Machine Learning is the most sought after topic in the field of software Industry and would be considered the most in-demand technical skill going ahead. A Machine Learning algorithm analyze historical data, build models & predict outcomes. Machine Learning has got vast industry transformation capabilities and is used across sectors for business efficiency. Over the years it has been widely used for image & face recognition, it is also used in the e-commerce domain in the form of recommendation engines. Through item-item collaborative filtering algorithm Amazon recommends programming titles to a software engineer and baby toys to a new mother. ML is also used for cyber security, public safety & healthcare domain.
What is machine learning?
A machine learning algorithm identifies patterns in the data that deliver insights and increase the chance of better predictions and decision making. There are different types of machine learning algorithm.
The point of supervised, machine learning is to fabricate a predictive model based on the input and output data. A supervised learning algorithm takes a known set of input data with a labeled output. Let’s say you want to classify images into 2 categories male & female, you train the machine with a set of labeled images and build a model when a new image is shown to the model as an input the machine can classify the image and put them under respective category. Classification and regression techniques are used to develop predictive models in supervised learning.
Unsupervised learning finds hidden patterns or intrinsic structures in data where the input data are without labeled response. We take the earlier examples of images, and this time there is no labeled information the machine tries to make a sense of the images to draw an inference. Clustering is the most common unsupervised learning technique.
Reinforcement learning is a trial & error learning where the reinforcement agent learns from the consequences of its actions. Take the example of how a dog is trained. If the dog follows instruction and obeys we encourage by giving biscuits, if they don’t we punish by scolding. In the same way, if the system works well a positive value is given (i.e. reward) if not a negative value is given (i.e. punishment). The system which gets punishment improves by a trial & error method .
Usage of Machine learning
Machine Learning has the ability to identify patterns that humans tend to overlook. Machine learning has earned a substantial spot in the core territory of user experience, however, its contextual usage often gets unnoticed. Amazon uses a machine learning algorithm that will “learn which reviews are most helpful to customers” — that is, which reviews are real and which ones are fake . DeepText an unsupervised machine learning algorithm is used by Facebook interpret the meaning of posts and comments . If someone says “I like apple”, would it mean an apple fruit or an apple smartphone? By leveraging several deep neural network architectures like convolutional and recurrent neural nets DeepText can perform word-level and character-level based learning .
Consider spam filtering in emails, we hardly notice it, but the machine does not, the machine is trained to carefully reads & track every incoming email and send it back to the respective folder. The first known mail-filtering program to use a naive Bayes classifier was Jason Rennie's ifile program, released in 1996 . Naive Bayes algorithm uses Bayes’ theorem. The below equation calculates the chance that event A has occurred given that event B occurred.
The Problem Statement
User Experience as a discipline has attained a state of maturity over the years. As years passed the domain got matured in the areas of the design principles, design process, user’s expectation from a product, technical literacy of the users and the advancement of technology frameworks. However, data is still considered a far fetching strategy in formulating UX design decisions. Data Driven decision is considered more reliable than relying on human empathy as it gives a solid scientific ground to support a hypothesis. In a typical UCD process data is taken into consideration only if there is need to do a usability testing. The fact is not ignored that behind any product ideation process both quantitative & qualitative methods are applied by marketing & design teams to execute a research. However, when it comes to user experience design we often rely on the designer intuition to solve a problem. There is a significant amount of bias involved in such design decisions. Bias is not, by definition, always a bad thing . Human begins has a bias which is often unnoticed, but it influences our decision making. A phenomenon called Confirmation bias is the most common one when there is a pressure to do research quickly we might validate a design decision with few users that we have built without necessarily challenging our assumption .
The future of User experience will be driven by data. Data certainly has the ability to make human lives better by crafting meaningful interactions through predictive systems. Practically in each circle of human
life, a specific amount of data gets exuded through connections which get caught in some frame. Dietary habits, reading habits, travel habits when given a choice between mountains or a beach, browsing habits, gym habits are all data which are getting stored digitally. Habit changes over time due to the influence of various factors, like getting exposed to a different culture, due to the occurrence of a life changing event, newer learnings, attaining a state of mental maturity. If this change factor can be captured in some form then it can assist a designer to analyze a user better at a given point of time.
In the dynamic world human being come across several events which shape their short term & long term thinking. A life changing event often have a long-lasting effect in turning an individual’s thought process, mental model, likes & dislikes. Age is another important factor to be considered while designing systems as age carries various cognitive & physical disabilities like the decline in memory, reduced attention span, disorientation, apathy, and calmness.
Designers rely on personas to build a system. Users Persona is a representation of a character and acts as a reference point for designers. While designing products for a specific user group a user persona helps identify certain characteristics which involves the demographic profile of the user, personality, pain points, motivations, and technology savviness. Imagine, due to the robustness of the application the final product reaches the end customers after a year from the date of inception. In this time period, the user groups whom we identified initially as the potential customers might have experienced a variety of incidents that had a long lasting influence resulting in a behavioral change. This may result in the sharp decline of the demand, as users may not wish to use the product due to a change in their life habits resulted from an event.
