Redefining Style: Bridging Technology and Expertise for Sustainable Fashion Choices
The waste and environmental pollution of the fashion industry is a major cause of concern. We are buying more and utilizing the clothes purchased far less than we were a few decades ago. Some reasons identified for this include an increase in online shopping, lack of knowledge of what looks good on us, poor fit, etc. Cheaper clothes due to fast fashion means that poor decision making has low financial consequences but high environmental consequences.
One existing approach to better decision-making is the use of automated suggestions from recommender systems. Recommended outfits are generated based on your previous purchase or what other customers have bought. There are 3 main challenges with this approach to apparel recommendation:
- Consumers desire some variety in outfits and may need professional help.
- People with similar taste may not necessarily look good in the same outfits.
- Style trends change faster than the systems can adapt.
A different approach employs the services of personal stylists with expertise in what outfits work for their clients. Stylists’ decision-making processes for picking outfits can be complicated. Factors such as the lifestyle, appearance (body shape, hair color, and skin color), occasion of wear, weather, age, personality, etc. are all considered in making decisions. Integrating such knowledge and expertise in automated systems might help overcome some of the disadvantages of current automated recommenders. It is therefore important to evaluate how well existing style advice is representative of consumers aesthetic preferences. To this end, under the supervision of my advisor, Lucy Dunne, my PhD dissertation focused on answering the research question:
Does knowing a lot about what kinds of clothes should go with what kinds of body shapes for typical women help predict if young adults will think these outfits look good?
To answer this question, we needed to first understand how experts predict “good” outfits based on the relationship between body and garment attributes. We then had to assess the statistical significance of experts’ outfit predictions and young adults’ outfit preferences. Given that multiple factors affected outfit selection, we chose to focus on one factor to evaluate in this study—body shape. Therefore, we chose a sequential exploratory mixed methods research design. Evaluation was conducted in three phases: a qualitative phase, an instrument design phase and finally, a quantitative phase.
Analyzing Existing Authors’ Outfit Recommendations
In the qualitative phase, we conducted a content analysis on five books on style advice to identify the underlying principles influencing style advice based on the body shape of the individual. We found that authors try to:
- Balance wider shoulders by visually increasing hips.
- Balance wider hips by visually increasing shoulders.
- Define the waist/midsection.
- Draw the eye to preferred body features.
- Create a taller, leaner appearance by elongating the frame.
Preparing the Sample
In the second phase we designed a survey to evaluate the agreement between young adults’ perception of aesthetics and the authors’ underlying principles identified in the qualitative study. We selected outfits using examples from the five books with a focus on the use of style, lines and fit. Internally, an outfit recommended based on the qualitative study’s findings was assigned a score of 1 and the other outfit not recommended was assigned a score of 0. We then created 3-dimensional simulations of the selected outfits in the CLO3D software using avatars based on 3D scans of people. We simulated all outfits in a bland grey color to reduce effects of color as a biasing factor in evaluations.
Survey Administration
The survey design required placing multiple pairs of outfits in front of young adults and asking them to select the preferred one. We had three main challenges with conducting this type of image-based A/B test:
- Ensuring participants could see both outfits in the pair clearly on screen to make the comparison.
- Accuracy of results by integrating attention checks in an image-based survey.
- Collecting large numbers of responses quickly for multiple outfit pairs.
The Connect platform was ideal for us as we were able to collect 250 responses very quickly. It integrated well with the Qualtrics software that was used for survey design and the participant screening process was very efficient at ensuring participants were taking the survey on a laptop so they could clearly see the outfits. The overall quality of responses was really good as most participants passed the attention checks.
Research Findings
We analyzed the results of the survey to compare participants’ outfit preference with authors’ suggested outfits. Our findings suggest that there might be some value in using authors’ advice to determine aesthetically pleasing outfits with a hierarchical structure where certain principles are more important than others. The principles might also be more important for some body shapes than they are for others with the subtlety of the change in garment attribute being an important factor to consider.
As an initial foray into an empirical evaluation of authors’ advice on aesthetics for body shapes, this study shows the importance of evaluating the rules we integrate into designing recommenders. It suggests the need for more formally defined theories for dressing advice. Since this was a narrowly defined study focusing on body shape and using 3D simulations, future work will seek to understand how integrating other personal features such as skin color, hair color, etc. affects aesthetic preference. An evaluation of the effects of other clothing attributes would also be beneficial.
References
Ajmani, S., Ghosh, H., Mallik, A., & Chaudhury, S. (2013). An Ontology Based Personalized Garment Recommendation System. 2013 IEEE/WIC/ACM International Joint Conferences on Web Intelligence (WI) and Intelligent Agent Technologies (IAT), 3, 17–20. https://doi.org/10.1109/WI-IAT.2013.143
Dahunsi, B. O., & Dunne, L. E. (2021). Understanding Professional Fashion Stylists’ Outfit Recommendation Process: A Qualitative Study. In N. Dokoohaki, S. Jaradat, H. J. Corona Pampín, & R. Shirvany (Eds.), Recommender Systems in Fashion and Retail (pp. 139–160). Springer International Publishing. https://doi.org/10.1007/978-3-030-66103-8_8
Landia, N. (2017). Building Recommender Systems for Fashion: Industry Talk Abstract. Proceedings of the Eleventh ACM Conference on Recommender Systems, 343–343. https://doi.org/10.1145/3109859.3109929
Landia, N. (2021, March 29). Why is Fashion Different? Part 1: Quantifying Fashion Data. Dressipi. https://dressipi.com/blog/why-is-fashion-different/
Vaccaro, K., Agarwalla, T., Shivakumar, S., & Kumar, R. (2018). Designing the Future of Personal Fashion. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems, 1–11. https://doi.org/10.1145/3173574.3174201
Zhang, J., Terveen, L., & Dunne, L. E. (2021). The Ensemble-Building Challenge for Fashion Recommendation: Investigation of In-Home Practices and Assessment of Garment Combinations. In N. Dokoohaki, S. Jaradat, H. J. Corona Pampín, & R. Shirvany (Eds.), Recommender Systems in Fashion and Retail (pp. 101–116). Springer International Publishing.
