×


GET IN TOUCH WITH US
Boston | London | Berlin | Mumbai

PR or Media inquires: communications@truefit.com

Corporate HQ: 617.848.3740
General Inquiries: 877.FIT.TRUE


HQ: Boston
60 State Street
12th Floor
Boston, MA 02109

View Map

Europe HQ: London
Thanet House
231-232 Strand,

London, WC2R 1DA


True Fit India
No.306, A-Wing, Boomerang, 
Chandivali Farm Road, Chandivali,
Mumbai, Maharashtra, India, 400072

Contact Us

We'd love to hear from you.

×

Schedule a Live Demo

 

True Fit Data Science: THREE Technical Papers Accepted at the 2019 ACM Conference on Recommender Systems!

September 12, 2019

We are very pleased to announce that we have three technical papers accepted at The ACM Conference on Recommender Systems (RecSys) September 16-20th in Copenhagen, Denmark! RecSys is the premier international forum for the presentation of new research results, systems and techniques in the broad field of recommender systems.

Not only did we have one paper accepted for the main conference, but two additional papers will be showcased at the FashionxRecSys Workshop.

Keep reading to learn about the impressive research our Data Science Team has been conducting in recent months.

“Users in the Loop: A Psychologically-Informed Approach to Similar Item Retrieval”

The paper titled, "Users in the Loop: A Psychologically-Informed Approach to Similar Item Retrieval" was accepted into ACM Conference on Recommender Systems. The papers submitted for the general conference have a 19% acceptance rate - a huge feat for the True Fit team! 

This paper explains how members of True Fit’s Data Science Team developed an algorithm that simulates how consumers make judgments about the similarity between products when shopping. Through this research, the team is better able to understand which product attributes matter most to consumers when they shop. 

One of the most frustrating experiences consumers face when shopping online is clicking on a product and realizing the item isn’t available in their size. In these instances, many retailers and brands have implemented “item-item” similarity algorithms, which find similar items to recommend to the user instead, in the hopes of helping the shopper find alternative products that they will likely love and keep. 

Often, item-item similarity algorithms are based on the idea of geometric distance — or how far apart two points are in space (think of the length of one side of a triangle). However, geometric distance doesn’t capture the uniqueness of human judgments of similarity. The research conducted in this paper explores how incorporating unique, human psychology into algorithms can lead to better product recommendations.

Amy Winecoff led the research for this paper, as the team integrated ideas from psychology and behavioral economics into a novel item-item recommendation algorithm. The team also developed a custom framework to test out the algorithm in large groups of users. Amy’s academic background in psychology and research methods was integral to developing the algorithm and testing framework that incorporates the ways real people think.

About the Authors 

Amy Winecoff: Senior Data Science Manager, True Fit
Matthew Graham: Senior Data Science Manager, True Fit
Florin Brasoveanu: Principal Scientific Software Developer,True Fit
Pearce Washabaugh: Senior Data Science Manager, True Fit
Bryce Casavant: Data Scientist, True Fit

"Assessing Fashion Recommendations: A Multifaceted Offline Evaluation Approach"

The paper "Assessing Fashion Recommendations: A Multifaceted Offline Evaluation Approach" will be featured in the FashionxRecSys workshop on Friday September 20, 2019. This research explains the importance of personalization, particularly for fashion retailers, in driving 1-to-1 shopping experiences to each and every shopper. 

Traditional recommendation systems are evaluated based on the relevance of the clothing that they recommend to shoppers (i.e., do shoppers “like” their recommendations). The research for this paper focuses on the unique connection that people have with fashion and apparel, and how consumers want clothing recommendations that are personalized (i.e., unique) to them in addition to simply being relevant to them. 

By only recommending popular items, a retailer or brand can provide recommendations that score very high on relevance, but these recommendations will not deliver a truly personal shopping experience because they do not also include options that are unique to the individual shopper’s preferences. 

Additionally, many fashion recommendations systems do not take into account the distinction between new versus existing users in their evaluations, which is a major problem given the large number of online shoppers who are buying clothes on a retailer or brand’s website for the first time.

Only by performing evaluations that take both relevancy and personalization into account in addition to distinguishing between new and existing users can retailers and brands truly understand how well their fashion recommendation systems are performing.

The research for this paper was led by Jake Sherman, a Data Scientist at True Fit who has worked on building and evaluating fashion recommendation systems and modeling shopper behavior.

About the Authors

Jake Sherman: Data Scientist, True Fit
Amy Winecoff: Senior Data Science Manager, True Fit
Chinmay Shukla: Data Scientist, True Fit
Su Zhang: Data Scientist, True Fit
Rhonda Textor: Head of Data Science, True Fit

“Automated Fashion Size Normalization”

The paper, “Automated Fashion Size Normalization” was a joint research effort by True Fit’s Data Science team and Applied Research Scientists from our lead investor, Georgian Partners

The research for this project was conducted over a sixteen-month period and highlights the close connection we have with our investors, and the power of Georgian Partner’s  IMPACT Team. The Georgian Partner IMPACT Team helps portfolio companies take advantage of foundational tech trends to create value and accelerate their business across a variety of workshops, research projects and more.

The research conducted for this paper focuses on how to teach a computer to normalize sizes. For a human, size normalization is straightforward; “S”, “SM”, “SML”, and even “P” are all recognized as abbreviations for the same size. However, a computer recognizes each of these variations for the same size as different entities.

The teams worked to replace a crucial, and expensive to maintain, prerequisite to make size recommendations by automating size normalization so all variations of different sizes were understood to mean the same common size. The system also recognizes discrepancies across geographies and is able to learn any differences that may exist between geographies. This paper will be featured at The FashionxRecSys workshop on Friday September 20, 2019.

About the Authors

David Wayne: Senior Data Science Manager, Consumer Engagement, True Fit
Eddie S.J. Du: Applied Research Scientist, Georgian Partners
Chang Liu: Applied Research Scientist, Georgian Partners

Have you heard of Matched Cohort Analysis? Check out our most recent Whitepaper to learn more about the importance of MCA in understanding the value of True Fit and consumer loyalty overtime.

Download