Modelling low-level interactions on the Web: contributions to online learning, familiarity, and search behaviour evolution

Abstract

Web technologies are constantly changing, and despite continuous advancements, the user still encounters difficulties interacting with Web platforms. Nowadays, researchers, Web designers, and website operators conduct extensive research on user Web interactions in order to better understand user behaviour, evaluate user experiences, and provide better support to users. Traditionally, user interaction research has relied heavily on controlled laboratory studies, surveys, and interviews. These techniques can potentially be intrusive, prone to bias, and lack ecological validity. While collecting and investigating users, Web interaction data provides an excellent alternative that is both unobtrusive, naturalistic, and ecologically valid. Web user interactions can be classified based on their level of abstraction, from the lowest level (e.g. finger pressing a key) to highest level (e.g. placing an order). We focus on low-level Web interactions, which are UI events generated by the user via user interfaces. There are several advantages of focusing on low-granularity data: it is easy to collect, contains a wealth of information, and potentially contains implicit user behavioural markers that are not observable in high-level data. Analysing fine-grained data, however, does present challenges; they can be noisy, less descriptive, and subject to high cardinality. There are also no well-established techniques for analysing low-level user interaction data. Thus, this thesis addresses this gap by proposing a data-driven methodology that combines data mining, qualitative analysis, and user modelling techniques in order to maximise the benefits of low-level data while mitigating its drawbacks. The proposed methodology and practical application of low-granularity user interaction data were demonstrated in this thesis through three studies, in the domains of online learning, Web familiarity, and search behaviour evolution. We identified user behaviours associated with the exploration of online learning forums, course materials, and Web navigation tools, and we were able to identify user groups with similar search behaviours and track changes in user behaviour over time. Additionally, our findings suggest that incorporating low-level user interaction into the analysis improves the explainability of user models. The significance of this work lies in the methodological contribution to the practical use of low-level user interactions in user modelling, as well as for the findings in the research areas of online learning, Web familiarity, and search behaviour evolution.

Date of Award	31 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Simon Harper (Supervisor) & Markel Vigo (Supervisor)