Working Memory and Learning Underlying Website Structure
description
Transcript of Working Memory and Learning Underlying Website Structure
Working Memory and Learning Underlying Website Structure
Steven Banas & Christopher A. Sanchez Cognitive Science & Engineering
Arizona State University
Information Gathering on the Web• The World-Wide-Web is complex and organized
in many different ways– Not all websites include navigational aids
• Without navigational aids users must rely on mental models created from information from various sources to guide their searching– i.e., Previous experience with domain or web
structure• Goal: match user’s mental model to actual
structure
Matching mental models• During search, prior knowledge must be
combined with incoming information to guide a users searching behavior– For example, previous experiences with Wikipedia
and similarities with the current page– Effortful and conscious process
• This process of combing incoming knowledge with previous knowledge has been shown to occur within the working memory system.
Working Memory• Working memory capacity (WMC) has emerged in the
past 30+ years as a powerful theory that predicts performance and behavior across a wide array of tasks.– Reading performance, g, science learning, anti-saccade, etc.
• Strongly tied to the notion of controlled attention– Ability to focus attention on relevant information and
either suppress or otherwise ignore task irrelevant information.
• More than just STM, as it includes aspects of both executive processing AND storage.
Working Memory and Web Learning
• Remember:– WMC predicts how well individuals connect discrete
concepts and make appropriate inferences – High WMC individuals have been shown to be better able
to retain information that is relevant and useful for integrating textual information, even in the face of related processing demands
• So…– Relative to the context of web search for understanding,
WMC should also predict learning from multiple web documents
– Integrating this information across discrete pages
Current study
• Participants (N=62) read a Wikipedia-like page on Plant Taxonomy
Website
• Hierarchical tree structure that contained 4 levels– 24 total pages– Each page ~ 500 words
• Navigated only using links– Links mirrored hierarchical structure of content
• Participants were not given a site map• Participants entered the website at the top
Screenshot of Website
Pre/Posttest Questions
• Participants rated their knowledge of plants and biology on a 1-5 scale
• Also completed– Hierarchical tree construction task.
• Place terms in correct location in hierarchy• More global measure of hierarchy
– Matching task• Choose item immediately connected in hierarchy• More local measure of hierarchy
• Completed tasks again after reading
Search Questions
• 18 short answer questions to be completed while searching the website
• Simple factual questions, drawn evenly from the entire website. – i.e. “What is the scientific name of clubmosses? “
WMC Measure
• Automated Operation Span task (AOSpan)
• Equation-letter strings were presented in sets of between 2 and 7 strings.
• Participants completed 3 trials of each set size, and the order of these sets was randomized.
IS 8/4 +1 =2? C
Hypotheses
• High WMC Individuals– Better able to construct a more accurate tree than lower
WMC individuals due to a better more robust mental model of the material and inferencing
– Better able to complete both the search questions and the tree construction due to the increased capability to handle both simultaneous tasks
• Low WMC Individuals– More taxed by the secondary search task, less likely to
develop an accurate mental model needed to complete the tree construction task
Results: Search Questions
• Overall, participants were able to adequately complete the search task (M=9.93, SD=3.73).
• Performance was not significantly correlated with – WMC (r(61)=.04, p>.05)– Knowledge of plants (r(61)=.07, p>.05)– Knowledge of biology (r(61)=.08, p>.05).
• Search Questions were more or less difficult regardless of WMC of prior knowledge
Results: Matching
• Significant improvement pre-post– F(1, 61)=34.99, ηp
2=.37, p<.01
Matching0
1
2
3
4
5
6
PrePost
Change in Matching Task• Hierarchical regression on improvement
– First block: WMC, knowledge of plants, and knowledge of biology
– Second block: interaction terms between WMC and both prior knowledge variables
• First block Results: – R2=.07, F(3, 61)=1.37, ns– No variables significant predictors
• Second block Results: – Interaction Terms did not significantly improve the fit of the
model• R2 change=.01, p>.05
Results: Tree Construction Task
• Participants did significantly improve pre to post – F(1, 61)=36.15, ηp
2=.37, p<.01).
Tree Construction0
1
2
3
4
5
6
PrePost
Change in Tree Construction Task• Hierarchical regression on
improvement– First block: WMC, knowledge of
plants, and knowledge of biology– Second block: interaction terms
between WMC and both prior knowledge variables
• First block Results: – R2=.12, F(3, 61)=3.71, p<.05– WMC only significant predictor of
learning gains gain (β=.35, p<.05)• Second block Results:
– Interaction Terms did not significantly improve the fit of the model
Discussion• Results show that WMC does influence how well individuals
learn and remember the underlying, non-explicit, structure of complex material. – High WMC individuals improved their implicit understanding of the
material on the website, lower did not– Effect was not mediated by prior knowledge
• Results are important for online learning environment designers as it shows that individual differences do impact how learners grasp implicit information
• Also shows that user control over what navigational tools are available to them would benefit the user experience and learning
Future Work
• Extend to other domains• Other relevant individual differences• Test-bed for creation of better navigational
tools and learning aids