Fostering a Learning Community in MOOCs Cindy E. Hmelo-Silver Carolyn P. Rosé Jeff Levy Indiana University Carnegie Mellon OfficeHours 543 Eigenmann Hall University5000 Forbes Avenue 817 W. Peachtree St Bloomington IN 47406 Pittsburgh, PA 15217 Atlanta GA 30308 1. Introduction • E4. Self-directed study A key hurdle preventing MOOCs from reaching their • E5. Evaluate their learning and performance. transformative potential is that they fail to provide a social Performance goals for facilitator environment that is conducive to sustained collaborative • P1. Keep all students active in the learning process engagement and learning. Correlational analyses from existing • P2. Keep the learning process on track MOOCs demonstrate a reliable connection between social • P3. Make the student’s thoughts and their depth of integration into the threaded discussions and course retention. The understanding apparent best evidence so far suggests an important connection between • P4. Encourage student reliance on selves and peers social support and retention consistent with findings in other types for direction and information. of online communities. If we can engineer MOOCs to achieve Strategies that can support these goals are shown in Table 1. greater success at providing affordances for sustained social Table 1. Example Strategies (adapted from [3]) engagement and learning, the potential for impact is immense. 2. Facilitating Small Groups Introducing facilitation into large scale threaded discussion in MOOCs requires extensions to existing wisdom on facilitation of small groups [1; 2]. We seek to understand how better to facilitate student-centered learning and student agency (and deep engagement) on a grander scale in online environments. We know a lot about strategies that can be used to support students in collaborative projects and problem based learning [3] but these have generally only been used on a small scale because they requires intense monitoring of the collaborative discourse for both collaboration and content (what we might term productive collaboration). Engineering facilitation requires attention at two levels: supporting the collaboration among learners, but also supporting facilitation for teaching assistants or other tutors. To accomplish larger scale facilitation, there are two important aspects to consider. One is identify markers of both collaborative activities and disciplinary content and/or practices. Some of these might include indicators such as making contributions, asking responding to questions, social network analysis, or indicators of transactivity. Learning analytics (LA) techniques are important to be able to accomplish this [4; 5]. The markers of disciplinary content and practices may be more difficult to measure. The second aspect is acting on what is For TA’s or any kind of automated agents to use these learned through LA. A first step might be a dashboard that might strategies, it will be important to have indicators so that whatever alert instructors as to where they might need to intervene as well the facilitation strategy needed, there is some way to first indicate as suggesting appropriate strategies. that attention is needed in a group. Some of this might be timed-- for example, having a group summarize where they are could be Research on these massive networked communities should done at intervals based on the ebb and flow of course assignments build on research on best practices in CSCL and problem-based learning (PBL; e.g., [3; 6-10]). This research suggests that certain and content difficulty. Some of the social prompts could be based instructional goals can support productive collaboration, and that on shallow indicators like participation and length of there are particular cognitive, metacognitive, and social strategies posts. Others would need more semantically oriented that can best serve particular kinds of goals. For example, in some indicators. For example, the first time that someone mentions or groups such as a PBL group, goals might include: proposes some complex idea, it would be reasonable to ask the group if they understand. Semantic indicators could compare what Educational Goals for students: is being discussed with key course objectives and past • E1. Construct causal explanations discussion. To the extent that these indicators can be identified, • E2. Employ effective reasoning and argumentation. these are other places where simple intervention might be quite productive. An area ripe for research is identifying these • E3. Identify knowledge limitations indicators, understanding what kinds of strategies might be generally useful and the extent to which these indicators are cognitive engagement. Conversely, measures of relationship loss amenable to computational solutions. predict higher dropout. Similarly, membership in subgroups where there is high attrition also predicts higher attrition. To In our computational work, we have been developing model relationship formation over time, we are experimenting techniques for extracting indicators within messages that tell us with probabilistic graphical models that combine text, social something about a student’s orientation at a time point (i.e., a network, and thread structure representation in order to identify week of participation within the course) towards the course, which coordinated group behavior that may indicate the formation