Marti A. Hearst

Professor

University of California, Berkeley

Learning@Scale

The goal of this research is to empower people by finding ways to improve learning by harnessing technology at scale and combining it with insights from learning science.

MOOCChat: Peer Learning in MOOCS

This project explores the benefits of bringing Social Interaction to Massive Open Online Courses (MOOCs). Students taking MOOCs benefit from being able to do their work anytime and anywhere, but miss out on the benefits of interacting with other students that in-person classrooms offer.

Peer learning has been shown in dozens of in-person classroom studies to improve learning, retention and morale across student demographics; the goal of this work is to see if the benefits seen in face-to-face peer learning can be replicated in MOOCs and SPOCs. As opposed to unstructured team projects, group interactions in peer learning are structured, often brief, and membership of groups is encouraged to be heterogeneous and rotated frequently. Our software automatically places remotely located students into synchronous groups for discussing the ideas behind study questions. In a series of experiments we have found that answers improve after small group discussions when participants are incentivized to help the entire group achieve the correct answer. We have performed initial tests in a >1000 person SPOC as well, with similar promising results.

Videos:

Peer Feedback on Design Assignments

In massive online courses, we need to get creative about how to give feedback to students about their creative work on assignments. We are developing methods to improve peer feedback on design problems. We suggest a new approach that combines the methods of heuristic evaluation and a simple usability test into one assessment that allows instructors and students to check for outliers in assessments and to rank different designs according to both qualitative and quantitative measures simultaneously.

Reputation Systems and Chat Rooms in MOOC Forums

MOOC Forums are the way students currently interact with one another in most MOOCs. We have experimented with introducing reputation features into MOOC online forums, finding faster response times and larger number of responses per posts, along with differences in how students ask questions, but no differences in grades, retention, or feelings of community. We also found that a course-wide chat system did not engage a significant slice of the student population.

Natural Language Processing and Education

Linking Textbooks To Related Materials

Online, open access, high-quality textbooks are an exciting new resource for improving the online learning experience. Because textbooks contain carefully crafted material written in a logical order, with terms defined before use and discussed in detail, they can provide foundational material with which to buttress other resources. As a first step towards this goal, we explore the automated augmentation of a popular online learning resource – Khan Academy video modules – with relevant reference chapters from open access textbooks.

Generating Questions from Ontologies

Ontologies provide a structured representation of concepts and the relationships which connect them. This work investigates how a pre-existing educational Biology ontology can be used to generate useful practice questions for students by using the connectivity structure in a novel way. It also introduces a novel way to generate multiple-choice distractors from the ontology, and compares this to a baseline of using embedding representations of nodes.

From Rationales to Dialogue: A Deep Learning Approach for Explaining Math Problems to Students

This project takes a novel approach to intelligent tutoring by utilizing deep learning techniques to present a student with \textit{an explanation} for the next step in a solution, \textit{conditioned on their current learning state}. We propose a three-step approach which: (1) generates an appropriate rationale conditioned on a set student difficulty, (2) identifies how a student's math knowledge is lacking based on prior problem answers and a neural model of their internal learning state, and (3) extends these two tasks to form a dialogue-based agent which determines the minimal viable hint to prompt the student to work toward the correct answer.

Publications

  • Laskowski, P., Karayev, S., and Hearst, M. How Do Professors Format Exams? An Analysis of Question Variety at Scale, ACM Learning@Scale 2018 (conditionally accepted).
  • Stasaski, K and Hearst, M.A., Multiple Choice Question Generation Utilizing an Ontology, BEA Workshop, EMNLP 2017   pdf   supplement
  • Hearst, M.A., Can Natural Language Processing Become Natural Language Coaching?, ACL 2015 . pdf
  • Milli, S., and Hearst, M.A., Augmenting Course Material with Open Access Textbooks, BEA Workshop, NAACL 2016. pdf
  • Hearst, M.A., Laskowski, P., and Silva, L., Evaluating Information Visualization via the Interplay of Heuristic Evaluation and Question-Based Scoring, CHI 2016, to appear. pdf   auxiliary materials zip
  • Coetzee, D., Lim, S., Fox, A., Hartmann, B., and Hearst, M.A., Structuring Interactions for Large-Scale Synchronous Peer Learning, CSCW 2015. pdf
  • Hearst, M.A., Coetzee, D., Fox, A., and Hartmann, B., All It Takes Is One: Evidence for a Strategy for Seeding Large Scale Peer Learning Interactions, Learning at Scale Work in Progress track, ACM Learning At Scale 2015. pdf
  • Coetzee, D., Fox, A., Hearst, M.A., and Hartmann, B. Should your MOOC Forum Use a Reputation System? CSCW 2014. pdf
  • Stephens-Martinez, K., Hearst, M.A., and Fox, A., Monitoring MOOCs: Which Information Sources Do Instructors Value?, ACM Learning At Scale 2014. pdf
  • Coetzee, D., Fox, A., Hearst, M.A., and Hartmann, B. Chatrooms in MOOCs; All talk and no action, ACM Learning at Scale 2014. pdf
  • Lim, S., Coetzee, D., Hartmann, B., Fox, A., and Hearst, M.A., Initial Experiences with Small Group Discussions in MOOCs, Work in Progress, ACM Learning at Scale 2014. pdf

Talks

Software

Open source software for MOOCChat is available. Two versions of the code have been developed. Use at your own risk; no support is offered.

Conference Organizing

ACM Learning@Scale conference: Profs. Hearst, Fox, and Chi co-founded this conference in 2014.

BayLan 2018: Prof. Hearst co-organized a Bay Area industry-academic conference on learning technology in March 2018 called BayLan. The conference sold out at 140 people.

Support

    This research is supported by a generous research award from Google's Social Interactions Program, the National Science Foundation under Grant No. IIS 1149799 and IIS 1210836, and from an AWS Machine Learning Research Award.