Capstone: To find a good topic

This is a story about what I do when all capstone topics get killed. Initially, I conduct a comparison approach to compare several topics I am interested in. After knowing that none of them is an option anymore, I take a step back to understand what I’m interested in and the pattern of it. This article is about how I conduct my top-down approach to search for my capstone topics.

Methodology journey map

Roadblock: All my topics get killed

I was initially passionate about the three topics below and conducted a detailed comparative analysis. However, when professors announced that NO HUMAN SUBJECT RESEARCH in Minerva, my excitement burned into ashes, because none of them fit into the description.

My topic comparison

Fight my way out Big List of Brainstorming

I knew that the difficulty of the capstone wasn’t only about the work itself. I will have lots of mental combat such as demotivation and procrastination, and my first challenge just came. There were two challenge

• I’m demotivated and don’t know what to do now.
• I cannot tell what topic I’m interested in anymore, because I am just desperate to find something.

At this stage, I might avoid the problem by procrastinating or easily settling on a topic I didn’t like just to pass it through. Neither of them was a good solution.

Though a capstone topic needed to be sharp and concise, since my option was limited, and my mind was constrained, I needed to find a way to break free. Hence, in one afternoon, I went to a cafe next to London MUJI, and decided to set aside all the constraints and worries in my mind, but brainstormed with my intuition and whatever related inspiration from research papers.

Big List of brainstorm

Topic Categorization

Thanks to Prof McAllister, she gave me two ideas to approach this scary big list, topic categorizations, and scope manipulation. Categorization involves three simple steps.

1. Create multiple dimensions to categorize the big brainstorm topics list
2. Find the relationships among categories.
3. Use the relationships to withdraw insights

1. Create multiple dimensions to categorize the big brainstorm topics list

Categorizations

Perspective/Beneficiary: Who needs this solution?

Since I’m doing an ed-tech project, I think of who will need and be benefited from this solution. Is it teachers, students, the business sector, or just my personal interests? I use this dimension to gauge the direction of the impact that can happen.

Deliverable/Process: Is this topic product, research, or methodology focus?

From seniors, I learn that even a meaningful topic can cause distress if the majority of the time you spend on the work process isn’t enjoyable. Therefore, I need to dissect the process to know what I will be doing, and also what output I will finally produce.

• Product focus: solutions or customization
• Research focuses

Interests: How much do I like about the project?

I complete this section purely based on my instinct. Even though my mind is clouded when I tried to analyze a specific topic, my instinct when facing a wide range of topics still function. I categorize all the questions by my gut feelings.

In class, I asked Prof. Stern, “How do you find your interests? After going through so many problems, I’m not sure what is fun for me anymore.”

Prof said, “ Fun is the topic that gives you guilty pleasure.”

When he said so, I immediately recall the moment of creating an algorithm and creating a curriculum. I learn that though the capstone topics might be new for me if we decompose the topic, elements that excite you might exist. The process of finding a suitable topic can be a prediction question. You use your personal data to predict a topic that you will enjoy.

2. Find the relationships among categories.

After instinctively categorizing all the topics based on the interests level, I use color code, the colored circle above, to see its relationships with other categories. Here, I marked the topics that I’m highly interested in with yellow circles. I derived conclusions such as I’m not interested in business-related topics.
There is even much more that one can apply data analysis skills to. For instance, you can encode high interests as 1, medium interests as 2, and low interests as 3. Now, a new dataset is created to find the correlation between interests and product focus.

3. Use the relationships to withdraw insights

I observed the pattern of the distribution of my high-interest topics in a different distribution, and I organize the following insights.

1. Don’t do business
2. It needs to be beneficial for me.
3. It needs to have the methodology I want to do.
4. High interests involve lots of personal experience.
5. Medium interests are either somehow personal or imaginative(things I don’t know e.g. VR/AR)
6. My capstone needs to be
- If it is research, it needs to be strongly tied with learning strategies.
- If it is a product, it needs to give me creative space that involves content creation.

Scope Manipulation

The advantages of manipulating the level of the topics

• High-level topic: Allow freedom for investigation. e.g. How can we understand important factors about students’ performance and formulate suggestions?
• Granular level topic: Connect to personal experience, easy to relate and imagine. e.g. How does e: Investigating the effects of professors’ linguistic feedback on student grades.

I learn that my old topic: How can we understand important factors about grade prediction and take an effective intervention? This isn’t a flexible enough topic, because the only effective intervention I was imagining was that a professor will talk to students about their problem when the machine predicts they are going to fail a class. This sounds quite dumb to me (How can a prof or a student cannot tell who is gotta fail in the class? 😂).

Prof McAllister inspired me with a similar but broader framing of the questions: How can we understand important factors about students’ performance and formulate suggestions?

The flexibility involves

• What is performance? Is it only pass/fail, or grades that I can classify into more categories(Example: GPA < 3.1 , 3.1–3.3, 3.3–3.5, 3.5–3.7, > 3.7)? This allows me to investigate how to improve someone with a 3.1 GPA to 3.5 for instance.
• What are the reasons for underperformance and overperformance? Is it because of time management (e.g. Start the preclass early)? Career interests(e.g. need to go to grad school)? Relationship with the prof?
• How to measure the important factors?
  ○ Time perspective: before, during, after class, or other activities that students can spend time on.
  ○ Social role perspective: profs, students, TAs, parents, students’ friends can all be factors of influence.
  ○ Functional perspective: curriculum designers, forum/platform designers, software engineers, grade/assessment designers
  Brainstorming all the potential factors give me a direction of where to collect data from.
• What is a suggestion? Is it an office hour, a structural study session, a one-on-one peer tutor system, a piece of advice from successful seniors, or a StackOverflow of how to approach the class?

Since every combination of those questions can be a topic, this process gives me a wide variety of potential. For instance, if we focus on

• What is performance? Pass/Fail.
• How to measure the important factors? Social role perspective: profs.

We can interview profs, and ask them what might be the red flag you will identify if a student is going to fail a class. We can extract some insights and dive into a more specific question, and use data analysis to verify or falsify the hypothesis.

Process journey map

My Next Step: Testing

From Topic categorization, I learn that if my topic is research-related, it needs to be strongly tied with learning strategies. Now, I can go test out if this topic is feasible for a data analysis project.

From Scope Manipulation, I learn that I won’t be stuck on an idea when I’m trying to dive into a more niche focus. If the niche topic doesn’t sound feasible, I can always manipulate the level of the scope. Make it more general, and try out different combinations. I assume the testing stage will be an iterative process to find a topic that is feasible, fit the constraints, and is aligned with my interests. I will try out one iteration a week.

Go testing now! 🔥