USER PREFERENCES IN RECOMMENDATION ALGORITHMS The Influence Of User Diversity, Trust, And Product Category On Privacy Perceptions In Recommender Algorithms Laura Burbach Johannes Nakayama, Nils Plettenberg, Martina Ziefle & André Calero Valdez 12 th ACM Conference on Recommender Systems (RECSYS 2018) 2 von 14
Introduction User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Spread of Recommender systems Domains: • E-Commerce, tourism, e-learning, people recommendation, group recommendation, information retrieval, search, media and communications, health, news recommendation User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Different recommender systems • content-based recommendation, • collaborative filtering, • hybrid forms, • trust-based recommendation, • social recommendation. • Rely on: • data from endusers, • meta-data on items. User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Different recommender systems Novel approaches: • more accurate • more sensitive Risk: data-misuse data How do How much I get useful information do recommen- I want to share? dations? Trade-off : Useful recommendation vs. distrust ? User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Content-based Age Gender CSE Product-Type Recommender Collaborative algorithm Key Question Hybrid Social Trust-based Institution- based Privacy Disposition concerns to Trust Trust technical personal data Distrust How do different product types and user usage Data Fear diversity factors influence the acceptance of recommender systems? User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Core-Hypotheses Willingness to share individual information Content-based Age Gender CSE Product-Type Recommender Collaborative depends on: algorithm Hybrid • product-type Social Trust-based • user diversity factors Institution- based Privacy Disposition concerns to Trust Trust technical personal data Ø Preference for different recommender Distrust usage Fear Data systems User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Research Design Content-based Age Gender CSE Product-Type Recommender Collaborative algorithm Hybrid Social Trust-based Institution- based Privacy Disposition concerns to Trust Trust technical personal data Distrust usage Fear Data User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Content-based Age Gender CSE Method – scenario-based online-survey Product-Type Recommender Collaborative mobile phone Content-based Age Gender CSE Content-based book Product-Type Age Gender CSE algorithm Recommender Content-based Age Gender CSE Product-Type Collaborative Product-Type Recommender Recommender algorithm Collaborative Collaborative algorithm Hybrid algorithm Content-based Hybrid Age Gender CSE Hybrid Product-Type Hybrid Recommender Social Collaborative algorithm Social Trust-based Social Hybrid Trust-based Institution- Social Social based Trust-based Institution- Privacy Disposition based concerns to Trust Trust-based Trust technical Privacy personal data Disposition Institution- concerns to Trust Trust technical Distrust personal data Institution- based usage Trust-based based Distrust Fear Data Privacy Disposition contraceptives Privacy Disposition usage Data Fear concerns to Trust concerns to Trust Trust technical Trust technical personal data personal data Distrust Distrust usage Institution- User Preferences in Recommendation Algorithms | | Laura Fear Data usage Burbach, M. A. | | 03.10.2018 Fear Data based Privacy Disposition concerns to Trust Trust technical personal data Distrust usage Data Fear
Method – recommendation algorithms I would like products that are similar to be recommended to me. Content-based Recommendation I would like to be recommended products that users like me have liked. Collaborative Filtering I would like to be recommended products that are both similar and liked by other users. Hybrid Recommendation User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Method – recommendation algorithms Saving my preferences of other users whose recommendations I like may be used for better product recommendations. Trust-based Recommendation The connections with my friends from my social media profile may be used for future recommendations. Social Recommendation Strongly Disagree Slightly Slightly Agree Strongly disagree disagree agree agree 1 2 3 4 5 6 User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Sample (N = 197) Study Online Survey N 197 Men | Women 50.8 % | 49.2 % Age 18 – 62 years; M = 31.2 years Computer self-efficacy M = 3.93 ( SD = 0.81) Privacy Concerns fear M = 2.90 ( SD = 0.96) Privacy Concerns data usage M = 2.23 ( SD = 1.01) Institution-based trust technical M = 3.03 ( SD = 0.88) Disposition to trust M = 3.08 ( SD = 0.64) User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Mixed effects model Within-subject factors: • product categories • recommendation algorithms Between-subject factors: • user diversity factors User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Results – Effects of product category on algorithm Acceptance of algorithms by product 6 5 Avg. agreement Product ● ● ● ● 4 book ● ● contraceptive ● ● ● ● ● 3 mobile ● ● 2 ● ● ● ● ● 1 collaborative content hybrid social trust Algorithm Error bars indicate 95% confidence intervall (Within-Subject) User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Results – User diversity and Acceptance for recommendation • higher acceptance of recommendation ( : M = 3.1, : M = 2.78, p < .001) • • Age ( r = -.05, p < .01) • Computer self-efficacy ( r = -.13, p < .001) • Privacy concerns ( r = -.35, p < .001) • Trust in technical infrastructure ( r = .34, p < .001) User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Key Results & Discussion Content-based Age Gender CSE Product-Type Recommender Collaborative algorithm different acceptance patterns Hybrid Social Trust-based Institution- • The more sensitive the product, the higher the preference for based content-based and collaborative filtering Privacy Disposition concerns to Trust Trust technical personal data • Trust-based and social approaches = generally rejected Distrust usage Data Fear • No user-factors influence the preference for any algorithms User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Discussion – Limitations & Further Research • Users never saw a working system • Understanding of „real“ requirements? • Is the data required for good recommendations? • Cultural bias ☛ Future steps: § Translation to other cultural groups and other countries, § Explanations in recommendations, § Other methods e.g. Conjoint-studies, § More product categories. User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Thank you for your attention! Recommender systems should use the least amount of data and should give useful recommendations From a user perspective some products are more sensitive than others Different recommendation algorithms are accepted for different products Any questions? Contact: Laura Burbach, M.A. Human-Computer Interaction Center (HCIC) RWTH Aachen University 52074 Aachen, Germany burbach@comm.rwth-aachen.de
Scales Computer self-efficacy 1. Technical equipment is inscrutable and difficult to control. 2. Before I work on a task with the help of technical equipment, I try to solve it in a different way. 3. I really enjoy cracking a technical problem. 4. I often have the feeling that technical devices do what they want. 5. I solve many technical problems rather by luck. 6. Most technical problems are so complicated that it makes little sense to deal with them. 7. It mainly depends on me and my abilities whether I solve a technical problem or not. 8. Even if resistances occur, I continue to work on a technical problem. User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Scales Privacy Concerns fear 1. I enter my personal data unwillingly in general. 2. I am afraid that my data is being misused. 3. I am afraid that my data is not secure. 4. I have made bad experiences with sharing my data. Privacy Concerns data usage 1. I often do not understand why online services need certain data. 2. I do not know what happens with my data. 3. I do not know who has access to my data. User Preferences in Recommendation Algorithms | | Laura Burbach, M. A. | | 03.10.2018
Recommend
More recommend
