《电子商务 E-business》阅读文献：Learning with personalized recommender systems A psychological view

Computers in Human Behavior 28(2012)207-216 Contents lists available at SciVerse Science Direct Computers in Human behavior ELSEVIER journalhomepagewww.elsevier.com/locate/comphumbeh Learning with personalized recommender systems: A psychological view Jurgen Buder Christina Schwind Knowledge Media Research Center, Konrad-Adenauer-Str. 40. 72072 Tubingen, Germany ARTICLE INFO A BSTRACT This paper explores the potentials of recommender systems for learning from a psychological point of Available online 29 September 2011 view. It is argued that main features of recommender systems(collective responsibility, collective intelligence, user control, guidance, personalization) fit very well to principles in the learning sciences. However, recommender systems should not be transferred from commercial to educational contexts recommender systems n a one-to-one basis, but rather need adaptations in order to facilitate learn are discussed both with regard to learners as recipients of information and learn cers of data Moreover, it is distinguished between system-centered adaptations that enal In educational contexts, and social adaptations that address typical information tions for the design of educational recommender systems and for research on e 2011 Elsevier Ltd. All rights reserved. 1 Introduction unrated items can be predicted. A common method to predict pref erences is through collaborative filtering( Sarwar, Karypis, Konstan, When we ponder over the movie that we would like to see next Riedl, 2001) which mostly comes in two varieties: In user-based weekend, or whether the new restaurant in town is worth checking filtering, the behavioral profile of a target user will be compared to out, we often rely on the experience and recommendations of the profiles of other users, and recommendations for a particulai friends and other people who we trust to be knowledgeable about item will be derived from those users who are most similar to our tastes and preferences. Getting good recommendations be- the target user. The second method is item-based filtering where comes an important issue when the number of viable options is the overall rating differences among items will be set against the too large to be perused by an individual person. Internet servers profile of a target user to arrive at personalized recommendations provide access to vast amounts of information, and consequently. Personalized recommender systems are often used in offering recommendations is one of the most pressing problems e-commerce(Schafer, Konstan, riedl, 1999). as the ability to sug for the design of electronic environments. It can be said that search gest products that are tailored to the needs and preferences of cus- engines provide recommendations, as a list of search results is or- tomers provides a unique selling point. However, in recent years dered through link analysis algorithms that show most linked-to, the potential of personalized recommender systems for non-com- and thereby most relevant Web pages on top (brin Page, 1998). mercial purposes has begun to be explored, e.g. in educational con- Similarly, a bestseller list on a commercial Website can be regarded texts. Several educational recommender systems have been as providing recommendations. However, in these cases the rec- designed that recommend a broad range of items, among them soft- ommendations are generic, i.e. different users receive the same ware functionalities( Linton& Schaefer, 2000). learning resources e or highly similar output. In contrast, personalized recommender the Web(geyer-Schulz, Hahsler, Jahn, 2001; Recker, Walker, stems try to achieve the gold standard of recommendations in Lawless, 2003), Web 2.0 resources( Drachsler et al., 2010), foreign able about a topic, but also takes the individual tastes and prefer- McGrath, Ball, 2005), test items and assignments( Rafaeli, Barak, ences of users into account an-Gur,& Toch, 2004), lecture notes(Farzan Brusilovsk Personalized recommender systems capture the traces that 2005 ) or entire courses(Farzan Brusilovsky, 2006). The applica users leave in an environment, either through page visits or expli- tions cover very different areas of learning and education like use it ratings of items, and they are based on the assumption that of library systems(Geyer-Schulz, Hahsler, Neumann, Thede page visits or high ratings are indicative of user preferences. From 2003), informal learning( Drachsler, Hummel, Koper, 2009), m ta about visited or rated items, preferences on not-visited or bile learning(Andronico et al, 2003), learning at the workplace (Aehnelt, Ebert, Beham, Lindstaedt, Paschen, 2008), or within health education( Fernandez-Luque, Karlsen, Vognild, 2009). Corresponding author. Tel. +49 7071 979 326: fax: +49 7071 979 100. E-imail addresses: ibudereiwm-kmrc de (. Buder), schwind @iwm-kmrc de(c. Many papers on personalized recommender systems focus on technical issues and problems, the ultimate question being: How 0747-5632/s- see front matter o 2011 Elsevier Ltd. All rights reserved doi:10.1016/chb2011.09002

Learning with personalized recommender systems: A psychological view Jürgen Buder ⇑ , Christina Schwind Knowledge Media Research Center, Konrad-Adenauer-Str. 40, 72072 Tübingen, Germany article info Article history: Available online 29 September 2011 Keywords: Recommender systems Learning abstract This paper explores the potentials of recommender systems for learning from a psychological point of view. It is argued that main features of recommender systems (collective responsibility, collective intelligence, user control, guidance, personalization) fit very well to principles in the learning sciences. However, recommender systems should not be transferred from commercial to educational contexts on a one-to-one basis, but rather need adaptations in order to facilitate learning. Potential adaptations are discussed both with regard to learners as recipients of information and learners as producers of data. Moreover, it is distinguished between system-centered adaptations that enable proper functioning in educational contexts, and social adaptations that address typical information processing biases. Implica￾tions for the design of educational recommender systems and for research on educational recommender systems are discussed. 