Transitions in search tactics during the Web‐based search process

Abstract

Although many studies have identified search tactics, few studies have explored tactic transitions. This study investigated the transitions of search tactics during the Web-based search process. Bringing their own 60 search tasks, 31 participants, representing the general public with different demographic characteristics, participated in the study. Data collected from search logs and verbal protocols were analyzed by applying both qualitative and quantitative methods. The findings of this study show that participants exhibited some unique Web search tactics. They overwhelmingly employed accessing and evaluating tactics; they used fewer tactics related to modifying search statements, monitoring the search process, organizing search results, and learning system features. The contributing factors behind applying most and least frequently employed search tactics are in relation to users' efforts, trust in information retrieval (IR) systems, preference, experience, and knowledge as well as limitation of the system design. A matrix of searchtactic transitions was created to show the probabilities of transitions from one tactic to another. By applying fifthorder Markov chain, the results also presented the most common search strategies representing patterns of tactic transition occurring at the beginning, middle, and ending phases within one search session. The results of this study generated detailed and useful guidance for IR system design to support the most frequently applied tactics and transitions, to reduce unnecessary transitions, and support transitions at different phases.

Beyond actions: Exploring the discovery of tactics from user logs

“…They found that participants used different search tactics depending on the phase of the search session. Similar to Chen and Cooper (2002) and Xie and Joo (2010) , our proposed method identifies search tactics from logs, and we analyse user search tactics in terms of their transition patterns based on a Markov chain model. However, instead of a heuristic comparison of pattern frequencies, we summarise tactic transition patterns into a single statistics in terms of entropy rate of Markov chains, with which analysis can be performed with standard statistical tools.…”

“…Further, this process allows the annotators to analyse each individual action or combination of actions in detail and in context. The drawbacks of this approach are, however: (1) similar to previous methods ( Chen & Cooper, 2002;Goodrum et al, 2003;Xie & Joo, 2010 ), it is labour intensive, although a labelling tool helps to better visualise and manipulate the sequences; (2) a strategy to resolve disagreement between annotators is needed. To address these shortcomings we explore alternative approaches, namely heuristic rule-based and statistical segmentation.…”

“…Transitions between tactics have been widely used as a means of analysing patterns of user search behaviours ( Chapman, 1981;Chen & Cooper, 2002;Han et al, 2013;Xie & Joo, 2010;Yue, Han, & He, 2012 ). Typically, Markov chains are used to model user search tactic sequences, and most frequent sequences of tactics are identified and compared from different user groups ( Chen & Cooper, 2002;Xie & Joo, 2010 ).…”

“…In practice, however, such abstractions can be time-and labour-consuming to create, making them too expensive to scale well. To bring log file and tactic representations closer, attempts have been made to make tactic representations from log files through hand coding ( Goodrum, Bejune, & Siochi, 2003;Xie & Joo, 2010 ); while such approaches do not scale to large and dynamic log files, they are well suited to smaller lab studies. Automatic methods for tactic identification from log files using unsupervised HMM have also been proposed ( Han, Yue, & He, 2013;Yue, Han & He, 2014 ) where tactics are modelled as hidden states and low-level actions are modelled as observables; however, the interpretation of the resulting hidden states remains unclear-a drawback inherent to unsupervised machine learning methods.…”

“…In this paper, we take a step beyond the above attempts ( Goodrum et al, 2003;Han et al, 2013;Xie & Joo, 2010 ) towards a solution for analysing user search behaviour at a tactical level, specifically, we seek answers to the following research questions:…”