Related Experiment Videos
Computerizing reading training: evaluation of a latent semantic analysis space for science text.
Christopher A Kurby1, Katja Wiemer-Hastings, Nagasai Ganduri
1Department of Psychology, Northern Illinois University, DeKalb, Illinois 60115, USA. ckurby@niu.edu
Behavior Research Methods, Instruments, & Computers : a Journal of the Psychonomic Society, Inc
|July 2, 2003
Summary
Domain-specific latent semantic analysis (LSA) effectively assesses reading strategies by correlating highly with human judgments. A science-focused LSA space offers advantages for analyzing student think-aloud protocols in science texts.
Area of Science:
- Educational Technology
- Natural Language Processing
- Cognitive Science
Background:
- Assessing student reading strategies is crucial for effective learning.
- Latent Semantic Analysis (LSA) is a computational method for analyzing semantic relationships in text.
- Domain-specific LSA may offer improved accuracy over general LSA in specialized fields.
Purpose of the Study:
- To evaluate the effectiveness of a domain-specific LSA in assessing reading strategies.
- To compare a science-specific LSA with a general reading LSA for this task.
- To explore the utility of LSA in providing feedback for self-explanation reading training (SERT).
Main Methods:
- Students engaged in self-explanation reading training (SERT) and provided think-aloud protocols after each sentence.
- Human raters (novice and expert) and two LSA spaces (general reading, science) evaluated protocol similarity to benchmark reading strategies (minimal, local, global).
- Cosine similarity scores from LSA were analyzed for correlation with human judgments and ability to differentiate strategy levels.
Main Results:
- The science-specific LSA space demonstrated a high correlation with human judgments of reading strategy similarity.
- The science LSA outperformed the general reading LSA in aligning with human assessments.
- LSA cosines could distinguish between different semantic similarity levels but struggled with local processing protocols.
Conclusions:
- A domain-specific LSA space is advantageous for assessing reading strategies, irrespective of its size.
- The findings support the application of science-specific LSA for developing computer-based SERT with online feedback.
- This approach has potential for enhancing science education through automated analysis of student comprehension.