4.2 Three-tier Architecture
4.2.3 Data Access Layer
At the bottom of the three-tier architecture is the data access layer where a physi- cal MYSQL database is hosted and stores all the data. The data being logged (de- tailed discussion in Chapter 5) with the app were well-defined and formed a schema, as they informed by the News Reader Typology in Chapter 3. The choice, there- fore, of a relational database (RDBMS) over a NoSQL database was preferred. The database is completely structured, organised in tables that define the main entities of the application such as users, news reading, news navigation, context, question- naires and others.
Another important decision that needed to be made for the database design was the use of stored procedures over hardcoded SQL statements in the application layer (presentation Habito News). The choice of stored procedures is reinforced by
4.3. Discussion 83 its nature to allow modular programming in which the emphasis is to separate the functionality of a program into independent and interchangeable modules, aligned with the three-tier architecture. Further, they allow faster execution, as they are pre- parsed statements checked syntactically and semantically thus reducing the time of execution. Finally, they can be used as a security mechanism in order to avoid any SQL injection vulnerabilities in the case of hardcoded SQL statements in the application.
In relation to users’ sensitive data, anonymity was taken into consideration as passwords are hashed using an MD5 encryption method before being stored in the database. Location data is also treated as sensitive data and it will be explained further in detail in Chapter 5 in the section of feature extraction, how anonymity was ensured.
4.3
Discussion
This Chapter introduced the architecture of the framework that was used to address the research questions of this thesis. In particular, it presented a three-tier architec- ture that separates the presentation from the services and the data layer. The com- ponents of our framework aim to address questions relating to automatic detection and recognition of mobile news reading behaviour and the automatic generation of variant user interfaces and interactions that would suit different news reader types. Chapters 5 and 6 will discuss in detail the implementation of the User Modelling component and server side (Services layer) functionality as well as user interface components such as the adaptation mechanism and the variant user interfaces gen- eration. Chapter 7 will present an evaluation study of Habito News in which all the components are being utilised in an attempt to demonstrate the application and examine the effectiveness of an adaptive mobile news app.
Chapter 5
User Modelling for Mobile News
Reading Interactions
The previous Chapter outlined the architecture of the adaptive news research plat- form and presented its main components that facilitate the exploration of the over- arching goal of this thesis.
This Chapter reports the construction and data acquisition of the User Mod- elling (UM) component, which is responsible for generating user models from users’ interactions with Habito News. It presents studies wherein the app was de- ployed through Google Play (‘in-the-wild’), and data analyses were performed in order to demonstrate the capability of automatically recognising patterns of news reading interactions and building models that are able to predict a user’s news reader type, and subsequently, construct a user’s profile.
The Chapter begins with presenting the data collection using Habito News app. In particular, it reports two deployment studies and discusses the data being col- lected. It then presents a hierarchical layered framework that was used to facilitate the analysis of those news reading interaction data collected from the app. The framework defines different levels of abstraction over the logged interaction data and incorporates knowledge from Chapter 3 (i.e. the news reading factors and the News Reader Typology - Section 3.5). The Chapter then explores two approaches of building user models; rule-based models and statistics-based models where the lat- ter utilises machine learning algorithms. Finally, the exploitation of the user model
in order to facilitate the automatic adaptation of Habito News app user interface is discussed.
The user model acquisition and the framework presented in this Chapter has appeared to UMAP ’18. (Constantinides, M., Dowell, J. A Framework for Interaction-driven User Modelling of Mobile News Reading Behaviour. In Proc. UMAP 2018.)
5.1
Motivation
This Chapter presents the development of the User Modelling component of the Adaptive News Research Platform. As discussed in Chapter 4, the User Modelling component is at the core of the platform and is responsible for the user model ac- quisition that will be used during the adaptation process. The Chapter seeks to investigate how behavioural data about user’s news reading interactions can be used to build a user model. In particular, it aims to address the research questions (RQ2 and RQ3 - Chapter 1). These are:
(a) “How can a smartphone app detect and learn individual patterns of news read- ing interactions?”
(b) “How can a smartphone news app exploit a user’s news reader profile to adapt its user interface and interaction?”
Motivated by people’s individual differences while reading news, this Chapter builds on and utilises the News Reader Typology proposed in Chapter 3 in order to propose a user modelling framework for mobile news reading interactions. The aim of the framework is to provide the methods and mechanisms for analysing and modelling news reading interactions in relation to the News Reader Typology. The Chapter explores different user model acquisitions and discusses how they can be exploited as part of the adaptive news research platform.