Embrace Quantitative Research
Recognizing the future by sitting in the present is remarkably a challenging proposition for a machine. To do that machine needs to interpret the present and apply futuristic variables to predict a future outcome. User research forms the core of UX Design. A well-formulated research ensures that the product matches the expectation of the target user group and bring user satisfaction. While research is a costly thing to do, it requires a step by step execution to get the most of it. One alternate approach which can save both time & budget is to incline towards Empathic Design approach. Empathic design is a way to arrive at outcomes and analyze situations by putting yourself in the shoes of the user. Mapping end user emotions to yours to get an outcome has become a common practice in almost all design engagement. A designer often uses empathy & concludes a research. This often leads to biases as the designer would have got influenced by his/her own cultural and emotional setting in taking a design decision.
Imagine as an UX researcher you are trying to gaze the usability of a navigation menu. What would be the best thing to do, it depends on few contextual factors like availability of time & budget, access to user groups, implications of the usability test, etc. A quantitative researcher would collect the task completion statistics, how much time a user has spent on completing a task. On the other hand, a qualitative researcher would probably float a survey and take a generic user opinion. While a quantitative research uses deductive approach and tries to establish facts & numbers a qualitative researcher would use inductive approach & look for generalization. A quantitative research solely relies
on data and use a statistical model to prove a hypothesis. Over time as the data keeps growing a pattern is identified which can help build a predictive model.
Identify Patterns in the Data
Imagine a user plays golf only on the weekend, or read books during the morning. A user does online shopping only when his car is out for servicing. A user logs into the internet at night to check personal emails. A user does reckless shopping every first week of the month. A user visits the local pub every Thursday. If all of the above-mentioned activities are repetitive then it becomes a pattern. Once you start mining the data you can find a pattern which further helps in understanding a user better and predict actions. Machine intelligence can analyze thousands of variables across terabytes of data, potentially uncovering subtle but significant relationships .
Measure the Change
Change is inevitable in every corner of the human ecosystem. However measuring the change is the challenging piece and require a scientific model to handle this obvious. Smart user experience design helps us find new opportunities and make associations that we would have generally missed . We take up the past illustration where a client used to do rash shopping each month has abruptly quit doing. What could be the conceivable component of this change? Could be because of a location change, or the impact of some extraordinary occasion or attainment of maturity. If a machine learning model is able to predict this change, then we can build anticipatory design pattern to create meaningful experiences.
The world is moving towards machine intelligence and artificial decision-making systems. While research being the key to initiate such framework design gathering and analysing quantitative data become increasingly important to drive such development. In this article, we highlighted the need to move towards a quantitative model based solution so that it is easier to analyze data, recognize patterns, & build predictive models.
Robin, "Reinforcement Learning," Artificial Intelligence, 26th November 2009. [Online]. Available: http://intelligence.worldofcomputing.net/machine-learning/reinforcement- learning.html#.WPDIxvmGPIU.
"Teaching machines to read/comprehend websites, recognize and group faces, and reject fake reviews," Kurzweil News, [Online]. Available: http://www.kurzweilai.net/google-facebook-amazon- advance-machine-learning-applications.
S. M. Patterson, "Understanding Deep Text, Facebook’s text understanding engine," Networkworld, 1 6 2016. [Online]. Available: http://www.networkworld.com/article/3077998/internet/understanding-deep-text-facebooks-text- understanding-engine.html.
A. Abdulkader, A. Lakshmiratan and J. Zhang, "Introducing DeepText: Facebook's text understanding engine," fCode, 2 6 2016. [Online]. Available: https://code.facebook.com/posts/181565595577955/introducing-deeptext-facebook-s-text-understanding-engine/.
"Naive Bayes spam filtering," Wikipedia, [Online]. Available:
"The Bias Blind Spot and Unconscious Bias in Design," Interaction Design Foundation, [Online].
M. Eckstein, "https://medium.theuxblog.com/overcoming-bias-in-research-and-product-design-
f35a0d92496d," The UXblog, 7 1 2017. [Online]. Available:
A. Kwong, "Smart user experiences: Machine learning and the future of enterprise applications,"
Oracle Blogs, 6 5 2016. [Online]. Available: https://blogs.oracle.com/VOX/entry/smart_user_experiences_machine_learning.
About the Authors
Alipta Ballav is a senior UX consultant at Harman. He is seasoned & accomplished UX professional having 17 years of rich experience across domains. His current research is in the area of predictive User Experience and how can we leverage predictive models in designing and immersive & engaging UX.
Deepak Kumar Jain is currently a software engineer at Harman. He specializes in front-end development. His interest lies in working on challenging and innovative projects impacting lives of billions around the world through technological and human-centric design solutions. Also, his other technical interests are Artificial learning, machine learning and Human Computer Interaction (HCI).