Relevance at a Distance - An Investigation of Distance ... - CiteSeerX

4 downloads 6430 Views 624KB Size Report
Department of Computer Science. University College Dublin ... and it clearly has the potential to improve the degree to which person- alized navigation can save ...
Relevance at a Distance - An Investigation of Distance-Biased Personalization on the Mobile Internet Barry Smyth, Kevin McCarthy & James Reilly Smart Media Institute Department of Computer Science University College Dublin Belfield, Dublin 4, Ireland [email protected] Abstract. The usability of mobile portals has been a major stumbling block since the advent of the mobile Internet and WAP handsets. Indeed poor usability is cited as a major contributing factor to the poor take-up of mobile Internet services amongst consumers. A key problem relates to the amount of time that users spend navigating to content as they browse mobile portals. Recent advances in personalization technology have the potential to solve this problem, and today a number of leading operators already provide their users with access to intelligent portals that are automatically personalized based on subscriber usage patterns. In this paper, we examine this so-called personalized navigation technology and propose ways in which it may be enhanced by combining structural properties of a mobile portal (such as the distance to content sites) with the access probabilities of users. We demonstrate that although such distance factors have proven successful in Web personalization, they are less beneficial when it comes to the personalization of mobile portals.

1 Introduction The mobile Internet (MI) has failed to live up to end-user expectations. Limited bandwidth, unreliable handsets, patchy content and poor usability have all contributed to this state of affairs. And although recent developments have seen significant improvements in bandwidth, handsets and content, usability remains a problem, particularly in relation to the navigation effort faced by users when searching for content in a typical mobile portal. For example, recent studies have highlighted how content services are usually positioned to be more than 16 clicks from the portal home page. In other words, to access a typical content service a user can expect to have to make 16 clicks on their mobile phone as they navigate through the portal, scrolling through menus and selecting options enroute [13]. The result is low levels of satisfaction from end-users and lackluster usage levels for mobile operators. Recently, however, a compelling solution has emerged that has been proven to have a dramatic impact on portal usability by significantly reducing the above navigation problem. This so-called personalized navigation solution applies personalization techniques to the navigation task. That is, instead of recommending individual content items to users, menus and menu options are recommended in such a way that users require an average of 50% fewer clicks to locate content, leading to significant increases in mobile usage [15, 16]. Very briefly, this personalized navigation approach estimates Pu (o|m),

the probability that a given user, u, currently in menu m, is looking for menu, o. Menu options are promoted to the user based on their past access probabilities. In this paper we focus on this probabilistic personalized navigation technique in the context of mobile portal personalization. In particular, we investigate whether or not performance improvements can be achieved by extending the basic probabilistic personalization model to take into account an options’s distance from the current menu as well as its access probability. For example, consider two menu options, o1 and o2 , both with the same access probabilities; that is, Pu (o1 |m)=Pu (o2 |m). But suppose that the distance from m to o1 is greater than the distance from m to o2 , that is Distance(m, o1 ) > Distance(m, o2 ), then shouldn’t o1 be promoted ahead of o2 because if correct a greater number of navigation clicks will have been saved? We call this distance-biased promotion and it clearly has the potential to improve the degree to which personalized navigation can save a user navigation effort. In fact evidence from Web personalization suggests that such an extension is likely to pay dividends [1]. However, we are conscious that traditional Webbased portals and current mobile portals are very different and what works on the Web does not always translate well to the mobile space. With this in mind, in this paper we examine the potential of distancebiased techniques in the personalization of mobile portals. In the next section we discuss the background to this research focusing in particular on recent developments in the mobile Internet and outlining past research related to the issue of navigation effort. Section 3 provides a review of our core personalized navigation strategy as detailed in [15, 16]. It also describes the click-distance model of navigation effort and explains how this can be used to bias personalized navigation with respect to portal distance. Finally, before concluding, in Section 4 we describe a recent evaluation to investigate the benefits of this distance-biased technique, based on a large-scale European portal and live user activity logs.

2 The Mobile Internet The mobile Internet refers to the delivery of data services across wireless networks for Internet-enabled handsets as implemented through a group of related infrastructure, protocol and device technologies. It allows the end-user to access various types of data services from their mobile handsets, including Web-style information content, email services, games etc. Access devices range

from limited, first-generation WAP (Wireless Application Protocol, see www.wapforum.org) phones to today’s sophisticated PDAs (Personal Digital Assistants) and so-called SmartPhones (see www.microsoft.com/smartphone). In the past the usability of mobile services has been compromised by limited device functionality, bandwidth, and content. Fortunately the new generation of mobile services (so-called 2.5G services) represents a significant improvement. The major bandwidth and content issues have largely been resolved, and the latest phones offer users significant interface and functionality improvements over earlier models. However, key portal usability problems remain, due to poor mobile portal design. Users find that they are spending too much of their time navigating to content because mobile portals are designed as fixed, complex hierarchies of menu options.

2.1 Mobile Internet Devices From a user experience viewpoint, one of the key features of the mobile Internet is the degree to which existing consumer devices represent a significant step backwards in terms of their functionality, at least when compared to the traditional Internet device (the desktop PC or laptop). In particular, presentation and input capabilities tend to be extremely limited on most mobile devices. For instance, a typical desktop PC, with a screen size of 1024x768 pixels, offers more than 10 times the screen real-estate of a PDA, and more than 20 times the screen space of second-generation Internet phones (eg. Imode and Vodafone Live! handsets or Microsoft’s SmartPhone). Mobile handsets are further limited in their ability to receive user input. The keyboard and mouse functionality of a modern PC are notably absent and the mobile phone numeric keypad makes it extremely difficult for user to input any quantity of information. From a mobile Internet viewpoint, these devices restrict selection features to simple scroll and select keys that allow the user to scroll through menu lists and perform selections. Some improvements are present in most PDAs, which tend to offer touch sensitive screens that are easier to manipulate. Nevertheless data input remains difficult at best.

2.2 Mobile Information Access These key differences that exist between mobile handsets and more traditional Internet devices, such as PCs and laptops, directly influence the manner in which users access information using these devices. For example, on the Internet today search has largely become the primary mode of information access. It is relatively easy for users to input search queries and search engines have improved significantly in their ability to respond intelligently to user needs. In addition the large screen sizes make it feasible for users to efficiently parse the longs lists of search results returned. In contrast, search is far more problematic on mobile devices. Entering queries is simply too time consuming and complex for the average user to tolerate and small screen sizes make it practically impossible for users to easily process the result lists returned. As a result, browsing is the primary mode of information access on the mobile Internet. Instead of searching for information, users attempt to navigate to information by using mobile portals. Today the vast majority of mobile Internet services are accessed via an operator portal with direct search constituting a small fraction (