2011 Elsevier Ltd. All rights reserved. 1. Introduction When we ponder over the movie that we would like to see next weekend, or whether the new restaurant in town is worth checking out, we often rely on the experience and recommendations of friends and other people who we trust to be knowledgeable about our tastes and preferences. Getting good recommendations be￾comes an important issue when the number of viable options is too large to be perused by an individual person. Internet servers provide access to vast amounts of information, and consequently, offering recommendations is one of the most pressing problems for the design of electronic environments. It can be said that search engines provide recommendations, as a list of search results is or￾dered through link analysis algorithms that show most linked-to, and thereby most relevant Web pages on top (Brin & Page, 1998). Similarly, a bestseller list on a commercial Website can be regarded as providing recommendations. However, in these cases the rec￾ommendations are generic, i.e. different users receive the same or highly similar output. In contrast, personalized recommender systems try to achieve the gold standard of recommendations in real life by mimicking a person who is not only very knowledge￾able about a topic, but also takes the individual tastes and prefer￾ences of users into account. Personalized recommender systems capture the traces that users leave in an environment, either through page visits or expli￾cit ratings of items, and they are based on the assumption that page visits or high ratings are indicative of user preferences. From data about visited or rated items, preferences on not-visited or unrated items can be predicted. A common method to predict pref￾erences is through collaborative filtering (Sarwar, Karypis, Konstan, & Riedl, 2001) which mostly comes in two varieties: In user-based filtering, the behavioral profile of a target user will be compared to the profiles of other users, and recommendations for a particular item will be derived from those users who are most similar to the target user. The second method is item-based filtering where the overall rating differences among items will be set against the profile of a target user to arrive at personalized recommendations. Personalized recommender systems are often used in e-commerce (Schafer, Konstan, & Riedl, 1999), as the ability to sug￾gest products that are tailored to the needs and preferences of cus￾tomers provides a unique selling point. However, in recent years the potential of personalized recommender systems for non-com￾mercial purposes has begun to be explored, e.g. in educational con￾texts. Several educational recommender systems have been designed that recommend a broad range of items, among them soft￾ware functionalities (Linton & Schaefer, 2000), learning resources on the Web (Geyer-Schulz, Hahsler, & Jahn, 2001; Recker, Walker, & Lawless, 2003), Web 2.0 resources (Drachsler et al., 2010), foreign language lessons (Hsu, 2008), learning objects (Lemire, Boley, McGrath, & Ball, 2005), test items and assignments (Rafaeli, Barak, Dan-Gur, & Toch, 2004), lecture notes (Farzan & Brusilovsky, 2005), or entire courses (Farzan & Brusilovsky, 2006). The applica￾tions cover very different areas of learning and education like use of library systems (Geyer-Schulz, Hahsler, Neumann, & Thede, 2003), informal learning (Drachsler, Hummel, & Koper, 2009), mo￾bile learning (Andronico et al., 2003), learning at the workplace (Aehnelt, Ebert, Beham, Lindstaedt, & Paschen, 2008), or within health education (Fernandez-Luque, Karlsen, & Vognild, 2009). Many papers on personalized recommender systems focus on technical issues and problems, the ultimate question being: How 0747-5632/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.chb.2011.09.002 ⇑ Corresponding author. Tel.: +49 7071 979 326; fax: +49 7071 979 100. E-mail addresses: j.buder@iwm-kmrc.de (J. Buder), c.schwind@iwm-kmrc.de (C. Schwind). Computers in Human Behavior 28 (2012) 207–216 Contents lists available at SciVerse ScienceDirect Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh

u e. a number of users that atings to her. Once this set of neighbors is database,and

do we manage to deliver the most accurate recommendation for the current purposes? This paper, however, takes a somewhat dif￾ferent approach: It explores the psychological aspects of personal￾ized recommender systems, with the ultimate question being: How do people react to and act upon recommender systems? This question will be addressed with particular emphasis on the use of recommender systems in educational contexts. Knowing the psychological impact of recommendations on users can be helpful for practitioners and researchers alike. If we have a better idea how people react to recommender systems, we can improve algorithms and interfaces in ways that make using the system more efficient and satisfactory. Understanding how users contribute data to recommender systems is important for practitioners, as problems like low participation can impede sys￾tem performance. From a research perspective, a better under￾standing of the psychological impacts of recommender systems can inform various fields, such as educational psychology (instruc￾tional design, educational technology), social psychology (persua￾sion, trust building), business administration (marketing), or computer science (machine learning, HCI). The paper is structured as follows: Section 2 explores how the key characteristics of personalized recommender systems fit into current thought in the learning sciences. Section 3 discusses spe￾cific requirements that recommender systems must fulfill in order to support learning processes, both with regard to two learner roles and two types of adaptation. This discussion leads to four conjectures about how recommender systems should be adapted for educational contexts. Section 4 integrates the findings, and pro￾vides an outlook on future research. 