AlgoSolve: Supporting Subgoal Learning in Algorithmic Problem-Solving with Learnersourced Microtasks

Title of the Paper

AlgoSolve: Supporting Subgoal Learning in Algorithmic Problem-Solving with Learnersourced Microtasks

Paper Information

• Domain: Computer Science Education and Algorithm Instruction
• Keywords: Algorithmic problem-solving, learnersourcing, subgoal learning, online education, programming learning, solution planning, learning support systems, cognitive load, student contributions, instructional design

Research Background and Problem

• Identified Issues or Challenges:

  1. Beginners often plan solutions while coding, lacking a comprehensive and structured understanding of the problem, which leads to failure or inefficiency in problem-solving.
  2. Subgoal learning has been proven to aid in developing comprehensive solution plans, but it relies on expert-created subgoal labels, which are scarce and costly resources.
  3. In self-learning environments, the lack of expert-created materials significantly limits the application of subgoal learning.

• Significance:

  1. Algorithmic problem-solving holds a crucial position in programming education and software design, with widespread applications in programming contests and online learning platforms.
  2. Subgoal learning supports effective knowledge transfer, helping learners solve similar problems more efficiently.

• Research Motivation and Related Work:

  1. Existing research demonstrates that subgoal learning enhances learners' understanding of problem-solving structures, particularly in algorithms.
  2. Learnersourcing methods can reduce reliance on expert support by involving learners in material creation, simultaneously improving their own skills.
  3. There is a need to design workflows that improve the quality of learnersourced tasks to ensure the generated materials are both educationally valuable and skill-enhancing for learners.

Solution

• Main Methods or Solutions:

  1. Proposing a two-stage learnersourcing workflow to support learners in generating high-quality subgoal labels:
     • Stage 1: Subgoal Voting Task, where learners compare the quality of multiple subgoal labels to identify high-quality ones.
     • Stage 2: Subgoal Labeling Task, where learners create initial labels, and the system provides high-quality label examples for comparison and improvement.
  2. Designing and implementing a prototype system, AlgoSolve, to support subgoal learning.

• Innovations:

  1. Replacing expert resources with learnersourcing methods, reducing the cost of creating high-quality subgoal labels.
  2. Introducing guided microtasks in subgoal learning to enhance learner engagement and material quality.
  3. Employing a Multi-Armed Bandit algorithm to dynamically select optimal subgoal label examples, balancing exploration and exploitation.

• Implementation Steps and Core Techniques:

  1. Initial Example Collection: Designing subgoal labeling tasks to allow participants to generate a pool of foundational labels.
  2. Microtask Implementation:
     • Subgoal Voting Task: Learners are shown code snippets with multiple label examples and asked to select the best description.
     • Subgoal Labeling Task: Learners create and refine their own labels by comparing them with peer examples recommended by the system.
  3. Dynamic Adjustment: Utilizing a Multi-Armed Bandit algorithm to dynamically adjust the recommended high-quality examples based on learner choices.

Research Outcomes

• Specific Results:

  1. The AlgoSolve system significantly improved the quality of subgoal labels generated by learners. Compared to the baseline, it better reflected the objectives and deeper meanings of the code.
  2. 31% of participants using AlgoSolve successfully provided complete solution plans, compared to only 6% in the baseline group.

• Comparison with Existing Solutions:

  1. Learner-generated labels from AlgoSolve were comparable in quality to expert labels, with better performance in explaining details and meeting beginner needs.
  2. Compared to a baseline condition without microtasks, the microtask approach significantly improved the quality of learners' subgoal labels.

• Experiments and Evaluation:

  1. Conducted a comparative experiment with 63 participants (algorithm problem-solving beginners), divided into microtask and baseline groups.
  2. Used the SOLO taxonomy to evaluate participants' solution plans, showing that microtasks significantly improved learners' ability to apply techniques to new problems.
  3. Statistical results indicated that the microtask workflow effectively enhanced the quality of learners' subgoal labels (p < 0.004, effect size d=0.13).

• Limitations and Future Directions:

  1. Limitations:
     • The sample size was relatively small (63 participants), primarily consisting of Korean participants, which may have influenced the results due to language factors.
     • The system's applicability to other algorithm topics was not extensively validated.
     • Some learners were unable to fully apply the learned techniques to generate effective complete solution plans.
  2. Future Directions:
     • Expanding the system's scale to validate its applicability to more complex algorithm topics.
     • Exploring automated tasks for generating subgoal scopes and hierarchies.
     • Introducing personalized tasks to enhance the learning experience and effectiveness for learners at different skill levels.