Educational Recommendation with Multiple Stakeholders Robin Burke DePaul University School of Computing Chicago, USA Email: rburke@cs.depaul.edu Himan Abdollahpouri DePaul University School of Computing Chicago, USA Email: habdolla@cdm.depaul.edu Introduction Education offers many opportunities for personalization and recommendation. Recent research has focused on person- alizing instruction for different learning styles, assisting teachers in assembling course materials, and other tasks [3]. One of the key characteristics of recommender systems research is an emphasis on personalization. Recommender systems are typically evaluated on their ability to provide items that satisfy the needs and interests of the end user. Researchers in educational recommendation have also made use of representations of learners’ knowledge state so that recommended materials can be used to correct misconcep- tions or convey missing concepts. In all of these formula- tions, the end user as the receiver of recommendations is, for the most part, the only consideration. We agree that such focus is entirely appropriate. Users would not flock to recommender systems if they believed such systems were not providing items that matched their interests. However, there are a variety of recommendation scenarios, including educational ones, where personalization to the user should not be the only consideration. For ex- ample, in reciprocal recommendation for online dating, the recommendations should be accepted to both participating users [5]. Another example is in digital advertising. The retrieval of a display ad in a real-time display advertising context depends not just on whether the ad is of interest to the user but, because advertisers pay for each impression, it also matters if the user is of interest to the advertiser [7]. Applications in Education Some educational settings share this multi-stakeholder char- acteristic. For example, in the iRemix educational social network, students are presented with a feed displaying media artifacts created by their peers for comment and reaction. Personalizing this feed based on prior activity is appropriate, but the system may also have the goal of equity, ensuring that all students get some feedback. Experimental results from [4] show that when students participate in online learning environment discussions and provide feedback to each other, they have more motivation for future creative activities. Students who do not receive feedback conversely have negative sentiments that can impact their future learn- ing. In this case, we can view the system as a stakeholder with an interest in ensuring equity. There are also educational contexts where reciprocal considerations come into play. For example, the Chicago City of Learning (CCoL) system is a catalog for the dis- covery of out-of-school time activities for high school and middle school students. Personalized recommendation and promotion of activities is helpful, so that students find out about programs of interest to them. In addition, however, the organizations offering such programs often have target “markets” in mind, much like advertisers. Consider the following example. A museum decides to offer a program in robotics targeted at 9th graders. The organizers are interested in gender equity in technology education and therefore consider a 50/50 ratio of girls and boys ideal. In addition, they know that some schools have the resources to offer enrichment activities that include robots and computer programming and others do not, so they want no more than 25% of the students to have prior exposure to these concepts. An ideal set of recommendations would therefore be ones likely to result in this distribution of students. If the system were to make recommendations only to boys with prior robotics exposure, it would not be helpful to the organization in meeting its goals, even if these users were the ones with the highest predicted ratings for such recommendations. Recommenation-based Promotion The problem of recommendation for multiple stakeholders was introduced in [1] where we discussed the importance of looking at recommender system as a multi-stakeholder environment and introduced the issue of maximizing social utility. We introduce here the challenge of representing mul- tiple stakeholders, and therefore, multiple utilities, within a recommender for the CCoL system. City of Learning acts as a middleman for a variety of educational experiences for students. On-line activities can, generally, be performed at any time, but off-line programs take place in a physical place where students must be present to participate. Space is generally limited and there may be more interested and eligible students than a program 2016 IEEE/WIC/ACM International Conference on Web Intelligence Workshops 978-0-7695-6039-7/16 $31.00 © 2016 IEEE DOI 10.1109/WIW.2016.18 62