2. Recommender systems and the learning sciences Designing and implementing workable recommender systems can be quite burdensome. Apart from a technological infrastruc￾ture that needs to store data about each possible combination of items and user, thereby generating substantial server load, a criti￾cal mass of users is one of the main roadblocks towards successful implementation (Glance, Arregui, & Dardenne, 1999). If the com￾munity of people who generate data is too small, recommenda￾tions become less precise. This begs the question of whether it is useful to implement personalized recommender systems in educa￾tional contexts. In order to answer this question, an example of a fictitious educational recommender system is introduced. This example will be used to illuminate main principles of recom￾mender systems design, and these principles will be compared to principles in the learning sciences. In our example, a psychology student is trying to find good re￾search literature for her Masters thesis. She logs into a digital li￾brary Website which operates a recommender system on academic publications. Let’s assume that she has never used the system before. The recommender might provide her with a list of the most popular publications on her thesis topic. This list would be similar to a bestseller list. Adjacent to each entry is a slider where she can rate each publication on a range from 1 (uninterest￾ing) to 5 (highly interesting). She reads through the list, and selects some publications that she knows. Interestingly, she dislikes some of the popular publications, and expresses this through low ratings. Though our student does not interact with other users of the rec￾ommender system, she is part of a larger community of others who have also selected and rated thousands of publications. As shown in Fig. 1, selecting entries and rating them constitutes the activity of individuals within the community. The recommender system then aggregates all the ratings from the community’s rating database, and filters this information according to specified algo￾rithms. For instance, if the recommender system employs user￾based collaborative filtering algorithms, one step is to define a so-called neighborhood for our student, i.e. a number of users that gave the most similar ratings to her. Once this set of neighbors is established, the system goes through all publications that our stu￾dent has not rated yet, and identifies those publications that re￾ceived the highest average ratings from the student’s neighborhood. In the recommender interface, the system provides an output of the top 10 publications; these items constitute the recommendations. As the student (and, by the use of similarity metrics, her neighborhood) have a non-standard taste, this list might differ strongly from the original, bestseller-like list. If a pub￾lication is recommended that the student does not know, she might order it. If she likes it (gives a high rating), she will become more similar to her neighbors; if the recommendation was bad and she gives a low rating, a new neighborhood might emerge, result￾ing in slightly different, adjusted recommendations. This ongoing cycle between individual activities (selecting, rating) and system activities (aggregating, filtering) rests on five principles of recom￾mender system design (see Fig. 1). First, recommender systems rely on collective responsibility. In our digital library example, the data on which book recom￾mendations are based were generated by a community of peers (Resnick & Varian, 1997). This is in contrast to offline contexts where recommendations often come from dedicated individuals like teachers, mentors, or reviewers. Recommender systems do not hand particular power to dedicated individuals, but shift responsibility and accountability towards the user collective. A similar principle of collective responsibility can be found in the learning sciences (Scardamalia, 2002) where many scholars have suggested moving from a traditional, teacher-centered education towards a peer-centered education (Brown et al., 1993). In peer￾centered education as well as in recommender systems, a power structure with flat hierarchies emerges. Moreover, in both fields it is assumed that peer efforts will lead to high-quality output: The learning from peer-centered education should be at least as high as in teacher-centered education; similarly, recommen￾dations derived from a community should at least be as good as those from dedicated experts. Second, recommender systems exhibit collective intelligence. For instance, if a particular book is recommended to our student, this recommendation cannot be traced back to the behavior of any individual user. Rather, it is the behavior of the user collec￾tive (or in the case of user-based collaborative filtering, the neighborhood) that is responsible for the recommendation. As it was shown empirically that computed recommendations are sufficiently correlated with the actual ratings of a user (Herlock￾er, Konstan, Borchers, & Riedl, 1999) it can be argued that these systems exhibit collective intelligence (Malone, Laubacher, & Dellarocas, 2009). This idea resonates with the notion of ‘‘group cognition’’ in the learning sciences, particularly in research on computer-supported collaborative learning (Stahl, 2006). Accord￾ing to this view, the output of a collaborative learning group, e.g. their discussions or the constructed artifacts, cannot be mean￾ingfully or completely traced back to individual group members, but rather arise through complex interactions among the constit￾uents (group members). It can be said that these emergent prop￾erties of groups can also be found in the way that recommender systems operate. Third, recommender systems are based on user control. A book that is suggested by a recommender system differs from a book that is a mandatory part of a course syllabus. Our student has the choice to follow the recommendation or not. Recommender systems preserve user autonomy, and they do not prescribe courses of action to be taken by a person. They typically support information search and retrieval, i.e. tasks of a self-directed, exploratory and often open-ended nature. In this regard, they cater to modern constructivist epistemologies in the learning sciences 208 J. Buder, C. Schwind / Computers in Human Behavior 28 (2012) 207–216

