International Journal of Mathematical, Engineering, Biological and Applied Computing
Review Article | Open Access | 10.31586/ijmebac.2022.429

The Advances in Recommendation Systems – Theoretical Analysis

Raja Marappan1,* and S. Bhaskaran1
1
School of Computing, SASTRA Deemed University, Thanjavur, India

Abstract

Most people can't subscribe to every direct-to-consumer platform today, and the number is growing. The platform's content and the user's experience influence the decision to subscribe or buy. Today's consumers anticipate instantaneously curated content exploration, acquisition, and consumption. Media firms actively seek to increase both click-through rate and profitability by enhancing the user experience and enticing customers to subscribe or buy premium content through recommender systems. The direct-to-consumer platforms may maintain user engagement after consumers have visited the contents by providing suggestions that make the most of the site's rich content catalogs. By bringing it to the attention of viewers based on their viewing habits, for instance, effective recommendation systems might boost earnings for underappreciated long tail content. This research explores various recommender system types currently in widespread usage with an analysis of some of the fascinating breakthroughs.

1. Introduction

Content-based and collaborative filtering (CF) are two broad umbrella terms encompassing many of these kinds of systems. Even though it is one of the simplest algorithms, content-based filtering can be beneficial in some situations [1, 2]. Users' stated or implied tastes and information about product attributes, form the basis of this method. These suggestion systems are convenient to construct, but they have difficulty adapting to new users whose tastes are unknown, and their recommendations can feel stagnant.

User, item, and rating tuples form the foundation of CF. As a result, it takes advantage of users' collective knowledge rather than their content. One of the first companies to use this strategy, Amazon.com, released a paper on the topic that went on to win a "best of time" award from the IEEE. CF is based on the idea that users with similar likes (as determined by their past interactions with objects) are likelier to have similar interactions with new items [3, 4].

CF outperforms content-based filtering in terms of diversity (the degree to which recommended items differ from one another), serendipity, and novelty. But CF is more, costly, complex and computationally intensive to create and manage. Some CF techniques, such as factorization machines, are lighter than others. The inability to make suitable recommendations for new content without substantial interaction data to train a model is known as the "cold start problem," and it affects CF as well.

These two "traditional" types of recommenders aren't the only ones out there; many modern suggestion systems use neural net structures. In some cases, a form of "CF" is used. Others develop systems to deal with data, making choices from the user's activity history and the way that history reflects the user's changing interests. Recurrent neural networks (RNN) were the original foundation for such systems. Recently transformer-based models incorporate self-attention to discover interdependencies in user-generated behavior sequences [5, 6, 7, 8].

While non-deep learning models like factorization machines remain in use, neural networks are more data and computation-intensive. For instance, Amazon SageMaker's library algorithms support several features.

2. Related works

Researchers have tried many different methods for developing recommender systems in recent years. The hybrid systems are becoming more and more common. Some of the more recent strategies are not exclusive and can be used in tandem with more conventional methods. Personalized recommendation services like Amazon's Personalize are one such example. Amazon's user-personalization preferred algorithm ("recipe") blends a more modern bandit-based technique with hierarchical recommenders [9, 10, 11, 12].

2.1. Bandit-based systems

Integrating bandit-based strategies into recommender systems is a hot topic in academia. Bandit algorithms are a recommendation that aims to strike a balance between probing uncharted territory and cashing in on available opportunities. The real-time adaptability has made them a popular replacement for traditional testing. This has the potential to solve the issue of starting from cold. Bandit algorithms have found various uses in recommender systems, and some have even made their way into production-ready systems, like Amazon's Personalize. Indeed, bandit algorithms could be used to choose amongst competing recommender systems in real- time, taking into account user responses to the recommendations offered by each system. Bandits are finding more and more use in systems that maximize the happiness of a wide variety of constituents (a "marketplace" consisting of users, marketers, platform owners, content creators, etc.). One other goal of a music content recommender system could be to "level the playing field" by giving them some suggestions. Providers of media content such as Spotify have investigated this strategy, as detailed in an engaging public presentation by one of Spotify's researchers [13, 14, 15].

Several different bandit-based strategies are recently implemented. Amazon Personalize, as noted before, offers a completely managed solution for this. Amazon SageMaker is a self-service option that provides ready-made libraries and algorithms for quick experimentation. The contextual bandit's algorithm can generate suggestions based on users' actions, like whether they clicked on a specific suggestion or not. A sample notebook is provided in the companion piece [6, 7, 8, 9, 10].

