Users' Resistance to Personalized Recommendations: Psychological Factors, and Possible Ways to Reduce It

Authors: Chen Yan, Wang Yong (Institute of Psychology, CAS)

Abstract--Personalized recommendation systems (PRS) are

promising applications on websites to recommend items to users

according to their unique preferences. Reseaches under the

priciple of computer sciences have developed numerous ways to

increase the accuracy of PRSs' prediction, but accuracy is not the

only indicator of the sucess of such a decison-aid tool. By digging

into the psychological factors relating to the user-system

interaction, the article discuss how three user factors - reactance ,

preference development, and innovativeness- could affact the

user acceptance of the system, and provides advices concerning

possible ways to reduce users' resistence to PRSs.

Keywords-Personalized recommendation system; user

resistance; psychology reactance; preference development;

innovativeness

I. INTRODUCTION

Recommendation systems (RS) are software agents that

recommend items (eg. products, services, or even people) to

match users' interests or preferences based on their particular

ranking algorithms[1]. They are now well welcomed by

websites. Best-seller lists, matching products

recommendations, or "persons you may know" by SNS

networks all can be defined as recommendation systems. One

reason for their widely adoption is that they may bring about

various benefits to the e-commerce websites. For example, they

could increase site sales and profits [2][3], and promote user's

impulse buying behavior [4]. They could also improve users'

satisfaction and loyalty [5][6].

There could be several ways to classify recommendation

systems. One of the approach is to classify them according to

the adaptivity of the recommendation lists, thus there could be

general recommendation and personalized recommendations

(PRS) [7]. The two recommendation systems both present lists

with recommended items, but with different ways to decide

which items to show. The general recommendation system, or

non-personalized recommend-ations, will recommend by

eliciting the interests or preferences of individual users' special

interest, preferences, and situations. In this way, different users

see different recommendation lists. The two types of

recommendation systems operate under different recommend

logic, and users' inner processing procedure may also differ.

This article should mainly discuss the issue related to PRSs.

II. FACTORS RELATED TO THE EFFICIENCY OF PRS

Existing studies mainly explore how to improve the

accuracy of the recommendations from the perspective of

computer science. At this point, accuracy becomes the main

concern of the system improvement work. However, the

overall process of the recommendation system and user

interaction is actually a complex decision-making process[8],

the accuracy of the system is just one of the antecedents to the

final result.

Anyway, researchers have started to investigate the impact

of the systems on users' inner psychological procedures such

as decision-making, and have found some of the factors

affecting this process. These factors can be roughly divided

into the system factors, and the user factors. System factors

include the type of the recommend algorithm[9], output quality

(see [10]), the nature of the item recommended, the nature of

the site launching the system[11], etc.; user factors include

users' preferences development level[9], involvement[8],

knowledge[12], gender[13], and perceived risk[14], etc.. The

joint effects of these factors will eventually lead to the

recommendation results- whether users are willing to follow

the recommendation to purchase, or to develop a sense of

satisfaction or loyalty to the system, or even the website that

launch the system.

For example, researchers suggest that users' preferences

may be constructed rather than retrieved[15]. The existence of

a PRS may change the way a user sees certain items if they

have been recommended, and the following confirmation

process might differ. Accuracy is just one of the system factors,

while user factors, such as preference development, will also

have influences. That's the reason why relying simply