For those of you keeping track of where Coveo’s research scientists are presenting – we are thrilled (and proud) to announce that Coveo will be at NAACL 2021 with two contributions, a theoretical paper, and an industry paper, both focused on building grounded language models in ecommerce. That’s right, both our submissions were accepted!
The 2021 Annual Conference of the North American Chapter of the Association for Computational Linguistics – (NAACL for friends) – originally planned in beautiful Mexico City – will be held online this year, June 6-11.
NAACL is one of the most prestigious venues for theoretical and applied research in Natural Language Processing technologies, with high-impact contributions from leading institutions and organizations around the globe. Ecommerce use cases feature prominently in the program, giants like Rakuten and Amazon are participating.
Our Contributions in Grounded Language Models
Our contributions share a simple thesis, and leverage a simple (in retrospective!) insight, as both papers go on and explore the consequences of the use of product vectors to build grounded language models:
- Query2Prod2Vec: Grounded Word Embeddings for eCommerce explains how product vectors provide accurate lexical representations for words such as “Nike,” “shoes” etc. The focus of the work is on short queries and, in particular, on the efficiency gain that our method provides compared with traditional NLP approaches based on co-occurrences.
- Language in a (Search) Box: Grounding Language Learning in Real-World Human-Machine Interaction focuses on compositionality and generalization: if a machine knows the meaning of “Nike shoes” and the meaning of “Adidas”, can it infer the meaning of “Adidas shoes” – even if it never saw this sequence of words before?
Acknowledgments
As with other research initiatives before – such as the SIGIR Data Challenge or last year’s RecSys paper – our NAACL 2021 papers are the result of a successful industry-academia collaboration. In particular, a huge thanks to our co-author Federico, from Bocconi University, without whom this achievement would not have been possible!