Adaptive User Interfaces

As part of our Interfaces & Interactivity module in my MSc, we had to sit a prior disclosure exam. Two weeks prior to the exam, we were given the essay questions, from which we were to choose to answer. I prepared by writing out my two answers in full, and since I conveniently have them on my laptop, I thought I’d share them here. This is the first one, which read: ‘Give an account of the essential concepts, potential uses, and progress to date in adaptive user interfaces.’ 

(Oh, and yes, this exam taught me that I am indeed fully capable of writing 3000+ words by hand in two hours.)

Adaptive user interfaces (AUI) are systems that change its structure, contents and elements according to the need and context of the user (Schneider-Hufschmidt, Küme, & Malinowski, 1993). AUIs are situated and aim to enable the user to become smarter and more empowered (Jameson, 2008). As computer systems today have become more diverse and commonplace – not just in the workplace or home, but also while on the move –, it is increasingly important for usability and user experience that the system adapts to the user rather than vice versa. This essay will consider core features of AUIs, describe some potential domains of use, and provide examples of these, evaluating their progress to date and key usability challenges.

User model acquisition and user model application are the two core features by which AUIs operate (Jameson, 2008). An AUI acquires a user model by learning and making inferences about a user’s behaviour with an interface. Information about the user can be obtained in various ways: with collaborative filtering, the system gathers preference information (e.g. product ratings) explicitly given by users; whilst with content-based filtering, the system uses e.g. mathematical models such as Bayesian algorithms or artificial neural networks, on the content characteristics itself (Billsus & Pazzani, 2007). The system may also use a hybrid method, whereby collaborative and content-based filtering are flexibly combined. The obtained user model is then applied by the system in deciding how to adapt the interface for the user.

One promising domain for AUIs is task management (Jameson, 2008), whereby the system helps users manage the demands of communication and information flow. This has been explored in research: Lilsys is an availability management system designed to offload the onus of providing unavailability cues from the recipient onto the caller in electronic communication (Begole, Matsakis, & Tang, 2004). Sensors capturing sound, motion and keyboard/mouse activity by the recipient’s workspace is interpreted by Lilsys and communicated to the caller in abstracted form as unavailability icons on the interface. While Lilsys advantageously enables the recipient to remain passive, yet continuously provide abstracted context cues, problems surfaced in user evaluation: the sensors cannot capture recipients’ true intent, e.g. whether they close the office door to signal unavailability or to block outside noises; and interruptions did still occur, albeit the caller would adapt their approach towards the recipient according to their unavailability icon. Nevertheless, Lilsys was found to enhance awareness of task interruptability in an office setting over a 6-month period, and provides a springboard for further research on how to leverage implicit cues from face-to-face communication into AUIs.

Another potential use of AUIs is supporting information acquisition (Jameson, 2008), for instance in reading news online (Billsus & Pazzani, 2007). Users may increasingly suffer from information overload by the plethora of online news sources. Personalising news content and navigation may reduce short-term memory load (Shneiderman, Plaisant, Cohen, & Jacobs, 2009) by providing dynamic news content and reflect multiple, long-lasting interests and preferences (Billsus & Pazzani, 2007). Billsus and Pazzani (2007) developed an adaptive news access system for mobile devices, which highlighted and prioritised news stories based on user-provided preferences over a 4-week period. Of note, adaptive news content is particularly relevant for mobile devices due to their limited screen estate. Moreover, a content-based user modelling approach was appropriately adopted, due to the latency problems associated with a collaborative modelling approach. Compared to static, non-adaptive news access, adaptive news access was found in user study to simplify access to interesting and relevant news stories. Top stories were also more likely to be selected by users when adapted from user modelling of their preferences, than when not. While this was an experimental study, the success of adaptive news access can be seen in e.g. RSS feeds and Google News. The latter aggregates news stories from thousands of sources, and creates a front page based on automatically identified common topics ranked according to estimated importance. This advantageously provides an unbiased news perspective without editorial input. However, such a zeitgeist approach may create a ‘filter bubble’, whereby users are only exposed to isolated and reinforcing information, preventing them from serendipitous information encounters (Pariser, 2011).

Users may further benefit from AUIs that recommend products and services on the basis of their web behaviour and preferences (Jameson, 2008). Recommendation systems that use collaborative filtering examine users’ social environments. Last.fm, for example, is an online musical application that records the music users listen to, and recommends musical artists and local events based on users’ listening habits and geographical location. Recommendation systems adopting content-based filtering include film review websites such as Rotten Tomatoes and IMDB (Internet Movie Database), which recommend films based on similar characteristics, e.g. genres, directors or actors. Netflix, which provides online media streaming, adopts a hybrid approach, by which a user model is created by combining viewing habits and ratings with similar film characteristics in film recommendations. Recommendation systems advantageously reduce the need for users to explicitly know what information is relevant to their need (Jameson, 2008). However, with collaborative filtering, it is important that users are aware of and consent to the system collecting their personal data, as lack of user control (Shneiderman et al. 2009) with regards to privacy and trust may lead to a negative user experience. Here, it is important to strike a balance between explicit and implicit user data collection, as the former, i.e. having the user enter their data directly themselves, may quickly become tedious and frustrating if not constrained to the essentials (Jameson, 2008).

