First Content Understanding and Generation for E-commerce Workshop

Call for Papers

Online experience on any e-commerce website is largely driven by the content customers interact with. The large volume of diverse content on e-commerce platforms, and the advances in machine learning provide opportunity for gathering insights from e-commerce content and build systems that can reliably generate content to improve shopper experience. The first workshop on content understanding and generation for the e-commerce aims to bring together researchers from industry and academia to discuss recent advances and challenges specific to these areas.

The workshop is a half-day event. We encourage submission of novel work-in-progress papers that show promising directions or demonstration systems, to facilitate discussion during the workshop. The workshop will solicit contributions related to the theme of supporting generation and curation of content for e-commerce which includes (but is not limited to):

Attending the Workshop

If you need visa to attend KDD in person, please request visa invite letter from ACM, as described here. The workshop is in person.

At least one author for the accepted paper should be registered for the workshop (even if presenting virtually). Authors can register for the conference here .

Important Dates

• Paper submission deadline: May 26th, 2022 June 9th, 2022
• Paper review begins: May 28th, 2022 June 11th, 2022
• Notification of decision: July 6th, 2022
• Camera-ready due: July 4th, 2022 July 21st 2022

Program

Virtual attendees can attend using this link . Workshop Agenda is as follows:

The following papers have been accepted as a talk/poster at the workshop. (*=corresponding author)

Five invited speakers shall talk about various topics associated with E-commerce content generation and moderation.

Aleix Martinez

Ohio State University

×

Shopping from images

The talk covers machine learning and computer vision algorithms to: 1. detect objects in cluttered and un-cluttered scenes, 2. classify these objects within a taxonomy of high-level and fine-gran object categories. and 3. identify objects visual similarity. Despite major advances, these three problems are still challenging. It will also review these difficulties and introduce our proposed solutions. I will the show results on a large scale application with clean as well as highly-cluttered real-world images.

Lydia Chilton

Columbia University

×

AI Tools for Design and Innovation

How can computational tools and AI help people be better at innovation and creative problem-solving? When solving a problem, people have the tendency to fixate on one problem or solution. If that one idea doesn’t work, they get stuck. To avoid getting stuck, the design process encourages people to have multiple ideas, and explore the space of possibilities before deciding on a problem or a solution. Although this works, it’s highly complex- requiring people to follow many threads at once. We show how AI and other computational tools can help simplify and speed up the most cognitively taxing aspects of the design process: 1. Collecting multiple partial solutions, 2. Synthesizing partial solution into multiple prototypes, 3. Quickly iterating on prototypes to produce an MVP.

Kristen Grauman

University of Texas at Austin

×

Style and Influence From In-the-Wild Fashion Photos

The fashion domain is a magnet for computer vision. New vision problems are emerging in step with the fashion industry's rapid evolution towards an online, social, and personalized business. Style models, trend forecasting, and recommendation all require visual understanding with rich detail and subtlety. I will present our work developing computer vision methods for fashion. Turning to the world stage, we investigate fashion forecasting and influence. Learned directly from photos, our models discover the “underground map” of a city based on the different clothes people wear, and they forecast what styles will be popular in the future by capturing how trends propagate across 44 major world cities. Building on this notion of fashion influence, we quantify which cultural factors (as captured by millions of news articles) most affect the clothes people choose to wear across a century of fashion photos.

Sam Wiseman

Duke University

×

Data-to-text Generation from Retrieved Nearest Neighbors

I’ll describe some recent work which attempts to generate text by directly cutting and pasting together pieces of text that have been retrieved from nearest neighbors, focusing in particular on a table-to-text generation scenario. The goal of this approach is to make text generation systems both more interpretable and easier to debug: problematic generations can be easily traced to the retrieved neighbors out of which they were created. I will also describe some ongoing work, which seeks to speed up the training of these cut-and-paste generation models.

Jacopo Tagliabue

Coveo