[Paper Review] Analysis of the Human-Computer Interaction on the Example of Image-based CAPTCHA by Association Rule Mining

Darko Brodić, Alessia Amelio|arXiv (Cornell University)|Dec 1, 2016

User Authentication and Security Systems7 references4 citations

TL;DR

This study analyzes human response times in solving image-based CAPTCHAs featuring facial expressions using association rule mining to uncover psychological and demographic influences on usability. It reveals that younger users (under 35) solve CAPTCHAs faster, with animated characters being the easiest and worried faces the most difficult, while higher education and internet experience moderate response times for older users.

ABSTRACT

The paper analyzes the interaction between humans and computers in terms of response time in solving the image-based CAPTCHA. In particular, the analysis focuses on the attitude of the different Internet users in easily solving four different types of image-based CAPTCHAs which include facial expressions like: animated character, old woman, surprised face, worried face. To pursue this goal, an experiment is realized involving 100 Internet users in solving the four types of CAPTCHAs, differentiated by age, Internet experience, and education level. The response times are collected for each user. Then, association rules are extracted from user data, for evaluating the dependence of the response time in solving the CAPTCHA from age, education level and experience in internet usage by statistical analysis. The results implicitly capture the users' psychological states showing in what states the users are more sensible. It reveals to be a novelty and a meaningful analysis in the state-of-the-art.

Motivation & Objective

• To investigate how demographic and behavioral factors (age, education, internet experience) influence response time in solving image-based CAPTCHAs.
• To explore the psychological impact of facial expressions (animated character, old woman, surprised face, worried face) on human recognition speed.
• To apply association rule mining to detect natural dependencies between user attributes and response time for improved CAPTCHA design.
• To contribute to cognitive psychology and HCI by identifying user states that correlate with faster or slower CAPTCHA solving.
• To support the development of more usable and user-adapted CAPTCHA systems based on empirical data and statistical patterns.

Proposed method

• Conducted an experiment with 100 internet users, collecting response times for four image-based CAPTCHAs with distinct facial expressions.
• Categorized users by age (below 35, above 35), education level (secondary, higher), and internet usage (low, middle, high).
• Applied association rule mining to identify frequent patterns and dependencies between user attributes and response time categories (low, middle).
• Calculated support, confidence, and lift values to evaluate the strength and significance of extracted rules.
• Used statistical analysis to assess the reliability of associations between demographic variables and solving speed.
• Prioritized rules with high lift and confidence to identify the most influential factors in response time variation.

Experimental results

Research questions

• RQ1 How does age affect the response time in solving image-based CAPTCHAs with facial expressions?
• RQ2 What is the influence of education level and internet experience on the speed of solving different types of image-based CAPTCHAs?
• RQ3 Which facial expression in image-based CAPTCHAs is most easily recognized, and which is most difficult to distinguish?
• RQ4 What are the key demographic and behavioral patterns that correlate with low or moderate response times?
• RQ5 How can association rule mining uncover hidden relationships between user characteristics and CAPTCHA solving performance?

Key findings

• Users under 35 years old consistently achieved low response times across all CAPTCHA types, indicating faster recognition of facial expressions.
• The animated character CAPTCHA was the easiest to solve, with 76% of users under 35 achieving low response time, followed by the old woman (67% of secondary-educated users above 35).
• The worried face CAPTCHA was the most difficult, with only 55% of users above 35 and secondary education achieving low response time, and 94% of those with higher education and middle internet usage achieving middle response time.
• Higher internet usage and higher education were strong predictors of faster response times, especially for complex facial expressions like the worried face.
• Association rules with high lift (e.g., 2.14 for above 35, higher education, middle internet usage → middle response time) revealed significant dependencies between user profiles and solving speed.
• The surprised face CAPTCHA showed no strong association with low response time, indicating it is less intuitive or more ambiguous for users.