Markov models of search state patterns in a hypertext information retrieval system

Abstract

The objective of this research is to discover the search state patterns through which users retrieve information in hypertext systems. The Markov model is used to describe users' search behavior. As determined by the log-linear model test, the second-order Markov model is the best model. Search patterns of different user groups were studied by comparing the corresponding transition probability matrices. The comparisons were made based on the following factors: gender, search experience, search task, and the user's academic background. The statistical tests revealed that there were significant differences between all the groups being compared. IntroductionThe purpose of the present study is to expand our knowledge of the actual use of hypertext systems. Specifically, I hope to discover the search state patterns through which users search for information in hypertext systems. By a search state, I mean an event in the search process, such as accessing the index or displaying an article. The questions to be investigated in this article are as follows: What is the distribution of different search states in a hypertext system? What is the probability of going from one state to another state? What states are most likely to follow one another? Do different user groups have different search state patterns? If so, how do they differ? Do differences in search tasks (general task vs. specific task) influence users' choice of states and, if so, how do they influence them? The main method for describing the search state patterns in this study is through the development of mathematical models as described below.The sequence of states that searches go through can be described mathematically by a Markov model, that is, by transition probability matrices. The Markov model is both descriptive and predictive. The datum in each cell of a transition probability matrix is the probability of going from the corresponding row state to the column state. Thus, a transition probability matrix describes a pattern of movements through states that can provide a map of the user behavior. An analysis of user behavior in this manner not only describes what proportion of a search is involved in string searching, index searching, etc., but also what states or actions are most likely to follow one another. The comparison of transition probability matrices across different user groups and search tasks will reveal the search patterns of the corresponding groups and tasks.It is necessary to define some concepts before reporting the details of this study. Hypertext is a network of nodes connected by links. A node is the basic unit used to store information, that is, a unit of text or a graph. A link is a relationship between two nodes. For example, a node that contains the citing paper and a node that contains the cited paper can be connected by the link citation. Since information is stored in nodes, users get information by visiting nodes. Because there are different types of links among nodes, there are different ways of moving from one node to the ot...

Stochastic modeling of usage patterns in a web‐based information system

“…Most importantly, Penniman found that significant differences in short sequences of activities do not account for differences in longer sequences of activities. Qiu (1993) used discrete-time Markov chains to describe user behavior. Eight search states were defined, and empirical testing showed that state transition behavior followed a second-order DTMC model.…”

Implications of User Characteristics in Information Seeking on the World Wide Web

“…This result coincides with findings of previous research. In a study investigating the navigational pattern on a hypermedia system, Qiu (1993b) discovered that Experienced searchers navigated the system in a nonlinear mode, whereas Novices adopted a more linear browsing approach to navigating the system.…”

“…They were typical types of information search tasks used in different studies (Marchionini, 1989;Qiu, 1993aQiu, , 1993b. For a known-item search task, there was a piece of target information that all participants were expected to retrieve.…”

Inferring user knowledge level from eye movement patterns

“…Information search in task sessions involves non-linear and conditionalized relationships (Wildemuth, 2004, Qiu, 1993. It is reasonable to suppose such relationships may affect the cognitive effort during the task session.…”