Improving Product Browsing whilst Engaging Users Stefano Padilla, Fraser Halley and Mike J. Chantler Texture Lab, School of Mathematical and Computer Sciences Heriot Watt University, Edinburgh, EH14 4AS, UK +44 (0) 131 451 4166 { S.Padilla, F.Halley, M.J.Chantler } @hw.ac.uk ABSTRACT We describe a new browsing system that improves online navigation of large numbers of objects which are otherwise difficult to label or split into categories. We use a two-stage process to obtain human similarity information which forms the core of the system and is used to create a resizable self-organizing map. This comprises a matrix of image stacks that provides a fast, intuitive and highly visual navigation environment. It can be easily implemented in sites, blogs and newsletters. We report the methods used to obtain the similarity data and derive the navigation system, and illustrate this with an example taken from the domain of abstract art. Categories and Subject Descriptors H.5.m [Miscellaneous]: Information Interfaces and Presentation (e.g. HCI) Miscellaneous. General Terms Documentation, Design, Human Factors. Keywords Navigation, categorization, engagement, Interactive, presentation, engagement, web, tool, visualization, gamification. 1. INTRODUCTION One of the core components of any website is the navigation (menus). In most websites, products can be easily categorized by type, functionality or colour. For example in a shoe shop the shoes will be categorized by gender, type, colour, size and brand. Known categories and common labels allow users to find products with relative ease and speed. However, objects that do not have obvious categories such as wallpapers, abstract art or patterns are notoriously difficult to navigate. In most cases users quickly become frustrated if they can't easily locate the object they are looking for. This can have a negative influence on shopping behaviour, Menon et al [1]. Therefore, we have investigated the use of community-derived similarity judgements to develop a new browsing system. The data are collected in a two stage process. First, free-grouping experiments are performed with a smaller (circa 100 object) subset to provide an embryonic navigation system which is then used to elicit information on the whole catalogue. The second stage can be ‘gamified’ as described by Deterding et al [2] to improve engagement and take-up. 2. CROWD BASED NAVIGATION It is well known that crowds have a relatively similar performance when compared to an expert on a given task (wisdom of crowds) [3]. In our method, we compile the knowledge from an online community to create a navigation which is intuitive to all users. We decided to use a browsing model based on self-organizing maps (SOM) [4] as described by Halley [5]. In the first stage we ‘bootstrap’ the process by creating a similarity matrix with a subset of a reduced number of samples (100) in a controlled environment. Individuals are asked to organise the subset into groups of ‘similar’ objects. A similarity matrix is derived in which the similarity score between any two objects is simply the number of times that the objects have been placed in the same group. This provides the data for construction of the ‘bootstrap’ navigation system comprising a self-organizing map (SOM) of a specified width and height (size is tuned to display space). The 100 objects are divided into a number of ‘image stacks’ that contain similar objects. Each image stack being represented by a single exemplar image, clicking on the exemplar reveals the contents of the stack. The exemplars are arranged in a grid in which the screen distance correlates with the mean similarity between the two stacks of objects. This model of navigation is simple for observers to grasp and has been proven to be an efficient browsing model [5]. We use the bootstrap system to derive similarity matrices for the complete catalogue using crowd-sourcing. Users are presented with a ‘new’ objects and asked to navigate to the closest matches in the bootstrap subset. These matches are used to calculate new entries in the similarity matrix. User Engagement in this second phase is particularly important. 3. GAMIFICATION OF THE TASK The main problem of crowd-based processes is finding a willing and focused audience. This is when "gamification" is important. Gamification is the term commonly used when video game elements are introduced to non-gaming systems to improve user's experience. Many websites now use gamification to engage and reward users. We incentivised the second stage of the similarity matrix derivation by creating a game environment where users see new object which they try to match with a similar object. Users are then rewarded for their contribution. The game-reward mechanic helps the enjoyment and engagement of the users. It is also important to note that users increase their sense of community, ownership and contribution [6] by using this mechanism. Digital Engagement '11, November 15 - 17, 2011, Newcastle, UK.