Adaptive UX Based Off of Individual Cognitive Models
From Verbal Analogies To Visual;
Adaptive UX Based Off of Individual Cognitive Models
Kai A. Horan
University of Texas
Dedre Gentner’s Structure Mapping: A Theoretical Framework for Analogy, (1983), established a method of mapping meaning, from the base of an analogy to the target, in order to communicate meaning beyond the literal words used. Gentner’s theory focuses on syntactic properties and utilizes a framework that supplies separation from literal meaning. The framework holds two principles; the focus is on the relations between two components, not just their attributes, and the “particular relations mapped are determined by systematicity, as defined by the existence of higher-order relations”, (p. 155). Gentner proceeds to explain that “degree of overlap” model, like used in Tversky’s (1977) model, is limiting as it works for “literal similarity comparisons” (Gentner p. 156). Furthermore, Gentner points out that often the “strength of an analogical match” is not directly dependent on the match’s attribute similarities but in other factors such as the known relation between the base components being able to describe a similar relation within the target. This governing structure still allows specific analogy one-to-one mapping but refers to the higher likelihood that dependent sets of predicates get carried over to the target domain. Gentner uses the example, “The atom is like the solar system” (p. 160), where the transferred base predicates are associations like “revolves around” and “attracts” but literal attributes like atmosphere, system, or center, etc., could potentially be transferred as overlapping similarities but are not in this framework. With this separation and consideration of relations earlier (semi-solely) attributes can be cataloged in this system but relations would be an after-process in an attribute comparison system. Also, when simply cataloging similarities, it would be hard to distinguish the intended meaning or decipher causal relationships due to the ability of the attributes to belong to many different inferences.
This shift in approach forged a new infrastructure of building blocks from which academics could create more precise systems when analyzing analogies and conveying relational meanings. According to Google search results, this paper alone has been directly cited over 6,100 times. Aside from citations and the evolving analogical studies, this classic paper has been used to further other areas of study such as education. Gentner herself evolved this research to apply structure mapping to metaphors with the knowledge that children can comprehend metaphors sooner than relational structures within domains (Gentner (1988). It also directly aided progress in categorization which can be applied to machine learning and other technical fields. For example, a key statement applicable to the User Experience (UX) field, Gentner refers to an observation that Ken Forbus and herself made of a subject’s evolving process when trying to articulate “the behavior of water flowing through a constricted pipe”, (p. 168). Gentner describes the subject’s narration as first a string of similarity matches, then analogies, and finally a declarative statement. Understanding the reasoning and behavior behind someone first using descriptive language, then trying to pair it with something s/he is already familiar with, and then drawing a conclusion, when attempting to comprehend something new, can aid the development of visual communications within a product. A designer could use this intuitive flow to help hasten a users thorough comprehension.
To elaborate more on this, first let me explain a little about my field. I am currently a Digital Product Designer and my fundamental task is to effectively communicate with an audience through digital means. In a more unpacked sense, product designers conduct user research in order to inform our designs with the goal of people being able to use them intuitively, resulting in a good user experience (this practice of UX can be applied to non-technology based products as well, ie. the Heinz bottle lid being on the bottom so it’s easier for a user to extract the ketchup). Our products are tools, whether they are intended for entertainment, work, or waking up, we are using them for an established purpose and expect them to work accordingly. It is the designer’s responsibility to think through the psychology of the flows and use-cases in which one may use the product. The aim is to make sure s/he does not have to employ extensive cognitive processing in order to complete her/his task. Here is the hitch, our research becomes increasingly diluted the further along the industry process you go. Intuition is a personal thing, some trends apply to the masses but when it comes down to how and when and for what purposes someone uses a smartphone, it is almost guaranteed that his/her use will differ from how/when/for what purposes the person to her/his right does. Initial UX research consists of mixed methods, the utilized assortment of which is chosen based on how each could inform on the issues the designer is charged with solving. These include tactics such as surveys, demographic analysis, analytics from current products in the wild, user interviews, user testing, etc. Each click noted as a data point in the research comes from an individual, let’s say it was Savanna’s click. We then take Savanna’s click along with hundreds of thousands of other clicks, find any similarities, extract general trends, and proceed to make design decisions that aren’t actually constructed by any one individual’s data set. Which produces a watered down good-enough experience for Savanna when she already has informed us on exactly how we could have made her experience better. However, there is a solution a few UX researchers and designers are starting to move towards and that is implementing an adaptive UX system. If we collect enough information on Savanna, and she allows us to collect insightful metrics, we could potentially create a model of her behaviors and preferences and then apply that to the product. Let us think in terms of accessibility and build from there, if Savanna is known to increase the magnification of her websites then we could alter the underlying grid and minimum type size of our product for her. If we knew she was colorblind, we could overhaul the hues and contrast ratings throughout the interface to make things easily distinguishable for her and better articulate the content hierarchy. If we knew she is more likely to swipe than tap when first interacting with a new application, we could change the navigation to open and close from a side drawer via swipe… and so on.