AUIs may have substantial potential outside the web domain as well, for instance in healthcare applications tailoring their information presentation according to user characteristics (Ramachandran, 2009). Healthcare users have a wide range of computer knowledge, from novice (nurses and doctors) to expert (system administrators), and a wide range of domain knowledge, from novice (patients) to expert (nurses and doctors). To address this diversity, usability of healthcare applications would benefit from adapting their interface according to the user. Ramachandran (2009) outlines two potential approaches that can be implemented in healthcare AUIs: adaptive presentation and adaptive navigation. With the former, the information content of the interface changes according to the user, such that e.g. a nurse or doctor perceives advanced medical terminology, while a patient accessing the same interface domain perceives more colloquial language. With adaptive navigation, the way navigational links are presented to achieve an interface goal adapts according to the user. A system administrator could be able to modify patient profiles efficiently by using e.g. command-based language, whereas a nurse should be provided with a graphical user interface incorporating medical terminology. While unimplemented as of yet, these adaptation concepts could potentially empower a wide range of users in using a same application, the challenge of which lies in how the application identifies the user.

While AUIs are both demonstrably and potentially advantageous in improving users’ management of information, there are further usability challenges to consider in their design. Spatial stability, i.e. that adaptation occurs only in a clearly confined area of the interface rather than in the entire system (Gajos, Czerwinski, Tan, & Weld, 2006), is important to maintain the usability principle of consistency (Shneiderman et al., 2009). For example, the Windows XP Start Menu, which can be personalised to prioritise the most frequently used programs, copy rather than move programs from the overall menu to the personalised menu. Enabling internal locus of control (Shneiderman et al., 2009) is also important: if the interface changes its content and structure frequently and without direct input from the user, this can reduce user satisfaction and performance (Gajos et al., 2006). A solution to this is to provide adaptability in addition to adaptivity, i.e. enable user control of content and structure change. For example, in comparing the automatically personalised Smart Menus introduced with Windows 2000 with the prototype of an equivalent user-personalised menu, users preferred the latter, as well as learned and navigated better with the adaptable menu (McGrenere, Baecker, & Booth, 2002). Finally, it is possible that adapting the interface prevents users from learning by exploration. In addition to creating ‘filter bubbles’ as mentioned above, a nurse, for instance, may be prevented from repairing a faulty healthcare application due to his lack of knowledge of the underlying system image.

In conclusion, AUIs support widely established usability principles such as internal locus of control and reducing short-term memory load (Shneiderman et al., 2009). Successful implementations of AUIs to date include aggregation of online news content, such as Google News, and systems providing recommendations for online media, such as Netflix. Promising avenues for near-future AUIs include task management systems for e.g. scheduling appointments and communicating human interaction unavailability in the workplace, as well as adapting the UI of healthcare applications according to the computer knowledge and domain knowledge of the user. There are important usability challenges to consider, however, such as the tradeoff between ease of use of UI personalisation via automatic (or implicit) user data collection and the associated privacy issues, as well as avoiding filter bubbles which may prevent serendipitous information encounters and user learning.

References

Schneider-Hufschmidt, M., Kühme, T., & Malinowski, U. (1993). Adaptive User Interfaces: Principles and Practice. (M. Schneider-Hufschmidt, T. Khme, & U Malinowski, Eds.) Adaptive User Interfaces Principles and Practice, 351. North-Holland.

Jameson, A. (2009). Adaptive interfaces and agents. (A. Sears & J. A. Jacko, Eds.) HumanComputer Interaction Design Issues Solutions and Applications, 105-130. CRC Press.

Billsus, D., & Pazzani, M. J. (2007). Adaptive News Access. (P. Brusilovsky, A. Kobsa, & W. Nejdl, Eds.) Lecture Notes in Computer Science: The Adaptive Web, 4321, 550 – 570.

Begole, J. B., Matsakis, N. E., & Tang, J. C. (2004). Lilsys: sensing unavailability. Proceedings of the 2004 ACM conference on Computer-supported cooperative work, 6(3), 511-514. ACM Press.

Shneiderman, B., Plaisant, C., Cohen, M., & Jacobs, S. (2009). Designing the User Interface: Strategies for Effective Human-Computer Interaction (5th Edition) (p. 624). Addison Wesley.

Pariser, E. (2011). The Filter Bubble: What the Internet Is Hiding from You. ZNet, 304. The Penguin Press HC.

Ramachandran, K. {2009). Adaptive user interfaces for health care applications. Retrieved from http://www.ibm.com/developerworks/web/library/wa-uihealth/ on 1 May 2012.

Gajos, K. Z., Czerwinski, M., Tan, D. S., & Weld, D. S. (2006). Exploring the design space for adaptive graphical user interfaces. Proceedings of the working conference on Advanced visual interfaces AVI 06, 201-208.

McGrenere, J., Baecker, R. M., & Booth, K. S. (2002). An evaluation of a multipleinterface design solution for bloated software. Proceedings of the SIGCHI conference on Human factors in computing systems Changing our world changing ourselves CHI 02, 4, 164-170. ACM Press.

Advertisements

One thought on “Adaptive User Interfaces

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s