Jakarta — Machine learning is transforming recommendation systems on e-commerce platforms, enabling them to better understand consumer behavior and provide more personalized product suggestions. This finding comes from a study by Stephen Gregorius Kurnia of Bina Nusantara University, Muhammad Rizki Perdana of Sekolah Tinggi Manajemen Informatika dan Komputer IKMI Cirebon, and Aldian Yusup of Institut Prima Bangsa Cirebon. Published in 2026, the study reviewed research on optimization strategies for machine learning-based recommendation systems in e-commerce. The researchers analyzed 10 selected scientific articles published between 2020 and 2025 and found that deep learning, hybrid recommendation models, sequential approaches, and large language model-based systems can improve recommendation accuracy and personalization.
The rapid growth of e-commerce has given consumers access to increasingly large product catalogs. While this provides more choices, it can also create information overload. Consumers may find it difficult to identify products that best match their needs and preferences when faced with thousands or even millions of products available online. Recommendation systems have therefore become an important technology for filtering information and presenting products that are more relevant to individual users.
Modern recommendation systems do not rely solely on product information. Machine learning can analyze browsing histories, product clicks, purchase transactions, reviews, ratings, and other user interactions. These data allow recommendation engines to generate more personalized suggestions and adapt to changing consumer interests. For e-commerce companies, this capability can also support customer engagement, retention, and revenue growth.
However, developing an effective recommendation system remains challenging. Problems such as data sparsity, cold-start situations, scalability, long-tail products, and rapidly changing consumer behavior can reduce recommendation quality. Systems must also process large volumes of information efficiently while maintaining fast response times. At the same time, companies must address privacy, algorithmic transparency, fairness, and ethical data use.
To examine these developments, Kurnia, Perdana, and Yusup conducted a Systematic Literature Review (SLR) using the PRISMA 2020 framework. Literature was collected from six academic databases: Scopus, ScienceDirect, SpringerLink, IEEE Xplore, ACM Digital Library, and Google Scholar. The search focused on publications from 2020 to 2025 and used keywords related to machine learning, recommendation systems, e-commerce, recommendation optimization, deep learning, and personalized recommendations.
The initial search produced 286 records. After removing 45 duplicate articles, 241 unique records remained for screening. A further 189 articles were excluded because they did not directly address machine learning-based recommendation systems in e-commerce. Of the remaining 52 articles assessed in full text, 42 were excluded because they did not meet the eligibility requirements. Ultimately, 10 studies were included in the final review. The complete selection process is presented in the PRISMA 2020 flow diagram on page 6.
The review found that machine learning has become a primary foundation of modern e-commerce recommendation systems. Traditional methods such as collaborative filtering and content-based filtering remain relevant, but recent studies increasingly use deep learning, hybrid models, graph-based learning, reinforcement learning, and large language model-based recommendation systems. These approaches allow recommendation engines to identify more complex relationships between users, products, and contextual factors.
One of the strongest findings concerns the effectiveness of hybrid recommendation systems. These systems combine multiple recommendation techniques so that the weaknesses of one approach can be addressed by another. The reviewed studies found that hybrid models can help overcome cold-start problems, data sparsity, and limited recommendation diversity. Deep learning-based optimization techniques were also found to improve precision, recall, and ranking performance.
User behavior is another important component. Recommendation systems can use browsing history, product clicks, purchase transactions, customer reviews, and navigation patterns to understand changing consumer preferences. Sequential recommendation and behavior prediction models allow systems to identify patterns in user activity over time. This means recommendations can respond not only to previous purchases but also to users’ current and evolving interests.
The study also highlights the growing role of large language models (LLMs), conversational AI, and multimodal recommendation systems. LLM-based systems can interpret user requests through natural language and consider conversational context when generating recommendations. Users can interact with recommendation systems more like virtual assistants, refining their preferences through dialogue.
Multimodal recommendation systems can go further by combining different types of information, including text, images, audio, and behavioral signals. By processing multiple forms of data, these systems have the potential to develop a more comprehensive understanding of consumer preferences and deliver more relevant and personalized recommendations.
For consumers, these developments can make online shopping more convenient. Instead of manually searching through enormous product catalogs, users can receive suggestions that are more closely aligned with their interests. For businesses, relevant recommendations can contribute to customer satisfaction, engagement, retention, and purchasing opportunities. The reviewed literature also indicates that behavior prediction models can improve recommendation relevance and purchase conversion rates.
However, increasingly sophisticated recommendation systems also introduce significant challenges. Data sparsity remains particularly problematic for new users and newly introduced products. Cold-start problems occur when systems lack sufficient interaction data to make reliable recommendations. Scalability is another concern because recommendation engines must process growing volumes of user and product data while maintaining real-time responsiveness.
Privacy, fairness, and transparency are also becoming increasingly important. Consumers and regulators want greater clarity about how recommendation algorithms generate suggestions and how personal data are collected and used. The researchers therefore emphasize that optimization should not focus exclusively on prediction accuracy. Recommendation systems also need to be transparent, fair, trustworthy, and responsible in their use of consumer data.
The researchers identify several directions for future development. Explainable recommendation systems could provide users with clearer reasons for why particular products are recommended, potentially increasing trust and acceptance. Reinforcement learning and real-time adaptive learning could make systems more responsive to changing consumer behavior. Meanwhile, LLMs and multimodal technologies could enable recommendation systems to understand more complex user intentions and preferences.
The study suggests that optimization has moved beyond simply improving prediction accuracy. Modern recommendation systems increasingly focus on personalization, behavioral analytics, real-time adaptation, natural-language interaction, and the integration of multiple data types. For e-commerce companies, these developments can support better customer experiences while strengthening competitiveness in an increasingly crowded digital marketplace.
Overall, the research by Stephen Gregorius Kurnia, Muhammad Rizki Perdana, and Aldian Yusup demonstrates that machine learning-based recommendation systems are becoming increasingly sophisticated. Deep learning, hybrid approaches, sequential recommendation, user behavior analytics, LLMs, conversational AI, and multimodal technologies are reshaping how e-commerce platforms understand consumers and present products. At the same time, businesses must address scalability, privacy, transparency, fairness, and computational efficiency to ensure that these systems remain effective and trustworthy.
Author Profiles
Stephen Gregorius Kurnia — Bina Nusantara University.
Muhammad Rizki Perdana — Sekolah Tinggi Manajemen Informatika dan Komputer IKMI Cirebon.
Aldian Yusup — Institut Prima Bangsa Cirebon.
Research Source
Article Title: Optimization of Machine Learning–Based Recommendation Systems on E-Commerce Platforms
Journal: East Asian Journal of Multidisciplinary Research (EAJMR), Vol. 5 No. 8, 2026, pp. 3519–3534.
DOI: https://doi.org/10.55927/eajmr.v5i8.284
Journal Website: https://journaleajmr.my.id/index.php/eajmr
0 Komentar