In a recent study by Ji, Yun, Lee, Kim, and Lim, An adaptable UI/UX considering user’s cognitive and behavior information in distributed environment, (2017), they measured the cognitive responses and modeled 122 users by running a series of both physical and cognitive tests. They then fed those results through a cognitive-based classification model and a behavioral-based classification model, both of those outputs went into the user profile and then adjustments were distributed among an array of devices. Next they monitored the interactions. Using that data with those individual profiles, they proceeded to develop a k-nearest neighbor algorithm, a decision tree algorithm, an artificial neural network algorithm, and a support vector machine algorithm in order to classify and predict the adjustments that could be made to improve experiences based on individual profiles. Specifically, they adjusted various combinations of elements within “5 categories, 22 sub-categories, and 63 sub-sub-categories of UI/UX” (p. 11). This resulted in high satisfaction and improved usability reports. This entire project was completed off of data we are able to extract from existing sources today. Some people are deterred by this and understandably-so. We still want to solve the application process though with the thought that people may welcome this tailoring feature as logical and helpful. An implementation example could be a contextual adjustment of a music application knowing that you are driving (due to monitoring location and cadence of movement) so it alters the interface to be easier to use at a glance with larger buttons and higher visibility/more concise labels. Then when you are using the same music application at 10:30 p.m. it would switch to a more informative layout allowing you to explore artists and create elaborate playlists.
The primary overlap between Gentner’s research and the current realm of UX is the shift from literal analogies to relational. For example, consider older versions of the mac notepad application, the interface emulated a yellow legal pad and the designers used a handwriting font. Originally well-received, if we skinned an application in the same way today it would come across as a lower-tiered and clunky product. In UX, we refer to this evolution as the user’s training. For example, a ecommerce power user will expect certain functionality of a site in order to have the experience be considered positive. Such functionality may include clicking on a product in order to gain more information about it, offering various payment options at checkout, receiving a confirmation email immediately after purchase, and getting a discount off of her/his first purchase if s/he subscribes to the brand’s newsletter. Where a novice’s expectations would most likely be considerably simpler (select item, add to cart, checkout) and s/he may be unaware of how to operate the power user’s preferred interface. The domain knowledge differs yet a designer today creates one product for both users. Gentner’s recognition of the hearer’s domain knowledge being a part of formulating the predicate set, and maybe more importantly the actual perceived meaning, (Gentner p. 159), is like a UX professional’s recognition that the end-user should help inform the form and function of the product. It is my opinion that one answer is implementing adaptive UX and UI based off of individual cognitive models. If we employed techniques like Gentner’s relational mapping to further flesh out the flows we are trying to replace with technology, we could develop a unique database per user to meet them where they are. This would tailor the experience so they could grow towards the latest version on their own time and if they wanted to, limiting pain points that may come from interacting with something foreign. Different people equates to different domain knowledge and different needs. It is our job as UX professionals to meet the user where they are and help improve their lives. If we start to incorporate predictive technology concerning a user’s mental model that could include how they may behave, how they make decisions, what associations they are likely to group, etc., we could better assimilate useful technology into more lives than just the highly trained.
Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155-170. http://dx.doi.org/10.1207/s15516709cog0702_3
Gentner, D. (1988). Metaphor as Structure Mapping: The Relational Shift. Child Development, 59(1), 47-59. doi:10.2307/1130388
Ji, H., Yun Y., Lee, S., Kim, K., and Lim, H. (2017). An adaptable UI/UX considering user’s cognitive and behavior information in distributed environment. Cluster Computing. 21. 10.1007/s10586-017-0999-9.
Tversky, A. (1977). Features of similarity. Psychological Review, 84, 327–352.