Applying NLP: Use Cases of Natural Language Processing in Commercial Applications

E7d6e390a90513756419be75a43609ca?s=47 finid
June 28, 2019

Applying NLP: Use Cases of Natural Language Processing in Commercial Applications

My presentation will look at how companies are applying NLP, the impact of deep learning on performance and methodology, then a deep dive on an automotive use case leveraging chat bot transcripts for regulatory reporting.



June 28, 2019


  1. None
  2. Big Data & AI Conference Dallas, Texas June 27 –

    29, 2019
  3. The Power to Answer™

  4. What We Do From large text data collections... ...we extract

    valuable information... create actionable knowledge.
  5. What is Natural Language Processing? “I saw her duck”

  6. What is Natural Language Processing? “I saw her duck”

  7. Text Classifcation Automotive Regulatory Reporting

  8. Reporting to NHTSA

  9. Filtering Chatbot Complaints Electrical Issue Engine Issue “My car does

    not start” “I turn the key and hear a sound from the engine, but it doesn’t start”
  10. Approach We implemented deep learning techniques… …with semantic features relevant

    to the domain… …resulting in an accuracy level of over 90%.
  11. Demonstration Relating to Current Automotive Issues Mike. Is my 2015

    CRV in the takata infator recall? 2HKRM4H54FH622678.
  12. Spelling, Grammar, and Capitalization Demonstration Tochukwu. 2008 Honda pilot. Pls

    explain dis code for me P0743. Pls wat is dis fault code P0743.
  13. Implicit Categorizing of Sub-Systems Demonstration Larry Christiansen. We have a

    2014 Honda CR-V that vibrates at highway speeds. Is there a sevice bulletin for this problem
  14. Size of Training Data • More is always better: system

    performance relative to training data size is task dependent • 5,000 annotated customer complaints for training and validation data • Possible to leverage other resources to overcome data sparsity issues using transfer learning
  15. Type of Deep Learning • Convolutional neural network with a

    softmax output layer for multi-class classifcation • Trained custom word embeddings for the automotive domain with an unsupervised model • Framework in place for transfer learning for data sparsity issues For a more in depth look, read Yoon Kim’s paper “Convolutional Neural Networks for Sentence Classifcation”
  16. Questions