2.2. Casual Interface

The goal of causal inference differs from that of classical statistics, which is to conclude relationships between variables under stable conditions; causal inference, on the other hand, is concerned with how and why these relationships might shift in response to outside influences such as interventions or alternative scenarios. On the other hand, a recommender system should not merely aim to replicate users' natural behavior but should also attempt to shape it. For this purpose, causal techniques are helpful, and they can often be implemented with very few adjustments to existing matrix factorization algorithms [7, 8].

An intervention calls for a causal inference question. Unobserved confounders, characteristics that influence which things users choose to interact with and how they rate them, are a significant difficulty in most casual inference questions. Other types of hybrid systems can be developed using causal inference. For instance, a group of Amazon researchers used this method to apply causal inference to a bandit-based system. By narrowing in on clients who can be swayed but otherwise wouldn't take action on their own, they discovered that tailored marketing yielded a higher return on investment when grounded in an understanding of causal effects [9].

2.3. Graph Neural-net based approaches

Graph Neural Networks (GNNs) differ from the strategies mentioned above since they are predicated on graph constructions that depict links between users and products. Given that there is no definite order to the objects a user could be interested in, GNNs may sometimes be preferable to sequence-based neural nets.

GNNs in recommender systems have been implemented using various architectural styles. The graph convolutional matrix-completion (GCMC) network [10] is an example of such an architecture. Matrix completion is expressed in a GCMC as a bipartite graph link prediction problem. The GCMC can then draw on other structured data sets, such as those found in social media, to inform its work. Performance constraints caused by the cold start problem can be alleviated when external information is coupled with interaction data.

The Deep Graph Package (DGL) is a helpful resource for anyone looking to use GNNs in AWS recommender systems. It is an open-source library designed to make GNN implementation simple. To utilize DGL with Amazon SageMaker, from Amazon (a fully-managed Docker container registry) as a deep learning container.

3. Conclusions

The field of recommender systems has grown substantially in the decades. While this is great for flexibility, picking a system for specific needs is much more challenging.

The following are the research directions in the recommendation systems [16, 17, 18, 19, 20]:

  • What business targets and measurements are needed to evaluate the system efficiency?" are just the tip of the iceberg. Diversity, serendipity, and originality can be taken into account in addition to the specific measures and coverage.
  • How can we solve the "cold start" issue that occurs with brand new users and products?
  • What is the optimal prediction lag time (and perhaps the maximum tolerable training duration)?
  • Whether the models can be scaled, and what kind of hardware is needed to enable training and servicing of the models efficiently. Again, this is highly dependent on the model's complexity and will significantly impact the final price tag.
  • How well can the model be explained? For many in the professional world, this is a must-have.

There may be a lot of choices to make during implementation as well. Amazon Personalize combines numerous ideas from current research to ease this decision-making process and relieves the burden of administering a recommender system at scale on data science and developer teams. These methods can be applied in Amazon SageMaker instead of Amazon Personalize if that service is inappropriate. Amazon SageMaker comes with proper prebuilt, open-source containers for contextual bandits, the DGL, and frameworks like TensorFlow and PyTorch.