of a will enable us to make a prediction about how likely it is that the subcommunity [13]. Beyond measures at the student level that student will drop out of the course on the next time point. The pick out students who are at risk and may need extra attention, we relevant levels of analysis are post, week, and trajectory within the use structural equation modeling techniques to identify the factors course. First, we extract indicators from individual that affect whether a thread that is started by a student who is messages. Then we aggregate messages within a week in order to reaching out for help will get a satisfactory response. In order to construct an indicator for a week. We then use a survival analysis reduce the load on instructors, we are developing matrix to model the probability that the student with drop out on the next factorization techniques to identify community members whose time point given the value of the variable. In this way, we can effort we may be able to enlist to respond to at risk threads [14]. identify risk factors that instructors should be aware of so that they can focus their efforts to support those students. Indicators Recent research in the learning sciences has begun to we have explored include social network analysis measures such address how to use LA productively [15]. At the same time, as authority and hub scores [2], indicators of social subgraph MOOCs can provide transformative opportunities for learners if membership [1] measures of motivation and cognitive we can identify and support learning communities within the engagement [11], and indicators of relationship formation and larger communities. Bringing the LA and learning sciences measures of loss of relationship due to attrition of other students communities together for discussion and joint efforts can provide [12]. Overall, we find that measures of high authority and hub opportunities to better understand how to facilitate learning scores as well as high numbers of formed relationships predict communities in this frontier. lower attrition, as do measures of high motivation and high 3. References [8] Chinn, C.A. and Clark, D.B., 2013. Learning through [1] Rosé, C.P., Carlson, R., Yang, D., Wen, M., Resnick, collaborative argumentation. In International L., Goldman, P., and Sherer, J., 2014. Social Factors Handbook of Collaborative Learning, C. Hmelo- that Contribute to Attrition in MOOCs. In Proceedings Silver, C.A. Chinn, C.K.K. Chan and A.M. O'Donnell of the First ACM Conference on Learning @ Scale. Eds. Routledge, New York, 314-332. [2] Yang, D., Sinha, T., Adamson, D., and Rosé, C.P., [9] Fischer, F., Kollar, I., Weinberger, A., Stegmann, K., 2013. Turn on, Tune in, Drop out: Anticipating Wecker, C., and Zottmann, J., 2013. Collaboration student dropouts in Massive Open Online Courses. In Scripts in Computer-Supported Collaborative Proceedings of the NIPS Data-Driven Education Learning. International Handbook of Collaborative Workshop (2013). Learning, Routledge, New York, 403-419. [3] Hmelo-Silver, C.E. and Barrows, H.S., 2006. Goals [10] Schwarz, B.B. and Asterhan, C.S., 2011. E- and strategies of a problem-based learning facilitator. Moderation of synchronous discussions in educational Interdisciplinary Journal of Problem-based Learning settings: A nascent practice. Journal of the Learning 1, 21-39. Sciences 20. [4] Mu, J., Stegmann, K., Mayfield, E., Rosé, C.P., and [11] Wen, M., Yang, D., and Rosé, C.P., under review. Fischer, F., 2012. The ACODEA Framework: Linguistic Reflections of Student Engagement in Developing Segmentation and Classification Schemes Massive Open Online Courses. submitted to for Fully Automatic Analysis of Online Discussions. International Conference on Weblogs and Social International Journal of Computer Supported Media. Collaborative Learning 7, 285-305. [12] Yang, D., Wen, M., and Rosé, C.P., under review-a [5] Gweon, G., Jain, M., Mc Donough, J., Raj, B., and Peer Influence on Student Attrition in Massively Open Rosé, C.P., 2013. Measuring Prevalence of Other- Online Courses. submitted to Educational Data Oriented Transactive Contributions Using an Mining. Automated Measure of Speech Style Accommodation. [13] Kumar, A., Palakodety, S., Wang, C., Wen, M., Rosé, International Journal of Computer Supported C.P., and Xing, E., under review. Large scale structure Collaborative Learning 8, 245-265. aware community discovery in online forums. [6] Bielaczyc, K., Kapur, M., and Collins, A., 2013. submitted to Knowledge Discovery and Data Mining. Cultivating a community of learners in a K-12 [14] Yang, D., Wen, M., and Rosé, C.P., under review-b. classroom. In International Handbook of Towards Increasing the Resolvability of Unresolved Collaborative Learning, C.E. Hmelo-Silver, C.A. Threads in MOOCs. submitted to International Chinn, C.K.K. Chan and A.M. O'Donnell Eds. Conference on Weblogs and Social Media. Routledge, New York, 233-249. [15] Martin, T. and Sherin, B., 2013. Learning analytics [7] Chan, C.K.K., 2013. Collaborative knowledge and computational techniques for detecting and building: Towards a knowledge creation perspective. evaluating patterns in learning: An introduction to the In International Handbook of Collaborative Learning, special issue. Journal of the Learning Sciences 22, C.E. Hmelo-Silver, C.A. Chinn, C.K.K. Chan and 511-520. A.M. O'Donnell Eds. Taylor and Francis, New York, 437-461.