Accenture Global Solutions Limited

Ranjan Kumar Jena

Manish Sharma Kolachalam

Manisha Dubey

Siddesha Swamy

Kanakalata Narayanan

Rajnikant Dutt

Suket Kumar Jain

A system and method provide a trained model that uses vectorized word embeddings that are averaged or summed to form representations for sentences and phrases. The representations are processed in a Siamese neural network including multiple LSTM stages to find semantically related matches in catalogs for non-catalog queries. The model is trained using catalog data and randomized data using a contrastive loss function to generate similarity metrics for catalog-non-catalog pairs.

Growing community of inventors

Odisha, India

Ranjan Kumar Jena