References

  1. Greg Linden, Brent Smith, and Jeremy York, Amazon.com Recommendations: Item-to-Item Collaborative Filtering, IEEE Internet Computing, January-February 2003, retrieved from https://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf[CrossRef]
  2. Larry Hardesty, The history of Amazon’s recommendation algorithm, Amazon Science, November 2019, https://www.amazon.science/the-history-of-amazons-recommendation-algorithm
  3. Qiwei Chen, Huan Zhao, Wei Li, Pipei Huang, and Wenwu Ou, Behavior Sequence Transformer for E-commerce Recommendation in Alibaba, May 2019, https://arxiv.org/abs/1905.06874[CrossRef]
  4. Bhaskaran, S., Marappan, R. Design and analysis of an efficient machine learning based hybrid recommendation system with enhanced density-based spatial clustering for digital e-learning applications. Complex Intell. Syst. (2021). https://doi.org/10.1007/s40747-021-00509-4[CrossRef]
  5. Yifei Ma, Balakrishnan (Murali) Narayanaswamy, Haibin Lin, and Hao Ding, Temporal-Contextual Recommendation in Real-Time, KDD ’20: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, July 2020, https://dl.acm.org/doi/10.1145/3394486.3403278
  6. Marappan, R., & Bhaskaran, S. (2022). Analysis of Network Modeling for Real-world Recommender Systems. International Journal of Mathematical, Engineering, Biological and Applied Computing, 1(1), 1–7. DOI: 10.31586/ijmebac.2022.283[CrossRef]
  7. Raja Marappan, S. Bhaskaran. (2022). Analysis of Recent Trends in E-Learning Personalization Techniques. The Educational Review, USA, 6(5), 167-170. DOI: http://dx.doi.org/10.26855/er.2022.05.003
  8. Rishabh Mehrotra, Personalizing Explainable Recommendations with Multi-objective Contextual Bandits, April 2019, retrieved from https://www.youtube.com/watch?v=KoMKgNeUX4k.
  9. Raja Marappan, S Bhaskaran. Datasets Finders and Best Public Datasets for Machine Learning and Data Science Applications. COJ Rob Artificial Intel. 2(1). COJRA. 000530. 2022.[CrossRef]
  10. Saurabh Gupta, Anna Luo, Bharathan Balaji, Siddhartha Agarwal, Vineet Khare, and Yijie Zhuang, Power contextual bandits using continual learning with Amazon SageMaker RL, August 2019, https://aws.amazon.com/blogs/machine-learning/power-contextual-bandits-using-continual-learning-with-amazon-sagemaker-rl
  11. Stephen Bonner and Flavian Vasile, Causal Embeddings for Recommendation, October 2018, Twelfth ACM Conference on Recommender Systems, https://arxiv.org/abs/1706.07639[CrossRef] [PubMed]
  12. S. Bhaskaran, Raja Marappan. (2022). Analysis of Collaborative, Content & Session Based and Multi-Criteria Recommendation Systems. The Educational Review, USA, 6(8), 387-390. DOI: http://dx.doi.org/10.26855/er.2022.08.009
  13. Yixin Wang, Dawen Liang, Laurent Charlin, and David Blei, Causal Inference for Recommender Systems, Fourteenth ACM Conference on Recommender Systems (RecSys ’20), September 2020, https://doi.org/10.1145/3383313.3412225[CrossRef]
  14. Raja Marappan, S. Bhaskaran. (2022). Movie Recommender Model Using Machine Learning Approaches. The Educational Review, USA, 6(7), 317-319. DOI: http://dx.doi.org/10.26855/er.2022.07.006
  15. Neela Sawant, Chitti Babu Namballa, Narayanan Sadagopan, and Houssam Nassif, Contextual Multi-Armed Bandits for Causal Marketing, October 2018, Proceedings of the 35th International Conference on Machine Learning, https://arxiv.org/pdf/1810.01859.pdf.
  16. Marappan, R. (2022). Open-Source Datasets for Recommender Systems Analysis. International Journal of Mathematical, Engineering, Biological and Applied Computing, 1(2), 49–51.[CrossRef]
  17. Rianne van den Berg, Thomas N. Kipf, and Max Welling, Graph Convolutional Matrix Completion, October 2017, https://arxiv.org/abs/1706.02263
  18. B. M. D. Ekstrand, J. T. Riedl, Collaborative Filtering Recommender Systems, 4(2011).[CrossRef]
  19. I. Portugal, P. Alencar, D. Cowan, The use of machine learning algorithms in recommender systems: A systematic review, Expert Syst. Appl. 97 (2018) 205–227[CrossRef]
  20. R. Goyal, A.K. Updhyay, S. Sharma, Trust Prediction Using Ant Colony Optimization and Particle Swar Optimization in Social Networks, Advances in Intelligent Systems and Computing: Proceeding (2019) 485–491.[CrossRef]
Article metrics
Views
238
Downloads
44
Citations
0

How to Cite

Marappan, R., & Bhaskaran, S. (2022). The Advances in Recommendation Systems – Theoretical Analysis. International Journal of Mathematical, Engineering, Biological and Applied Computing, 1(2), 52–55. Retrieved from https://www.scipublications.com/journal/index.php/ijmebac/article/view/429

Copyright

Copyright © 2023 by authors and Science Publications. This is an open access article and the related PDF distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