J Buder, C Schwind/Computers in Human Behavior 28(2012)207-216 Community 1. Collective responsibility Il. Collective intelligence Rating database System aggregate filter Individual Recommender interface V, Personalization IV Guidance lll, User control Fig. 1. Flow chart of the recommendation process. Principles of recommender systems are embedded. that also stress the importance of self-regulated learning 3. A psychological account of educational recommender (Boekaerts Minnaert, 1999), or discovery learning(Bruner, 1961). systems Fourth, recommender systems provide guidance. The student in the digital library example is not faced with a huge list of all pub Much attention on recommender systems has been devoted to lications on her thesis topic, but already receives a filtered list of issues of technical implementation, mathematical modeling, and ose titles that are most relevant to her search. By giving direc performance metrics(Adomavicius Tuzhilin, 2005 ) However. tions and offering hints that a user may or may not take into ac- there is a growing awareness that non-technical issues should be count, recommender systems are equivalent to an information taken into account in order to personalized recommender signpost(Konstan& Riedl, 2003). Providing guidance is also a cen- systems, particularly if these systems are applied in non-standard tral issue in the learning sciences as too much learner autonomy settings like education. Consequently, some authors began theoriz can be perceived as burdensome without some form of explicit ing about recommender system by including educational consider- or implicit structuring. As a consequence, principles in the learning ations(Drachsler, Hummel, et al, 2009: Manouselis, Drachsler, sciences often suggest using scaffolds(Vygotsky, 1978). scripts Vuorikari, Hummel, Koper, 2011: Tang McCalla, 2005: Wang (Kollar, Fischer, Hesse, 2006), or awareness functionalities 2007). The present paper also focuses on recommender systems (Engelmann, Dehler, Bodemer, Buder, 2009)in order to provide in educational contexts, but it is novel in taking a psychological uidance for self-regulated activities. The key is to strike a delicate point of view on the topic. relatively little is known about ho lance between autonomy and guidance so that guidance neither people react to and act upon information presented via recon becomes too immaterial nor too directive mender systems, and a psychological account might offer valuable Fifth, recommender systems are personalized, hints on barriers and potentials. gest items that are adaptively tailored to the nee and preferences of a user. As mentioned in the light on various issues that have to be taken into account when example, the recommendations for our student were not a gen- designing for educational recommender systems. The account is eric, bestseller-like list of most popular publications, but con- structured along a distinction that was made by Xiao and Benbasat sisted of items that were personalized with regard to her taste. (2007)in a conceptual paper on recommender systems in The notion of personalization also plays an important role in e-commerce contexts. However, while these authors put the tech- the learning sciences: Different learners do not benefit to the nology into the center by distinguishing between input character same degree from uniform types of instruction(Cronbach istics (data that a recommender system gets)and output Snow, 1977), and there is general consensus that instructional characteristics(data that a recommender system displays), we material should be adapted to the knowledge, the needs, and make the same distinction from a learner viewpoint. In other the abilities of learners. Consequently, learning technologies such words, our account distinguishes between a recipient role where as intelligent tutoring systems(Anderson, Boyle, Reiser, 1985) learners are confronted with recommended items and a producer or adaptive hypermedia environments(Brusilovsky, 2001)tailor role where learners generate data that are the basis for system information to the needs and abilities of learners. Recommender computations. The distinction between different roles(recipient systems are based on the same general idea by matching their vS producer)serves as a structural element for the remainder of output to a users historically developed profile this paper. For each role, two issues of recommender system adap Of course, the identified principles of the learning sciences- tation for educational contexts will be discussed The first issue re shifting responsibility towards peers, harnessing collective intelli fers to system-centered adaptations: In order to work properly gence,enabling user control, providing scaffolds, and tailoring to educational contexts, recommender systems must provide the needs, abilities, and interests- are embedded within many infor- right kind of information so that learning from recommendations mation technologies, but personalized recommender systems com- is enabled(recipient role). Moreover, proper functioning of recom- bine all of these principles. In this regard, they exhibit features that mender systems requires that user generate data on which system have the potential to leverage learning processes. computations can be performed(producer role). Apart from these However, the fit of recommender systems into learning con- basic, system-centered adaptations the second issue explored for texts by no means implies that they can be transferred from their recipient roles and producer roles pertains to social adaptation current, mostly commercial context into educational contexts on a Human information processing in general, and learning in particu one-to-one basis. Rather, they must be adapted to the peculiarities lar can be characterized by bounded rationality(Simon, 1959) of educational scenarios. Section 3 addresses the issues that have Navigation and selection of items in a recommender system(red to be taken into account in order to fruitfully apply recommender pient role)and rating of items (producer role) are influenced by a tems in the educational realm number of social psychological factors that can be linked to

