Telling Stories
OCEL.AI uses a storytelling approach to create use cases that will guide users through the process of machine learning. These steps include storytelling, data sharing, ML modeling, ML applications, social and cultural implication, and ethics.
[ivory-search id=”3639″ title=”AJAX Search Form”]Q1: What is a use case?
The purpose of Phase 1 Storytelling is to create a “persona” (if we use the terminology in advertising), or a “feature story” (if we use the terminology in journalism). If you are studying computer science, it is called a “user story that has been used as a means of guidelines in the development of software systems.” An example of user story is shown in https://www.visual-paradigm.com/guide/agile-software-development/what-is-user-story/ and https://www.youtube.com/watch?v=nE8ALJ2M004
In advertising, we create a “persona” to summarize consumer insights, and guide the creative process. Here is an example of a “persona” in advertising:
“John is a 37-year-old man living in Kansas City, Missouri. John has four children, works two jobs, and is married to Amanda, a stay-at-home mom. John is also a pianist of a band that plays for his church. The entire family goes to church most Sundays….”
In journalism, we have “feature stories” like this:
“Dressed in jeans, …, Beyoncé is sitting on a sofa in her assistant (and cousin) Angela’s Manhattan high-rise apartment. On a sunny May afternoon, wearing very little makeup, she looks even younger than her then 23 years. … Upstairs, we’re waiting for a pizza and talking about shoes… .”
Q2: How do machines learn? What is a right question to train machines?
No matter what your research question is, small or big, specific or general, the way how “machines” (computers) learn, is to learn something useful iteratively or repeatedly from relevant data, for maybe 100,000 times. So, a machine cannot directly answer a question like this:
“ What are the solutions to inequality of education resources in KC metro?”
Rather, a machine learns from data (e.g., images, texts, audio, video, etc.) representing what the education resources are available to Cabel, Jordan, Jaelon, Anteda, …, etc. until it sees patterns and models. These patterns and models can eventually address the issue of inequality in education resources and recommend solutions. In other words, the “right” questions for machine learning are:
- What is the best after school program for Cabel’s kids?
- Which summer break camp should Anteda send her children to?
However, we should provide relevant data (such as Cabel’s info – age, income, home, work schedule, or school information, etc.).
In journalism or advertising, it is not uncommon for us to conduct interviews and consumer research to create a “persona” or write a feature story. Using a role-playing approach, you want to write use cases that can answer questions like this: “How do I choose the best spring break camp for my kids?”
So, if you want to know “who are impacted the most by contaminated drinking water in Kansas City?”, the question of your use case is: How do I know whether my drinking water is contaminated?
Q3: How useful is a use case to machine learning?
If you are in advertising and journalism, you may want to ask whether a use case based upon just “my experience” is good enough for machines.
The answer could be Yes! It is likely that my story shares similarities with stories of other people in the same social stratum. My story provides clues to machines to learn, particularly regarding key aspects, factors, and processes. As a result, the machine can use the frame of “my story,” to read more stories from similar people, and acquire the pattern of the decision-making process, and thus make predictions/generalization.
The answer could be No! How about people who are outside of my social stratum? For example, I am living in the best school district. My accessibility to education resources is much greater than those in the other parts of the city, and I may not even care about inequality of education resources. It just does not matter that much to me! In contrast, a single mom who works two jobs and lives in the urban center is desperate to find a safe, affordable, and educational preschool for her kids where such resources are sparse.
“My story” is not her story!
No worries. You simply need to do the same thing that we often do in audience research: talking to a single mom who fits that profile, and creating a “use case” based upon her experience.
Q4: What is a good first exercise?
However, in this initial round you will build the ML application based on your own story and data about yourself. Your model will give you an answer like “X school is the best for you” based on your story/context. What we are doing is a context-aware ML model. Thus, as the first step, it is easier to start with “my story,” and run through all the steps and phases for right now.
In the future, we can write use cases like “Cabel’s story,” “Emily’s story,” etc. when we are more familiar with the steps and phases of machine learning.
Q5: What are the domains for use cases?
Use cases can be about anything, like vacation, shopping, education, healthy lifestyles, communities, etc. …. The answer depends upon the domain areas of your course.
If you are designing a use case exercise for Data Journalism, here are a few examples:
- A student project may be about mental health issues in KC. The use case question could be: which psychiatrist should I see in KC?
- If you are interested in neighborhood and housing issues like affordable housing. The use case question could be: “Which apartment is the best for me to rent?” “Me” could be a low-income single mom.
- Another example is gun crimes in Kansas City. The use case question could be: “where is the safest neighborhood for me to live in Kansas City?”
- If the social issue of interest is urban poverty, the use case question could be: “What should I do to lift my kids out of poverty?”
If the subject of the course is advertising, the domain areas for the use case could be about consumer goods, fashion goods, personal electronics, vacations and trips, etc..
Q6: What is a framework to write use cases?
The well-known “5Ws + H” journalism storytelling technique could be a good framework: Who, What, When, Where, Why, and How.
You can use this 5Ws+H template to start your first use case: Start Here.
There are examples of use cases:
- EduKC: k-12 educational camps during the spring break
- How can I buy a used car as a college freshman?
- Affordable and safe housing resources for low-income students and residents in the Ypsilanti
- Road condition in Michigan
Who, what, and where, by what helpe, and by whose,
Why, how and when, doe many things disclose.
— The Arte of Rhetorique, 1560
Photo by You X Ventures on Unsplash