Diogo Goncalves

Patrick John Chia, Giuseppe Attanasio, Federico Bianchi et al. · stanford

The steady rise of online shopping goes hand in hand with the development of increasingly complex ML and NLP models. While most use cases are cast as specialized supervised learning problems, we argue that practitioners would greatly benefit from more transferable representations of products. In this work, we build on recent developments in contrastive learning to train FashionCLIP, a CLIP-like model for the fashion industry. We showcase its capabilities for retrieval, classification and grounding, and release our model and code to the community.

Patrick John Chia, Jacopo Tagliabue, Federico Bianchi et al. · stanford

Product discovery is a crucial component for online shopping. However, item-to-item recommendations today do not allow users to explore changes along selected dimensions: given a query item, can a model suggest something similar but in a different color? We consider item recommendations of the comparative nature (e.g. "something darker") and show how CLIP-based models can support this use case in a zero-shot manner. Leveraging a large model built for fashion, we introduce GradREC and its industry potential, and offer a first rounded assessment of its strength and weaknesses.

Diogo Goncalves, Liweu Liu, Ana Magalhães

Using embeddings as representations of products is quite commonplace in recommender systems, either by extracting the semantic embeddings of text descriptions, user sessions, collaborative relationships, or product images. In this paper, we present an approach to extract style embeddings for using in fashion recommender systems, with a special focus on style information such as textures, prints, material, etc. The main issue of using such a type of embeddings is its high dimensionality. So, we propose feature reduction solutions alongside the investigation of its influence in the overall task of recommending products of the same style based on their main image. The feature reduction we propose allows for reducing the embedding vector from 600k features to 512, leading to a memory reduction of 99.91\% without critically compromising the quality of the recommendations.

Diogo Goncalves, Miguel Costa, Francisco M. Couto

Nowadays, Cable TV operators provide their users multiple ways to watch TV content, such as Live TV and Video on Demand (VOD) services. In the last years, Catch-up TV has been introduced, allowing users to watch recent broadcast content whenever they want to. Understanding how the users interact with such services is important to develop solutions that may increase user satisfaction , user engagement and user consumption. In this paper, we characterize, for the first time, how users interact with a large European Cable TV operator that provides Live TV, Catch-up TV and VOD services. We analyzed many characteristics, such as the service usage, user engagement, program type, program genres and time periods. This characterization will help us to have a deeper understanding on how users interact with these different services, that may be used to enhance the recommendation systems of Cable TV providers.

Diogo Goncalves, Miguel Costa, Francisco M. Couto

Recommendation systems are being explored by Cable TV operators to improve user satisfaction with services, such as Live TV and Video on Demand (VOD) services. More recently, Catch-up TV has been introduced, allowing users to watch recent broadcast content whenever they want to. These services give users a large set of options from which they can choose from, creating an information overflow problem. Thus, recommendation systems arise as essential tools to solve this problem by helping users in their selection, which increases not only user satisfaction but also user engagement and content consumption. In this paper we present a learning to rank approach that uses contextual information and implicit feedback to improve recommendation systems for a Cable TV operator that provides Live and Catch-up TV services. We compare our approach with existing state-of-the-art algorithms and show that our approach is superior in accuracy, while maintaining high scores of diversity and serendipity.