"1. INTRODUCTION. An unprecedented alignment of forces in the United States—from President Obama to state governors to private foundations—is calling for America to regain lost educational ground by once again having the highest proportion of students graduating from college by 2020. Concurrently, however, educational funding is shrinking. How can we make large gains in student success while spending less? By leveraging the sort of intelligent, automated computer “recommendation engines” proven successful by companies like Amazon, Netflix, Pandora and Apple. South Orange County Community College District (SOCCCD), a two-college district in Southern California with 43,000 students, has created Sherpa, an academic recommendation engine that combines human expertise and predictive analytics to provide students with the right information at the right time to facilitate better academic decisions. Sherpa uses time, event, or locationbased “triggers” to deliver multimodal (email, SMS, voice, textto-speech, or Facebook announcements) personalized communications such as: Keywords: Sherpa, Recommendation Engines, Personalization, Student Success - Helping students find acceptable alternatives when their preferred courses are full - Targeting at-risk students for academic interventions. - Tailoring information about campus events to individual interests At this session, we discuss the compelling nature of personalized online services, outline our software development process and provide a live demonstration of the Sherpa system. 2. DEVELOPMENT. 2.1 Precursors. Previously, SOCCCD had developed MySite, an enterprise academic web portal, and My Academic Plan (MAP), an online academic planning tool that has been used by students to create over 107,000 academic plans since it went online in April 2007. Though MySite and MAP were successful, they, like nearly all systems in higher education, were passive in nature. We wanted a more proactive system capable of assisting students’ decisionmaking processes in a manner that would “nudge” them toward making better-informed academic decisions. Initially, Sherpa was envisioned as a proactive academic planning tool that would focus on course selection. However, the more we discussed such a system, the more expansive our vision became. We realized that if we built a platform rather than an isolated product--i.e., a recommendation architecture (see below) rather than a specialized system--it could provide guidance on a wide range of decisions including student services and specific instructional content. 2.2 Modalities. Though our two-college district has 43,000 students, the California Community College System, with 2.7 million students in 112 colleges, is the largest system of higher education in the United States. From the outset, we wanted to create a system capable of scaling up to serve millions of students. In addition, we wanted the system to be capable of delivering nudges using multiple communication modalities: our web portal, personalized RSS feeds, email messages, text messages, voice calls, text-tospeech audio, mobile device apps, or a custom Facebook application. 2.3 Nudges The term “nudge” was chosen deliberately to reflect the openaccess nature of community colleges, where students are rarely forbidden from taking any classes they desire. Sherpa includes three categories of nudges: Courses, Information, and Services. - The Courses module provides assistance in finding open course sections during class registration. Currently, its decision rules are codified by human subject matter experts; soon, other rules will be generated by data mining legacy data in order to base course recommendations on the academic performance of academically-successful students with similar college transcripts. - Information channels provide data feeds to students based on whether their personal attributes match attributes the author of the information felt would be relevant. - The Services module presents students with personalized links to online services such as course registration, book purchasing, or Matriculation. Nudges are created by subject matter experts using Boolean operators. First, a target population is created by concatenating rules (e.g., [At-Risk Athletes = [Student Athlete] + [GPA < 2.0]). Next, trigger conditions are set. Then, a message to the target population is crafted. Nudges can be print, audio, or video-based. Delivery of nudges can be triggered by dates, actions, or locations. - Date-based messages can be set to be delivered on absolute (xx/xx/xxxx) or relative dates (e.g., “three days before this individual student’s registration appointment”). - Actions are triggered by data changes, such as the appearance of a student’s grade in the Student Information System (SIS) or a class status change. - Location-based services await completion of Phase II of our Mobile App project, whereupon students who opt-in to sharing GPS information from their mobile devices will be able to have nudges triggered by geographic location (e.g., a student walking by the library receives a text informing his/her that the book they requested through interlibrary loan has arrived). 2.4 Design Team and Methodology. SOCCCD utilizes the Agile SCRUM software development methodology, which maximizes user involvement and flexibility. User involvement is maximized by using including administrators, staff, faculty, and students on the development team; flexibility is aided by creating software via an iterative series of three-week “Sprints”, the object of each Sprint being a functioning module of software. The Sherpa team includes a vice president, dean, public information officer, administrative assistant, a technology support staff member, student services manager, outreach specialist, instructors, and most importantly, students. 2.5 Feedback Mechanisms. In addition to the broad constitution of the design team, other feedback mechanisms include Quality Assurance Testing, User Acceptance Testing, usability studies conducted with TechSmith’s MORAE software, focus groups, and an annual survey. In addition, each nudge is accompanied by a 1-5 star rating system and a comment box; the former is used to automatically rank nudges according to students’ perception of their relative importance and the latter is reviewed regularly to fine-tune nudges for clarity. 2.6 Training. Since the Sherpa project is driven by a focus on students, we thought it appropriate to have students introduce the system to their peers. After consideration, we rejected print-based training in favor of short videos available on our MySite web portal: - In this video, students from the Sherpa design team describe the kinds of problems they face in college and how Sherpa can help solve those problems. www.youtube.com/watch?v=hIZIvgwsHM - In this video, students introduce the changes to the MySite web portal necessitated by the Sherpa project. They don’t mention Sherpa in the video because the MySite portal acts as the web “face” of Sherpa: www.youtube.com/watch?v=-oMlahqo4iQ 3. RESULTS. - Sherpa helped students who were closed out of a class find an acceptable alternative class 6,606 times since its deployment in the Fall 2011 semester. - Sherpa is now used as the messaging engine for our custom-created student information system (SIS), and is Instead of generic announcements, announcements are personalized for each student and integrated with our MySite web portal, either by the student’s inclusion in the target set for a given nudge or by allowing students to opt-in to various communication “channels” (e.g., Admissions, Athletics, Matriculation)."