UW 2022 thesis project ideation
Project Overview
In the last post, I suggested the pain points of the current recommendation system. The existing system is based on the categorized products. This means the system can’t consider the user’s design preference. Therefore, I was recommended a random design product that did not reflect my design preference. From this, I wanted to focus on an enhanced user experience on the recommendation system so that users can get more refined results on the shopping app.
Ideation
MBTI Test
MBTI test — Myers-Briggs Type Indicator is a personality type test, which classifies people’s personality into 16 types and defines them as one of them through tests.
As I mentioned in the last post, through my thesis, I would like to classify people’s color preferences and let them know their type. So that people can recognize their preference type and apply it to their everyday life.
However, in the last quarter, the methodology was a little vague about how I could make them know their design preference. In this quarter, I focused on the “Why” this project needs and How” can I do this project.
Building a chatbot
The last December, we had the In the MDes Poster Show session; the main feedback I’ve got was, “What is your final deliverable going to be?” Through the poster show, I realized that my thesis concept is challenging to explain and uncertain for others. Therefore, I tried to decide the final output first. Through ideation with my committee members and AI expert, I determined the final deliverable as a chatbot that can diagnose users’ color preferences on the interior design.
Methodology — Clustering
To apply the ML technology to design work, I decided to try using “Clustering.”
Clustering refers to dividing an object into several clusters (subgroups) when it is given. Through this process of grouping objects, the goal of clustering is to ensure that members within a cluster are close to each other or similar to each other and that members between two different groups are not far from each other.
When we study history, We first distinguish by a period of history and then explore the events in it. This is because it is easier to understand each case in the significant flow.
From the learner's perspective, this strategy same goes for machine learning too. The system divides the data into large groups as the first step in data analysis. Then it gives a sense of belonging to each data, which is called “labeling.
This is what the clustering result looks like. You can see the given data were classified into three groups based on their feature. In machine learning, the degree of similarity of data values may be calculated and expressed as a distance. If the data values are similar, they are located at a close distance. Conversely, if the data values are different, they will have to be located far away from each other.
Advanced Research
I found two papers at IEEE that classified the given visual data based on the correlation features.
This paper presents a comprehensive study of deep correlation features on image style classification. Inspired by that, correlation between feature maps can effectively describe image texture, and we design various correlations and transform them into style vectors, and investigate classification performance brought by different variants. (Chu and Wu, 2018)
Chu and Wu classified the oil painting style based on their visual features.
In this paper, we deal with handbag recognition. It is a challenging problem due to the inter-class style similarity and the intra-class color variation. We focus on developing discriminative representations of handbag style and color (Wang et al. , 2016)
Wang et al. designed a system that can recognize the handbags based on their color and features.
Through these researches, I could find the possibility of the invisible notion, “the style and feature,” clustering through AI technology.
Exploration
- storyboard
After the conversation with my thesis chair, I realized the necessity of using the scenario of the chatbot. The primary situation when users need my chatbot would be the move-in situation because people usually have to decide the new interior of their new place. I focused on this case and have been developing the storyboard.
This is the final storyboard I worked on in the winter quarter. I reflected my friend’s case a lot so that many people could empathize with it. However, I keep finding a more specific and convincing scenario.
2. Building a database
Based on the use case scenario, I also designed the whole system flow. My thought on diagnosing the type was mainly two ways. The first was analyzing the user’s photo album and diagnosing their color preference with AI technology. The second way was to upload their current room picture on the chatbot app, and it analyses that picture and lets the user know their color preferences. Based on the advice from the AI expert, one of my committee members, I decided to take the second method because it is technically more realistic.
Also, I tried to use ML technology. Actually, in this process, I experienced some barriers. I decided to use the most popular AI coding platform Colabbortive Laboratory (Colab) which is serviced by Google. For using Clustering, the database is essential.
First, build a data base and then, put it into the Colab as program library. Since I needed to make an image data base, I used a program named “Extreme Picture Finder”. This is the program that can automatically download a huge amount of similar images on the multiple websites.
I collected about 7000 pictures of living through this program. However, I found some problem on this. I asked for an advice on my project to one of my friends who know to use AI with Colab.
He said, if I want to build a database, I have to refine and label on all of those images. So to build my own bata base, I have to adjust the size and perspective of about 7000 images. In addition, I have to modify all of those names based on the standards that I set for the project.
I thought this is physically impossible work I can done within about 3 months, So I decided to use similar type of ready-made database for my project.
Next Step
I received feedback on my paper at the final paper meeting with the committee members this winter. They said that what I’ve done so far is too much biased towards engineering based. As a designer, I need to more focus on the user-experience more than what technology I will use. So I rebuilt the plan for next step.
Why user need to use this? I have to more focus on the user study and specific problems. Drill more about the user persona is necessary. I only made user storyboard based on the nearby experiences for the project. The convincing case is needed for my project. Also, I need specific benefits that users can get from using the app I’m going to make.
Reference
Kim J, Lee J-K. Stochastic Detection of Interior Design Styles Using a Deep-Learning Model for Reference Images. Applied Sciences. 2020; 10(20):7299. https://doi.org/10.3390/app10207299
W. -T. Chu and Y. -L. Wu, “Image Style Classification Based on Learnt Deep Correlation Features,” in IEEE Transactions on Multimedia, vol. 20, no. 9, pp. 2491–2502, Sept. 2018, doi: 10.1109/TMM.2018.2801718.
Y. Wang, S. Li and A. C. Kot, “On Branded Handbag Recognition,” in IEEE Transactions on Multimedia, vol. 18, no. 9, pp. 1869–1881, Sept. 2016, doi: 10.1109/TMM.2016.2581580.
