AbstractThe amount of digital information available in the Internet and various Intranets often causes information-overload, significantly increasing the amount of time and cognitive resources needed to acquire relevant and accurate information. When searching for information to address complex problems, users spend significant amount of time clicking through search-query results and reformulating the search if the results are not satisfactory. This is a tedious and challenging task and has a negative impact on the global economy.
Personalisation and contextualisation techniques intend to address the abovementioned problem. Such techniques help to derive additional information from the history of the past user interactions (user profile) or the current context of interaction. This information is used to refine the search results in order to narrow down the scope of search queries for better results.
One of the key requirements for the development of personalised or contextualised search utilities is consistency of human behaviour. It is only possible to predict user future preferences and actions if they are correlated with past behaviour. This fact is frequently ignored in the current IR literature even though empirical evidence clearly illustrates that humans are very inconsistent when interacting with information. This leads to very low predictive validity of existing contextualisation/ personalisation IR implementations.
This thesis hypothesises that HIB could be consistent under certain contextual and task specific conditions. The thesis claims that a large proportion of our daily information activities are highly consistent and also meet the definition of habitual behaviours. In other words, even though empirical evidence clearly illustrates that on average human information behaviour is
very complex and dependent on thousands of factors, there will exist a small group of daily activities which are highly consistent and can be supported effectively by modern IR utilities. As a consequence the development of highly effective personalised or contextualised solutions are feasible.
In order to prove the above mentioned hypothesis User Study 1 (diary study)was carried out. User Study 1 revealed that a significant proportion of our daily information interaction is indeed consistent (49%) and a significant proportion of this can be classified as habitual (41.9%). User Study 1 also confirmed that the behaviour of participants is consistent only when the same tasks are carried out in the same context and under the same emotional user state. Finally User Study 1 confirmed that the behavioural consistencies are highly individual and affected by a number of external factors. However there exists no HIB model or research methodology which can help to identify the external factors systematically and take advantage of them during IR system design. Therefore this research proposes an ”Integrated Framework for HIB in-situ” which intends to fill the above mentioned gap in knowledge.
The framework was designed to support the development of Information Retrieval utilities based on consistent behaviour of clearly defined user groups and their problems. In order to illustrate its applicability a single user group was selected. The proposed framework was successfully applied to the problem of work-related activities of software engineers. The framework allowed for identification of a more specific user group of software developers and narrowed down the investigated task to code development/debugging. In the next step the framework allowed for shortlisting a number of behaviours which had a significant potential for consistency. The consistency of the shortlisted behaviours and their correlation to relevance was verified through User Study 2 (questionnaire). The key shortlisted behaviours were further analysed through User Study 3 (fully automated, long lasting ethnographic study) which allowed for the identification of factors that have a key impact on behavioural variance. The analysis revealed a number of consistent behaviours (implicit-feedback indicators) that can be used for
prediction of document relevance. Importantly User Studies 1 and 3 vali-dated the research hypothesis on a specific case study of software engineers .However the proposed framework is based on very basic cognitive mechanisms responsible for the human decision-making process and as a consequence is highly generalizable to other user and problem groups.
|Date of Award||2013|
|Supervisor||Rahat Iqbal (Supervisor), John Halloran (Supervisor) & Anne James (Supervisor)|