Towards an Ethical Recommendation Framework Dimitris Paraschakis Dept. of Computer Science Malm¨ o University SE-205 06 Malm¨ o, Sweden Email: dimitris.paraschakis@mah.se Abstract—The goal of our study is to provide a holistic view on various ethical challenges that complicate the design and use of recommender systems (RS). Our findings materialize into an ethical recommendation framework, which maps RS development stages to the corresponding ethical concerns, and further down to known solutions and the proposed user-adjustable controls. The need for such a framework is dictated by the apparent lack of research in this particular direction and the severity of consequences stemming from the neglect of the code of ethics in recommendations. The framework aims to aid RS practitioners in staying ethically alert while taking morally charged design decisions. At the same time, it would give users the desired control over the sensitive moral aspects of recommendations via the proposed “ethical toolbox”. The idea is embraced by the participants of our feasibility study. Index Terms—recommender systems, ethics, ethical recommen- dation framework I. I NTRODUCTION In the era of big data, when the vast information overload makes it challenging for consumers to locate the products or services they need, recommender systems (RS) offer a helping hand in filtering the data. The notion of recommendations is built on real-life experiences and therefore perceived by humans as something inherently positive. User studies indeed show that the mere fact of labelling items as “recommenda- tions” increases their chances of being consumed [8], [23]. Whenever this fact is exploited for reasons beyond serving user needs, an ethical problem arises. Formally, ethics can be defined as “the study of morality” [47], where morality in turn can be defined as “the system whose purpose is to prevent harm and evils” [16]. As we show later, in the context of recommender systems these “evils” may include privacy intrusion, identity theft, behavior manipulation, discrimination, offensive / hazardous content, misleading information, etc. Thus, we can define recommendation ethics as the study of the moral system of norms for serving recommendations of products and services to end users in the cyberspace. This system must account for moral implications stemming from both the act of recommending per se, and the enabling technologies involved. A holistic view on recommendation ethics is currently lacking in the field of RS, despite the massive research that it attracts nowadays. According to the recent study by Tang & Winoto [46], there exist only two publications ([39], [43]) that specifically address the problem of ethical recommendations. Still, they only focus on particular problems in particular applications. Concerns over the lack of holistic approach have also been brought up by Friedman et al. [15] who studied the privacy aspects of recommender systems - the subject that has drawn the most attention in the ethical discourse around the practices of big data. Another recent paper by Koene et al. [24] points to the striking research imbalance in the area of personalized RS. The authors note that the strong emphasis on the commercial success of recommender systems contrasted with the considerable neglect of moral values, has a potential risk of a future public backlash against this research area. Our study is also in line with the rapidly growing attention to Fairness, Accountability, and Transparency (FAT) in machine learning. Recommendation ethics is a multifaceted problem, which relates to several interconnected topics that we broadly group into the ethics of data manipulation (privacy, anonymity, cen- sorship issues), algorithm design (algorithmic biases, behav- ior manipulation, discrimination issues), and experimentation (fairness, awareness, informed consent issues). In Section II, we touch upon all these topics in relation to recommender systems. In Section III, we outline an ethical recommendation framework that serves two purposes: a) it provides a roadmap for an ethics-inspired design of RS; b) it proposes a toolbox for manual tuning of morally-sensitive components of RS. We evaluate our proposal in Section IV by analyzing the results of the conducted survey. Section V concludes our work. II. THEORETICAL BACKGROUND: ETHICAL CHALLENGES This section discusses various ethical challenges around recommender systems that have been identified in the related literature. A. Data collection and filtering Both types of personalized information filtering used in recommender systems - collaborative and content-based - require the construction of a user profile, expressed in either item attributes (metadata), or user interactions (ratings, pur- chases, etc). Unique user profiles make it possible to tailor recommended items to user’s individual preferences. Collecting behavioral data is often done in the absence of informed consent. Public surveys point to the great demand for “do not track” tools that would help users gain control over the data collection process [12]. Often, a notice about data logging is hidden inside the Terms of Service (ToS) [24], which users are expected to read and agree with. However, 978-1-5090-5476-3/17/$31.00 2017 IEEE