that also stress the importance of self-regulated learning (Boekaerts & Minnaert, 1999), or discovery learning (Bruner, 1961). Fourth, recommender systems provide guidance. The student in the digital library example is not faced with a huge list of all pub￾lications on her thesis topic, but already receives a filtered list of those titles that are most relevant to her search. By giving direc￾tions and offering hints that a user may or may not take into ac￾count, recommender systems are equivalent to an information signpost (Konstan & Riedl, 2003). Providing guidance is also a cen￾tral issue in the learning sciences as too much learner autonomy can be perceived as burdensome without some form of explicit or implicit structuring. As a consequence, principles in the learning sciences often suggest using scaffolds (Vygotsky, 1978), scripts (Kollar, Fischer, & Hesse, 2006), or awareness functionalities (Engelmann, Dehler, Bodemer, & Buder, 2009) in order to provide guidance for self-regulated activities. The key is to strike a delicate balance between autonomy and guidance so that guidance neither becomes too immaterial nor too directive. Fifth, recommender systems are personalized, i.e. they sug￾gest items that are adaptively tailored to the needs, interests, and preferences of a user. As mentioned in the digital library example, the recommendations for our student were not a gen￾eric, bestseller-like list of most popular publications, but con￾sisted of items that were personalized with regard to her taste. The notion of personalization also plays an important role in the learning sciences: Different learners do not benefit to the same degree from uniform types of instruction (Cronbach & Snow, 1977), and there is general consensus that instructional material should be adapted to the knowledge, the needs, and the abilities of learners. Consequently, learning technologies such as intelligent tutoring systems (Anderson, Boyle, & Reiser, 1985) or adaptive hypermedia environments (Brusilovsky, 2001) tailor information to the needs and abilities of learners. Recommender systems are based on the same general idea by matching their output to a user’s historically developed profile. Of course, the identified principles of the learning sciences – shifting responsibility towards peers, harnessing collective intelli￾gence, enabling user control, providing scaffolds, and tailoring to needs, abilities, and interests – are embedded within many infor￾mation technologies, but personalized recommender systems com￾bine all of these principles. In this regard, they exhibit features that have the potential to leverage learning processes. However, the fit of recommender systems into learning con￾texts by no means implies that they can be transferred from their current, mostly commercial context into educational contexts on a one-to-one basis. Rather, they must be adapted to the peculiarities of educational scenarios. Section 3 addresses the issues that have to be taken into account in order to fruitfully apply recommender systems in the educational realm. 3. A psychological account of educational recommender systems Much attention on recommender systems has been devoted to issues of technical implementation, mathematical modeling, and performance metrics (Adomavicius & Tuzhilin, 2005). However, there is a growing awareness that non-technical issues should be taken into account in order to improve personalized recommender systems, particularly if these systems are applied in non-standard settings like education. Consequently, some authors began theoriz￾ing about recommender system by including educational consider￾ations (Drachsler, Hummel, et al., 2009; Manouselis, Drachsler, Vuorikari, Hummel, & Koper, 2011; Tang & McCalla, 2005; Wang, 2007). The present paper also focuses on recommender systems in educational contexts, but it is novel in taking a psychological point of view on the topic. Relatively little is known about how people react to and act upon information presented via recom￾mender systems, and a psychological account might offer valuable hints on barriers and potentials. In the following, we propose a conceptualization that sheds a light on various issues that have to be taken into account when designing for educational recommender systems. The account is structured along a distinction that was made by Xiao and Benbasat (2007) in a conceptual paper on recommender systems in e-commerce contexts. However, while these authors put the tech￾nology into the center by distinguishing between input character￾istics (data that a recommender system gets) and output characteristics (data that a recommender system displays), we make the same distinction from a learner viewpoint. In other words, our account distinguishes between a recipient role where learners are confronted with recommended items and a producer role where learners generate data that are the basis for system computations. The distinction between different roles (recipient vs. producer) serves as a structural element for the remainder of this paper. For each role, two issues of recommender system adap￾tation for educational contexts will be discussed. The first issue re￾fers to system-centered adaptations: In order to work properly in educational contexts, recommender systems must provide the right kind of information so that learning from recommendations is enabled (recipient role). Moreover, proper functioning of recom￾mender systems requires that user generate data on which system computations can be performed (producer role). Apart from these basic, system-centered adaptations the second issue explored for recipient roles and producer roles pertains to social adaptations. Human information processing in general, and learning in particu￾lar can be characterized by bounded rationality (Simon, 1959). Navigation and selection of items in a recommender system (reci￾pient role) and rating of items (producer role) are influenced by a number of social psychological factors that can be linked to Fig. 1. Flow chart of the recommendation process. Principles of recommender systems are embedded. J. Buder, C. Schwind / Computers in Human Behavior 28 (2012) 207–216 209

