Events

Dr. Keongtae Kim
Associate Professor
Department of Decision Sciences and Managerial Economics
The Chinese University of Hong Kong

With the increasing adoption of recommendation systems (RSs) in the mobile commerce, academic research has increasingly paid more attention to the question of how the impact of mobile RSs on consumer decisions and sales differs from that of conventional PC-based RSs’. On one hand, information-overload and search costs are known to persist in mobile platforms due to technical characteristics and idiosyncrasies of mobile devices such as a small screen size (Ghose 2017). On the other hand, as mobile RSs increasingly provide location-based recommendations, a smaller number of local products and services appear in a consumer choice set, successfully limiting the adverse effects of information overload (Adamopoulos et. al., 2021). As academic research has provided conflicting views on the effectiveness of mobile RSs, the impact of mobiles RSs on local economy has largely been unanswered. This research attempts to fill such gap by studying the impact of mobile RSs on local restaurants in South Korea.

Prior work on (desktop) RSs has examined the impact of RSs on consumers at different purchasing stages (e.g., Lee and Hosanagar 2019). Broadly speaking, the literature has distinguished the impact of RSs into three stages—exposure (when a product is recommended to a consumer), view (when a consumer clicks on a product page), and purchase (when a consumer purchases a product conditional on view). We follow a similar methodology and examine how a mobile location-based RS affects the three stages of local restaurant choices. We then delve into whether RSs benefit restaurants that are already popular or help consumers find niche establishments that are less known but better meet users’ tastes.

To examine this question, we collaborate with a leading search engine company in South Korea and obtain a dataset on about 17,000 distinct local restaurants located in 10 major districts (‘dongs’) 1. In April 2018, the company launched a location-based recommendation service, a real-time place recognition and recommendation service offered via its mobile application. The service curates local dining places to users based on its proprietary AI system offering optimized recommendations according to the focal users’ current location and time, and personal attributes and tastes. Our data span two months before and size months after the official launch.

We then exploit the official launch of the location-based RS as a natural experiment in studying the causal impact of making product recommendations on consumer search and selection behavior (i.e., clicks and visits). Though the launch of a recommendation service is a national event that has affected all the registered establishments, its impact has not been shared equally. A restaurant that is located near other restaurants is less likely to be recommended because the service bases its recommendation on the location of a user. In particular, the service recommends restaurants within a 400-meter radius of a user. Hence, we define treatment group to be establishments that have fewer than the median number (i.e., 314) of restaurants in a 400-meter radius. Control group is, then, restaurants that have more than 314 restaurants within 400 meters. One immediate concern for such research design is that sparsely populated restaurants are located in economically depressed regions and the users in such regions are not comparable to users in more prosperous regions. Though we have no direct data on user demographics, our district-level data at least allows us to compare users within the same district. In fact, we show that observable characteristics such as the distance to the closest subway station are not different between treatment and control restaurants. We then leverage a difference-in-differences analysis to measure the impact of introducing location-based RS on consumer exposures, clicks and visits.

Contrary to the conventional findings in the literature that PC-based recommendation systems favor more popular items and create a rich-get-richer effect for already successfully products, our empirical findings highlight the positive impact of recommendation systems on niche restaurants by enhancing visibility and conversion. Specifically, we find that the national launch of the location-based RS makes new recommendations for previously unexposed restaurants rather than increases the number of recommendations on previously recommended restaurants. In addition, we further find that the location- based RS induces more consumers to visit niche restaurants with higher quality measure. Together, we conclude that a location-based mobile recommendation system can well promote under-represented products without sacrificing consumer welfare as long as niche suppliers provide enough information on their product quality.