Impact of Recommender Systems on Consumers’ Purchase Intention

Businessman holding a cell phone with a cloud of notification icons like recommendations, comments, etc. on top of it.

“Being digitally connected substantially changes the way companies compete and succeed” (Jung, Kraft, 2017). With advancements in information technology, e-commerce plays an important role in our daily lives and provides consumers with a great deal of convenience. However, consumers tend to become overwhelmed by the ever-growing amount of product-related content they receive, and suffer from information overload, which may lead to poor decision making. To cope with this information overload, recommender systems are increasingly becoming a key tool in e-commerce. The main feature of recommender systems is their ability to analyze user behavior, especially which item a particular user might be interested in, and further provide users with item suggestions according to their preferences and behaviors (Lu, Wu, Mao, Wang, & Zhang, 2015, p. 12). To generate recommendations, recommender systems follow the process that consists of three phases. In the first phase, the system collects extensive user information to gain a deeper understanding of the target users and lay a solid foundation for future phases. This input can be either explicit feedback, such as product ratings, or implicit feedback, such as purchase history. In the second phase, the recommender system filters and utilizes the user data collected in the first phase by adopting a machine learning algorithm. Lastly, the system predicts and presents item recommendations to the users (Isinkaye, Folajimi, & Ojokoh, 2015, pp. 263-264).

Based on the data filtering and rating estimation methods, recommender systems are usually divided into the following categories: content-based, collaborative and hybrid filtering systems. In the content-based filtering technique, recommendation is generated by matching up the contents of items and the users’ profiles. The content of items is represented as a set of descriptions which areextracted from the features of items. The user profile consists of the information about the interests, preferences and needs of the user and is constructed by analyzing the features of items that the user has purchased or viewed in the past (Lops, Gemmis, & Semeraro, 2011, p. 75). Collaborative filtering considers the opinions of other users as important factor and proposes recommendations to a user based on the analysis of other users who have similar preferences (Sivapalan, Sadeghian, Rahnama, & Madni, 2014, p. 163). This approach requires historical data and past ratings of users and works by building a database made up of users’ preferences for items and searching the users who have more in common with the target user (Cheng, Wang, 2014, p. 290). Hybrid filtering was proposed by combining two or more recommendation techniques to achieve higher performance and to avoid the limitations of the individual techniques it combines (Isinkaye, Folajimi, & Ojokoh, 2015, p. 269). Most commonly, the hybrid filtering approach combines the strengths of the content-based and collaborative filtering approaches to optimize recommender systems (Bagherifard, Rahmani, Nilashi, & Rafe, 2017, p. 1777).

This study – conducted as part of a final thesis at Munich Business School – sets out to examine the factors related that affect consumers’ purchase intention toward the products suggested by recommender systems in online shopping and gain a holistic view of recommender systems from the consumers’ perspective. To this end, ourresearch model integrates the unified theory of acceptance and use of technology (UTAUT) model proposed by Venkatesh, Morris, Davis, & Davis (2003) with the internet users’ information privacy concerns (IUIPC) model proposed by Malhotra, Kim, & Agarwal (2004). The UTAUT model helps to understand users’ perceptions and behavioral intentions towards recommender systems, while the IUIPC model explores users’ privacy concerns regarding recommender systems. Furthermore, to better understand what makes a recommendation useful to consumers, this research model includes the three factors that influence performance expectancy, i.e., recommendation accuracy, novelty, and diversity.

Research Model Recommender System

Research Model (illustrated by the authors)

Hui-Chieh Chen Portrait
About Hui-Chieh Chen
Hui-Chieh Chen is a master's student in the International Business master's program at MBS and conducted the described study as part of her master's thesis supervised by Prof. Dr. Hans H. Jung and Prof. Dr. Alexander Suhm.
Avatar photo
About Prof. Dr. Hans H. Jung 50 Articles
Hans H. Jung has been holding the professorship for Marketing at Munich Business School since 2012. After his graduation, he worked several years as a manager and consultant for premium car manufacturers in Germany and abroad. Since 2011, he has been working for the management consulting agency UNITY AG as Senior Manager.
Avatar photo
About Prof. Dr. Alexander Suhm 4 Articles
Prof. Dr. Alexander Suhm studied mechanical engineering at the Technical University in Karlsruhe, Germany, and obtained his doctorate in 1993 with a dissertation on new product development methods. Afterward, he joined Softlab GmbH (BMW Group) as a management consultant for the automotive industry and for mechanical and plant engineering, leading the Automotive Management division before he was appointed Director BMW at Softlab Ltd. At this stage he was also responsible for founding the management consulting company Nexolab GmbH (BMW Group) which he led from 2001 until 2007. In 2007 Prof. Dr. Suhm joined UNITY AG as a partner in the automotive division. Until 2017 he was responsible for the branch in Munich and Managing Director of the Chinese subsidiary. Today he is working in supervisory and advisory board mandates, is coaching start-ups and teaches in the bachelor's and master's programs at Munich Business School.