10 J Buder, C Schwind/ Computers in Human Behavior 28(2012)207-216 bounded rationality. For instance, we do not always attend to the fact that humans show preferences for particular types of informa- Ther munity could benefit from such an activity (DaM.. 2005): tion, and these inherent biases are not always conducive to information from which we learn most (tang Mccalla and we do not always contribute information even if an entire learning. Several ways of adapting recommender systems are explored that are based on ideas such as increasing the persua erefore, educational recommender systems can b siveness of recommendations, or providing counter-intuitive by introducing social adaptations that facilitate those inf recommendations processing biases that are conducive to learning or attenu biases that are detrimental to learning. These distinctions result in four structural elements(recipient 3.1.1. System-centered adaptation role vs producer role: system-centered adaptation vs social adap Whereas classical recommender systems in e-commerce try to on)for the following sections. In Sections 3. 1 and 3. 2 these is- adapt to the taste of a user, educational recommender systems will be discussed based on theoretical considerations as well should be personalized with regard to learner knowledge and as empirical results from various fields of research. Table 1 gives learning activities. For a number of reasons, learner knowledge an overview of the literature that informed the following and learning activities are more difficult to assess than user taste (Drachsler, Hummel, Koper, 2009): Learning is a gradual process extending over a longer stretch of time. In commercial contexts. effectiveness of a recommender system can be assessed by captur 3. 1. Recipient role ng whether a customer has purchased a recommended item. In contrast, learning does not have clear-defined and measurable Relatively little is known about how recommendations are per-"learning events"that immediately provide information about re ceived by users. Sections 3.1.1 and 3. 1.2 describe issues pertaining ommender system effectiveness. Not only are constructs like to the learners'roles as recipients of information. First, Section 3. 1. 1 knowledge and activities difficult to assess, they are also con- on system-centered adaptation addresses the fact that in classical stantly changing, and they rest on multiple sequential dependen- e-commerce scenarios recommendations are tailored to user taste cies, i. e at any given time there can be items that are too easy or (Schafer et al., 1999). In contrast, for educational contexts recom- too difficult for a learner. This creates numerous situational con- endations must be tailored to learner knowledge and learner straints: An expert in a domain needs different recommendations activities. Second, Section 3. 1.2 on social adaptation refers to the than a novice: different learning styles (e.g. reproducing of reviewed studies about recommender systems. Field Finding ent role and system-centered adaptation Computer science Drachsler, Hummel, Koper Conceptua tional technology Reflects on differences between recommenders for learning vs. commerce Drachsler. Hummel. van den Nadolski et al. (2009) 2 Educational technology Hybrid system leads to higher efficiency in learning Computer science/ Collaborative filtering and hybrid systems outperform no recommendations Recipient role and social adaptation McNee et al. (2006) Makes a case that personalities are ascribed to recommender Schwind et al. (2011a) Educational psychology N=123) lower evaluation Schwind et al.(2011b) pirical (lab experiments, Educational psychology eference- inconsistency reduces confirmation bias and leads to N=210) Tang and McCalla (2005) y endations are not always liked st(preference-inconsistency) Yoo and Gretzel (2011) Social psychology Discusses persuasion of recommender systems through source characteristics Producer role and system-centered adaptation Task transparency leads to higher acceptance(ma McNee et al.(2003 HCI ser control in sign-up increases loyalty(makes a case for explicit Schein et al. (2002 al (simulatio Computer science gues for implicit elicitation to overcome cold-start Xiao and Benbasat(2007) Conceptu ntroduces distinction between implicit vs. explicit elicitation cer role and social adaptation riment, N=268 Herlocker et al. (2004) HCI Makes a case that motivation for contribution can differ strongly Ling et ts,N=2715) Ludford et aL (2004) Social psychology Utility instruction increases rating activity xperiment, N=245) Rashid et al. (2006) Empirical (field cl social psychology Utility interface increases rating activity experiment, N= 160) Note: Classifications into type of study and findings are reported only as they pertain to this pa

bounded rationality. For instance, we do not always attend to the information from which we learn most (Tang & McCalla, 2005); and we do not always contribute information even if an entire community could benefit from such an activity (Dawes, 1980). Therefore, educational recommender systems can be improved by introducing social adaptations that facilitate those information processing biases that are conducive to learning or attenuate those biases that are detrimental to learning. These distinctions result in four structural elements (recipient role vs. producer role; system-centered adaptation vs. social adap￾tation) for the following sections. In Sections 3.1 and 3.2 these is￾sues will be discussed based on theoretical considerations as well as empirical results from various fields of research. Table 1 gives an overview of the literature that informed the following discussion. 3.1. Recipient role Relatively little is known about how recommendations are per￾ceived by users. Sections 3.1.1 and 3.1.2 describe issues pertaining to the learners’ roles as recipients of information. First, Section 3.1.1 on system-centered adaptation addresses the fact that in classical e-commerce scenarios recommendations are tailored to user taste (Schafer et al., 1999). In contrast, for educational contexts recom￾mendations must be tailored to learner knowledge and learner activities. Second, Section 3.1.2 on social adaptation refers to the fact that humans show preferences for particular types of informa￾tion, and these inherent biases are not always conducive to learning. Several ways of adapting recommender systems are explored that are based on ideas such as increasing the persua￾siveness of recommendations, or providing counter-intuitive recommendations. 