  1. Greg Linden, Brent Smith, and Jeremy York, Amazon.com Recommendations: Item-to-Item Collaborative Filtering, IEEE Internet Computing, January-February 2003, retrieved from https://www.cs.umd.edu/~samir/498/Amazon-Recommendations.pdf[CrossRef]
  2. Larry Hardesty, The history of Amazon’s recommendation algorithm, Amazon Science, November 2019, https://www.amazon.science/the-history-of-amazons-recommendation-algorithm
  3. Qiwei Chen, Huan Zhao, Wei Li, Pipei Huang, and Wenwu Ou, Behavior Sequence Transformer for E-commerce Recommendation in Alibaba, May 2019, https://arxiv.org/abs/1905.06874[CrossRef]
  4. Bhaskaran, S., Marappan, R. Design and analysis of an efficient machine learning based hybrid recommendation system with enhanced density-based spatial clustering for digital e-learning applications. Complex Intell. Syst. (2021). https://doi.org/10.1007/s40747-021-00509-4[CrossRef]
  5. Yifei Ma, Balakrishnan (Murali) Narayanaswamy, Haibin Lin, and Hao Ding, Temporal-Contextual Recommendation in Real-Time, KDD ’20: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, July 2020, https://dl.acm.org/doi/10.1145/3394486.3403278
  6. Marappan, R., & Bhaskaran, S. (2022). Analysis of Network Modeling for Real-world Recommender Systems. International Journal of Mathematical, Engineering, Biological and Applied Computing, 1(1), 1–7. DOI: 10.31586/ijmebac.2022.283[CrossRef]
  7. Raja Marappan, S. Bhaskaran. (2022). Analysis of Recent Trends in E-Learning Personalization Techniques. The Educational Review, USA, 6(5), 167-170. DOI: http://dx.doi.org/10.26855/er.2022.05.003
  8. Rishabh Mehrotra, Personalizing Explainable Recommendations with Multi-objective Contextual Bandits, April 2019, retrieved from https://www.youtube.com/watch?v=KoMKgNeUX4k.
  9. Raja Marappan, S Bhaskaran. Datasets Finders and Best Public Datasets for Machine Learning and Data Science Applications. COJ Rob Artificial Intel. 2(1). COJRA. 000530. 2022.[CrossRef]
  10. Saurabh Gupta, Anna Luo, Bharathan Balaji, Siddhartha Agarwal, Vineet Khare, and Yijie Zhuang, Power contextual bandits using continual learning with Amazon SageMaker RL, August 2019, https://aws.amazon.com/blogs/machine-learning/power-contextual-bandits-using-continual-learning-with-amazon-sagemaker-rl
  11. Stephen Bonner and Flavian Vasile, Causal Embeddings for Recommendation, October 2018, Twelfth ACM Conference on Recommender Systems, https://arxiv.org/abs/1706.07639[CrossRef] [PubMed]
  12. S. Bhaskaran, Raja Marappan. (2022). Analysis of Collaborative, Content & Session Based and Multi-Criteria Recommendation Systems. The Educational Review, USA, 6(8), 387-390. DOI: http://dx.doi.org/10.26855/er.2022.08.009
  13. Yixin Wang, Dawen Liang, Laurent Charlin, and David Blei, Causal Inference for Recommender Systems, Fourteenth ACM Conference on Recommender Systems (RecSys ’20), September 2020, https://doi.org/10.1145/3383313.3412225[CrossRef]
  14. Raja Marappan, S. Bhaskaran. (2022). Movie Recommender Model Using Machine Learning Approaches. The Educational Review, USA, 6(7), 317-319. DOI: http://dx.doi.org/10.26855/er.2022.07.006
  15. Neela Sawant, Chitti Babu Namballa, Narayanan Sadagopan, and Houssam Nassif, Contextual Multi-Armed Bandits for Causal Marketing, October 2018, Proceedings of the 35th International Conference on Machine Learning, https://arxiv.org/pdf/1810.01859.pdf.
  16. Marappan, R. (2022). Open-Source Datasets for Recommender Systems Analysis. International Journal of Mathematical, Engineering, Biological and Applied Computing, 1(2), 49–51.[CrossRef]
  17. Rianne van den Berg, Thomas N. Kipf, and Max Welling, Graph Convolutional Matrix Completion, October 2017, https://arxiv.org/abs/1706.02263
  18. B. M. D. Ekstrand, J. T. Riedl, Collaborative Filtering Recommender Systems, 4(2011).[CrossRef]
  19. I. Portugal, P. Alencar, D. Cowan, The use of machine learning algorithms in recommender systems: A systematic review, Expert Syst. Appl. 97 (2018) 205–227[CrossRef]
  20. R. Goyal, A.K. Updhyay, S. Sharma, Trust Prediction Using Ant Colony Optimization and Particle Swar Optimization in Social Networks, Advances in Intelligent Systems and Computing: Proceeding (2019) 485–491.[CrossRef]