannach, D.; Ludewig, M.; Lerche, L. Session-based item recommendation in e-commerce: On short-term intents, reminders, trends, and discounts. User Model. User-Adapt. Interact. 2017, 27, 351–392. [Google Scholar] [CrossRef]
- Wang, H.; Du, Y.; Jin, C.; Li, Y.; Wang, Y.; Sun, T.; Qin, P.; Fan, C. GACE: Learning graph-based cross-page ads embedding for click-through rate prediction. In Neural Information Processing; Springer Nature: Singapore, 2024; pp. 429–443. [Google Scholar] [CrossRef]
- Ludewig, M.; Jannach, D. Evaluation of session-based recommendation algorithms. arXiv 2018, arXiv:1803.09587. [Google Scholar] [CrossRef]
- Hidasi, B.; Karatzoglou, A.; Baltrunas, L.; Tikk, D. Session-based recommendations with recurrent neural networks. arXiv 2016, arXiv:1511.06939. [Google Scholar] [CrossRef]
- Li, J.; Ren, P.; Chen, Z.; Ren, Z.; Ma, J. Neural attentive session-based recommendation. arXiv 2017, arXiv:1711.04725. [Google Scholar] [CrossRef]
- Wu, S.; Tang, Y.; Zhu, Y.; Wang, L.; Xie, X.; Tan, T. Session-based recommendation with graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, Honolulu, HI, USA, 27 January–1 February 2019; Volume 33, pp. 346–353. [Google Scholar] [CrossRef]
- Schnabel, T.; Swaminathan, A.; Singh, A.; Chandak, N.; Joachims, T. Recommendations as treatments: Debiasing learning and evaluation. arXiv 2016, arXiv:1602.05352. [Google Scholar] [CrossRef]
- Yang, L.; Cui, Y.; Xuan, Y.; Wang, C.; Belongie, S.J.; Estrin, D. Unbiased offline recommender evaluation for missing-not-at-random implicit feedback. In Proceedings of the 12th ACM Conference on Recommender Systems, Vancouver, BC, Canada, 2 October 2018. [Google Scholar] [CrossRef]
- Steck, H. Item popularity and recommendation accuracy. In Proceedings of the Fifth ACM Conference on Recommender Systems, RecSys 11, Chicago, IL, USA, 23–27 October 2011; pp. 125–132. [Google Scholar] [CrossRef]
- Hidasi, B.; Karatzoglou, A. Recurrent neural networks with top-k gains for session-based recommendations. In Proceedings of the 27th ACM International Conference on Information and Knowledge Management, Torino, Italy, 22–26 October 2018; pp. 843–852. [Google Scholar] [CrossRef]
- Rendle, S.; Freudenthaler, C.; Gantner, Z.; Schmidt-Thieme, L. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the UAI’09: 25th Conference on Uncertainty in Artificial Intelligence, Montreal, QC, Canada, 18–21 June 2009; pp. 452–461. [Google Scholar]
Dataset | Diginetica | RetailRocket | 30MU |
---|---|---|---|
Actions | 264 k | 210 k | 640 k |
Sessions | 55 k | 60 k | 37 k |
Items | 32 k | 32 k | 91 k |
Avg. Session Length | 4.78 | 3.54 | 17.11 |
Dataset | Metric | SKNN | GRU4REC |
---|---|---|---|
DIGINETICA | HitRate@20 | 0.3978 | 0.7007 |
MRR@20 | 0.1424 | 0.3455 | |
RetailRocket | HitRate@20 | 0.4351 | 0.6175 |
MRR@20 | 0.1383 | 0.2589 | |
30MU | HitRate@20 | 0.7293 | 0.7803 |
MRR@20 | 0.4187 | 0.6241 |
Dataset | Metric | SKNN | GRU4REC | 1.0 | 0.9 | 0.8 | 0.7 | 0.5 | 0.2 1 |
---|---|---|---|---|---|---|---|---|---|
DIGINETICA | HitRate@20 | 0.500 | 0.497 | 0.504 | 0.548 | 0.553 2 | 0.550 | 0.540 | 0.520 |
MRR@20 | 0.175 | 0.166 | 0.175 | 0.183 | 0.186 | 0.187 | 0.188 | 0.180 | |
RetailRocket | HitRate@20 | 0.548 | 0.551 | 0.550 | 0.581 | 0.581 | 0.578 | 0.577 | 0.565 |
MRR@20 | 0.311 | 0.300 | 0.320 | 0.326 | 0.326 | 0.325 | 0.323 | 0.310 | |
30MU | HitRate@20 | 0.355 | 0.4057 | - | - | 0.420 | - | - | - |
MRR@20 | 0.09 | 0.194 | - | - | 0.160 | - | - | - |
Dataset | Metric | Ensemble, Dynamic Weight | 1.0 | 0.8 | 0.5 1 |
---|---|---|---|---|---|
DIGINETICA | HitRate@20 | 0.556 2 | 0.504 | 0.553 | 0.520 |
MRR@20 | 0.186 | 0.175 | 0.186 | 0.180 | |
RetailRocket | HitRate@20 | 0.586 | 0.550 | 0.581 | 0.565 |
MRR@20 | 0.324 | 0.320 | 0.326 | 0.310 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Mo, Y.; Wang, H. Session-Based Recommendation Method Using Popularity-Stratified Preference Modeling. Mathematics 2025, 13, 960. https://doi.org/10.3390/math13060960
Mo Y, Wang H. Session-Based Recommendation Method Using Popularity-Stratified Preference Modeling. Mathematics. 2025; 13(6):960. https://doi.org/10.3390/math13060960
Chicago/Turabian StyleMo, Yayelin, and Haowen Wang. 2025. "Session-Based Recommendation Method Using Popularity-Stratified Preference Modeling" Mathematics 13, no. 6: 960. https://doi.org/10.3390/math13060960
APA StyleMo, Y., & Wang, H. (2025). Session-Based Recommendation Method Using Popularity-Stratified Preference Modeling. Mathematics, 13(6), 960. https://doi.org/10.3390/math13060960