Hack Session: Deep Learning for Search in E-Commerce

Nov 15, 2019


Auditorium 1

55 minutes

Deep Learning

Machine learning is used in almost every part of the system at major search engines like Google, Bing. However, most e-commerce websites are powered by search engines which provide excellent ROI and help in retaining and finally converting the user for a sale. In this hack session, come and witness how to use NLP based Deep Learning models to effectively design search platforms with a focus on e-commerce use case.


Section 1: Query Understanding

Description and Jupyter-notebook demonstration:

NLP based Deep Learning Models for finding the intent of a Query in a particular taxonomy/categories:

  • Multi-Label/Multi-Class Classification Model from scratch in Keras,
  • Feature Engineering in Spark Scala and pandas, Keras Functional APIs details in TF 2.0,
  • ImageNet moment of NLP – Latest invention in Word embeddings – ELMO and BERT

NLP based Deep Learning Models for Query Tagging with entities like Brand, Color, Nutrition, product quantity, etc. using Named Entity Recognition

  • Sequence modeling using Convolutional Random Fields (CRF),
  • Building a custom model in Tensorflow Estimator API,
  • Traditional Word Embeddings like Glove, Fasttext, etc.

Section 2: Related and Refined Searches

Description and Jupyter-notebook demonstration:

  • Related Searches: Building Sequence to Sequence (Seq2Seq) model using Long Short Term
    Memory (LSTM) concept of Deep Neural Network for predicting next search keywords,
    Sequential APIs details in Keras
  • Refined Searches: Similarity Search based on FAISS(Facebook AI Research) for giving a similar refined search keyword,
    comparing different word Embeddings e.g. word2vec, Fasttext, glove, etc. in popular AI framework such as gensim.


Section 3: ML Model Deployments in Production

  • Serving the model in production using Tensorflow and Keras


Key Takeaways:

  1. Learn about Natural Language Processing techniques like Word-Embeddings – BERT / ELMo, Bi-LSTM Networks.
  2. Application of NER and seq2seq modeling in the search domain.
  3. Get to know about various Multi-Class / Multi-Label Classification Problems related to Search Domains
  • Atul Agarwal

    Software Development Engineer

    Walmart Labs

  • Sonu Sharma

    Software Engineer

    Walmart Labs

Copyright 2019 Analytics Vidhya. All rights reserved