Determining e-commerce product recommendation systems utilizing mcdm methods

Abstract

With the continuous advancement of technology, the number of companies operating in this sector is steadily increasing. Especially after the COVID-19 pandemic, people began to change their lifestyles and lean more towards the digital world. Prior to the pandemic, people used to go to cinemas to watch movies or physically visit shopping malls to find desired products, but with the pandemic, people started meeting their personal needs through digital platforms. This trend has led to the growth of the e- commerce sector along with entertainment sectors like music and internet streaming. E-commerce is a developing sector witnessing a constant increase in both the number of customers and sellers. The expansion of product ranges on e-commerce platforms with the increase in the number of sellers has contributed to the growth of the sector. Maintaining their positions in the market and expanding their market shares for e- commerce companies depend on their primary objectives of ensuring customer satisfaction and providing the infrastructure to serve more customers. Therefore, it is of great importance that customers can easily find the products they desire and that their experiences on the platform are smooth and efficient until the completion of the purchase. In the United States, frequent shoppers spend an average of 44 hours and 35 minutes per month on e-commerce sites. Factors hindering customers from completing their shopping experiences on e-commerce platforms include slow site speed, difficulty in finding desired products, return policies, and inadequate customer service. Thus, it is crucial that the products recommended by recommendation systems are preferable to customers. Recommendation systems are digital tools that utilize a set of algorithms, data analysis, and artificial intelligence to create suitable recommendations for customers. Nowadays, these systems learn from customer profiles and take product ratings into account to generate recommendations. However, with the increasing number of products and customers, the area of recommending desired products to customers has become one that needs to be developed and personalized. This study aims to reduce the time it takes for e-commerce customers to find the products they want to purchase online and to enhance their overall shopping experiences. Recommendation systems are developed based on whether customers are interested in the products suggested by the system and are then presented to the customer. Therefore, to achieve this goal, the techniques of e-commerce recommendation systems and the factors influencing customers' purchase decisions have been examined separately. Recommendation system techniques include (i) personalization ability of the system, (ii) ability to solve cold start problem, (iii) ability to recommend relevant products, (iv) real-time processing capability, (v) quality of the recommendation system, and accuracy of the recommendation system factors. Factors influencing customer purchase behavior are categorized into seven headings: (i) product images, (ii) product attributes, (iii) customer reviews, (iv) product photos shared by customers, (v) product score given by the customers, (vi) result of height, size, weight filtering options, and (vii) finding the searched product. According to literature research, e-commerce product recommendation system techniques include both traditional methods and other techniques that some e-commerce companies do not use. In this study, 8 recommendation system techniques were examined. These are; (i) content-based filtering technique, (ii) popularity-based filtering technique, (iii) user-based collaborative filtering technique, (iv) item-based collaborative filtering technique, (v) model-based collaborative filtering technique, (vi) social network collaborative filtering technique, (vii) retrieval augmented generation technique, (viii) hybrid approaches technique. Since recommendation systems develop based on customer behavior, separate surveys were presented to both e-commerce customers and experts familiar with recommendation system techniques. 73 e-commerce users were asked to make pairwise comparisons of the specified criteria, and the geometric mean of the given answers was calculated to determine the importance degree of the factors affecting customers' online clothing purchase behavior using the AHP method. The consistency ratio was found to be 0.004, and the model was found to be significant. Customer reviews were observed to be the most important criterion affecting purchasing behavior. The other criteria following in order of importance are, respectively, product photos shared by the customer, finding the searched product, product score given by the customer, result of height, size, weight filtering options, product attributes, and product images. For the pairwise comparison of the factors of recommendation system techniques that suggest purchasable products to customers, an online survey was conducted with 6 experts in this field. These 6 people are experts in recommendation system techniques across multiple sectors, including e-commerce. The ages of these 6 people range from 26 to 49, with 1 woman and 5 men. The experts have at least 5 years of work experience, and their positions in the company are senior, lead, and head. Their areas of work are product management and data science. The geometric mean of the experts' responses was calculated using the AHP method. The consistency ratio was found to be 0.023. According to the experts, the most important criterion among the factors of recommendation system techniques is the accuracy of the recommendation. The accuracy of the recommendation is the ability to provide recommendations based on customer needs and preferences. The other criteria following in order of importance are, respectively, quality of the recommendation system, personalization ability of the system, ability to recommend relevant products, real-time processing capability, and ability to solve the cold start problem. In the survey asked to the experts, they were asked how well the relevant option performs concerning the relevant criterion of the recommendation system technique that suggests purchasable products to customers. The arithmetic mean of the answers given by the 6 experts was calculated, and the TOPSIS method was applied. The weights assigned by the experts to the criteria were used when applying this method, and positive and negative ideal solutions were determined. As a result of calculating the relative closeness, the alternatives were ranked. According to the experts, the hybrid approaches technique was identified as the best alternative. This was followed by model-based collaborative filtering technique, retrieval augmented generation technique, user-based collaborative filtering technique, item-based collaborative filtering technique, content-based filtering technique, social network collaborative filtering technique, and popularity-based filtering technique. According to the study results, the accuracy, quality, and personalization capability of the recommendation system play an important role in recommendation system techniques. Customer reviews and product photos shared by customers significantly influence customer purchase behavior. Among recommendation system techniques, hybrid approaches are supported as the most suitable option. Hybrid approaches are techniques that can work by taking product reviews into account and can improve themselves. However, the implementation of hybrid approaches can be costly and difficult. The more data it includes, the better results it produces. The implementation of the technique is complex and may be time-consuming. Therefore, its applicability in small-scale companies may be more difficult compared to large-scale companies.