Detecting memory and structure in human navigation patterns using Markov chain models of varying order

Area of Science:

• Web Science
• Information Science
• Computer Science

Background:

• Human web navigation is frequently modeled using Markov chains, assuming a memoryless property where the next page depends only on the current page.
• This memoryless assumption underpins algorithms like Google's PageRank.
• Recent research suggests higher-order Markov chains, incorporating browsing history, might better model navigation, but their complexity is a barrier.

Purpose of the Study:

• To investigate the appropriate order of Markov chain models for human web navigation.
• To analyze the memory and structure of web navigation at both page and topical levels.
• To compare the utility and complexity of different Markov chain inference methods.

Main Methods:

• Development and application of advanced inference methods to determine optimal Markov chain order.
• Analysis of web navigation data, distinguishing between page-level and topical-level transitions.
• Comparison of memoryless (first-order) and higher-order Markov chain models.

Main Results:

• The complexity of higher-order Markov chain models often outweighs their utility for page-level navigation.
• The memoryless assumption is violated at the topical level, revealing observable regularities in topic transitions.
• Distinct structural differences were found between goal-oriented and free-form navigational datasets.

Conclusions:

• The memoryless Markov chain model remains a practical choice for page-level web navigation analysis.
• Topical-level web navigation exhibits memory, indicating the need for more context-aware models.
• Future research should focus on contextual studies to better understand nuanced human navigation patterns.