PH.D DEFENCE - PUBLIC SEMINAR

Enhancing User Engagement with Online Information: The Impacts of Online Advertisement Layout and Serendipitous Recommendation

Speaker
Ms Cui Wei
Advisor
Dr Vaibhav Rajan, Assistant Professor, School of Computing


13 Oct 2021 Wednesday, 02:00 PM to 03:30 PM

Zoom presentation

Abstract:

The massive amount of information on the Internet has brought great challenges to both online marketers and website visitors. On the one hand, it is difficult for online marketers to present their marketing information to consumers by just passively waiting for consumers to search; on the other hand, website visitors are struggled to discover the products of their interest among hundreds, thousands, or even millions of alternatives on the Internet.

To alleviate these two concerns for online marketers and website users, two information technologies have been invented: online advertising and recommender system. This thesis focuses on these two information technologies, identifies practical challenges for each technology and proposes corresponding solutions.

Study One aims to investigate the impact of different visual designs of online static advertising on consumer attention and ad effectiveness. Typical static ads use both text and pictures to present products' information. While prior literature mainly studies the effects of ad content, there is a paucity of research exploring how the spatial relationship between ad text and picture affects consumer attention to the ad. This study focuses on a fundamental, yet under-studied design factor: the placement of ad text. Results from a field experiment and two eye-tracking laboratory experiments suggest that an online ad with text separate from picture will lead to a longer attention duration on the ad and enhance ad effectiveness than that with text embedded in ad picture.

Study Two focuses on the other information technology: recommender system. Recommender systems enable users to discover items of interest from a large set of alternatives. Most recommender systems employ accuracy-oriented algorithms to predict user preferences. Overemphasis on accuracy leads to monotony in the items predicted, resulting in low customer satisfaction. Hence, to improve user experience, it is crucial to inject serendipity into recommendations by discovering users' latent preferences and recommending items that are both relevant and unexpected. However, there is a lack of serendipity-related labelled data and previous serendipitous recommendation algorithms are often unable to strike a trade-off between serendipity and accuracy. We address the challenges by presenting a new serendipity-related dataset and proposing a single-objective and a multi-objective deep learning-based serendipitous recommendation algorithms. Our experiments on the self-collected dataset demonstrate that our models improve over state-of-the-art methods, in both recommendation accuracy and serendipity prediction.

Taken together, this thesis aims to contribute to Information Systems and Human-computer Interaction literature by exploring how to smartly design online information technologies to better engage consumers. The findings of the two studies extend our understanding of the design of online advertising and recommender systems and have important practical implications for online marketers and platforms.