3.1.1. System-centered adaptation Whereas classical recommender systems in e-commerce try to adapt to the taste of a user, educational recommender systems should be personalized with regard to learner knowledge and learning activities. For a number of reasons, learner knowledge and learning activities are more difficult to assess than user taste (Drachsler, Hummel, & Koper, 2009): Learning is a gradual process extending over a longer stretch of time. In commercial contexts, effectiveness of a recommender system can be assessed by captur￾ing whether a customer has purchased a recommended item. In contrast, learning does not have clear-defined and measurable ‘‘learning events’’ that immediately provide information about rec￾ommender system effectiveness. Not only are constructs like knowledge and activities difficult to assess, they are also con￾stantly changing, and they rest on multiple sequential dependen￾cies, i.e. at any given time there can be items that are too easy or too difficult for a learner. This creates numerous situational con￾straints: An expert in a domain needs different recommendations than a novice; different learning styles (e.g. reproducing Table 1 Overview of reviewed studies about recommender systems. Study Type Field Finding Recipient role and system-centered adaptation Adomavicius and Tuzhilin (2011) Conceptual (review) Computer science Introduces context-aware algorithms Burke (2002) Empirical (simulation) Computer science Compares different hybrid recommender algorithms Drachsler, Hummel, & Koper (2009) Conceptual Educational technology Reflects on differences between recommenders for learning vs. commerce Drachsler, Hummel, van den Berg, et al. (2009) Empirical (field experiment, N = 250) Educational technology Hybrid system leads to higher efficiency in learning Nadolski et al. (2009) Empirical (simulation) Computer science/ educational technology Collaborative filtering and hybrid systems outperform no recommendations Recipient role and social adaptation McNee et al. (2006) Conceptual HCI Makes a case that personalities are ascribed to recommender systems (basis for persuasion) Schwind et al. (2011a) Empirical (online experiment, N = 123) Educational psychology preference-inconsistency reduces confirmation bias, but leads to lower evaluation Schwind et al. (2011b) Empirical (lab experiments, N = 210) Educational psychology Preference-inconsistency reduces confirmation bias and leads to better elaboration Tang and McCalla (2005) Conceptual Educational technology Argues that educational recommendations are not always liked most (preference-inconsistency) Yoo and Gretzel (2011) Conceptual Social psychology Discusses persuasion of recommender systems through source characteristics Producer role and system-centered adaptation Kramer (2007) Empirical (experiments; N = 363) Marketing Task transparency leads to higher acceptance (makes a case for explicit ratings) McNee et al. (2003) Empirical (field experiment, N = 163) HCI User control in sign-up increases loyalty (makes a case for explicit ratings) Schein et al. (2002) Empirical (simulation) Computer science Argues for implicit elicitation to overcome cold-start Xiao and Benbasat (2007) Conceptual Marketing Introduces distinction between implicit vs. explicit elicitation Producer role and social adaptation Harper et al. (2007) Empirical (field experiment, N = 268) Social psychology Social comparison increases rating activity Herlocker et al. (2004) Conceptual HCI Makes a case that motivation for contribution can differ strongly Ling et al. (2005) Empirical (field experiments, N = 2715) Social psychology Goal setting and utility instructions increase rating activity Ludford et al. (2004) Empirical (field experiment, N = 245) Social psychology Utility instruction increases rating activity Rashid et al. (2006) Empirical (field experiment, N = 160) HCI/social psychology Utility interface increases rating activity Note: Classifications into type of study and findings are reported only as they pertain to this paper. 210 J. Buder, C. Schwind / Computers in Human Behavior 28 (2012) 207–216

16 211 ion)(Ent-2003). Moreover, involving learners into the very process of recom- mendations caters well to the spirit of learning as a constructive collaborative activity. This leads to our first conjecture: for re are in orde g activities. rT itiit h -osdo arners in of these two strategies ld be that learner nd affective reactions e design of lars hold nity to fee mendatior strategy is le n edu- hybrid rec strong ner involver ner control higher satisfac retzel,2011).For

orientation, achieving orientation, and meaning orientation) (Ent￾wistle, 1988) might require different recommendations; recom￾mendations might ideally take metacognitive skills and strategies (Weinstein & Mayer, 1986) into account; and they should be adapted to the goals of a learner (Boekaerts, 1998) – for instance, a learner who wants to find an explanation on a specific algebraic problem needs different recommendations than a learner who seeks for general resources on algebra. As Drachsler, Hummel, & Koper (2009) maintained, educational recommender systems would ideally be able to situationally identify those items that cor￾respond to a learner’s zone of proximal development (Vygotsky, 1978), the level of ability that the learner is able to master through scaffolding. In other words, in order to adapt to learner knowledge and learning activities, recommender systems must be context￾aware (Adomavicius & Tuzhilin, 2011). The classical approach of collaborative filtering through the analysis of simple ratings might not be very helpful, as a high rating could mean that a learner found an item easy, or challenging, or fun. However, there are dif￾ferent ways to achieve context-awareness. A first strategy to make recommender systems context-aware is to use machine intelli￾gence, e.g. through so-called hybrid recommender systems (Burke, 2002). These combine recommender algorithms like collaborative filtering with content-based filters and/or learning modeling tech￾niques (Brusilovsky, 2001). By way of our digital library example from Section 2, a publication recommender system could use a hy￾brid model that combines rating information with metadata. For instance, if our student gave a high rating for an article, the system could automatically increase the probability that other publica￾tions from the same author are recommended. Ontologies, tags and metadata can be used to describe learning items in more de￾tail, and modeling techniques can be used to describe learners in more detail. Having ontologies can also help to address the prob￾lem of sequential dependencies among learning items, and they might pave the way for systems that do not recommend isolated items, but actual learning paths (Drachsler, Hummel, & Koper, 2009). As to date, there are few examples of hybrid educational recommender systems that go beyond a prototype development, let alone a full system evaluation. However, in a detailed computer simulation study, Nadolski et al. (2009) found that different types of recommender systems yielded much better results (graduation percentages, user satisfaction, graduation times) than no recom￾mendations. Further, the authors found that hybrid recommender systems outperformed purely rating-based and purely ontology￾based recommender systems, although not by a significant margin. A real-world investigation of hybrid educational recommender systems compared a group using a hybrid personalized recom￾mender system for learning activities with a no-recommendation control group (Drachsler, Hummel, van den Berg, et al, 2009). In a usage study covering 4 months, they found that groups using the recommender system did not complete more activities, but completed them faster, exhibited a greater variety of learning paths, and expressed higher satisfaction. These examples show that hybrid educational recommender systems are likely to have measurable effects on learning-related variables. A second strategy to increase context-awareness does not rely on machine intelli￾gence, but on involving learners into the recommendation process. For instance, learners could choose among different learning paths depending on their learning styles or concrete learning goals. Moreover, dialogs could be provided that give learners an opportu￾nity to feed back on the situational adequacy of received recom￾mendations. While it appears that the learner involvement strategy is less popular among system designers than the use of hybrid recommender systems, it should be noted that active lear￾ner involvement might have additional benefits. For instance, lear￾ner control and customizability of system output are related to higher satisfaction and trust (McNee, Lam, Konstan, & Riedl, 2003). Moreover, involving learners into the very process of recom￾mendations caters well to the spirit of learning as a constructive and collaborative activity. This leads to our first conjecture: 3.1.1.1. How to achieve system-centered adaptation for recipi￾ents? Recommender systems must be context-aware in order to correctly diagnose learner knowledge and learning activities. This can be accomplished either through machine intelligence (hybrid recommender systems) or through involvement of learners into the recommendation process itself (customization, feedback loops). It is an empirical question which of these two strategies is superior, but a tentative conclusion could be that learner involvement has additional educational benefits. 3.1.2. Social adaptation Information processing has a social dimension: It is colored by attitudes, judgments, stereotypes, and affective reactions (Bandura, 1986). This lends a social dimension to the design of educational recommender systems as well. Some scholars hold that computers are social actors (Nass, Moon, Morkes, Kim, & Fogg, 1997), or that they are persuasive technologies that can exert social influence (Fogg, 2003). More specifically, the idea that recommender systems are perceived as social actors is sup￾ported by the observation that users ascribe a personality to them (McNee, Riedl, & Konstan, 2006). Personalized recom￾mender systems mimic a knowledgeable person, a person that does not only have information about a huge number of items, but also about the tastes and preferences of a user. However, we do not always follow a recommendation by a human being, and of course the same might apply to recommendations from a recommender system. This raises questions about the condi￾tions under which the selection of recommended items can be influenced. In order to answer these questions, it is helpful to re- flect on biases in human information processing. Some of these biases are conducive to learning and can be put to good use by making recommendations more appealing. Other biases in information processing are detrimental to particular types of learning, so recommender systems should be designed to over￾come these biases. We now turn to conducive biases in the con￾text of literature on persuasion, followed by detrimental biases in the context of selective exposure literature. Dual-process models of persuasion have outlined the boundary conditions that determine whether people are more or less in￾clined to follow a persuasive message such as a recommendation. According to the elaboration likelihood model (Petty & Cacioppo, 1986), the degree to which a persuasive message is elaborated de￾pends on a recipient’s motivation and ability to process the mes￾sage. Low personal relevance of the message topic undermines motivation, whereas distraction during processing impedes ability. If motivation and ability are high, messages are carefully scruti￾nized, and persuasion mainly depends on so-called message char￾acteristics; in contrast, if motivation and/or ability are low, persuasion mainly depends on so-called source characteristics (McGuire, 1969). As motivation and ability are not directly control￾lable, design of educational recommender systems should try to unfold persuasive power through message characteristics and source characteristics. As to message characteristics, the variable that is most often associated with them is argument strength. For instance, the elaboration likelihood model predicts that under conditions of high elaboration (high motivation and ability), a strong argument becomes persuasive, whereas a weak argument is likely to be rejected. As a consequence, the item pool of an edu￾cational recommender system should contain as many strong arguments as possible. A second way to influence the persuasive￾ness of a recommender system is through source characteristics, i.e. perceived attributes of a sender (Yoo & Gretzel, 2011). For J. Buder, C. Schwind / Computers in Human Behavior 28 (2012) 207–216 211