Anthill Inside 2018

On the current state of academic research, practice and development regarding Deep Learning and Artificial Intelligence.

Participate

Deep Learning - An example implementation

Submitted by Krishna Bhavsar (@krishnabhavsar) on Tuesday, 17 April 2018


Preview video

Section: Full talk Technical level: Beginner

Abstract

In this talk I intend to showcase one of the problems I solved using Deep Learning framework recently. Resume Classification for a recruitment consultant agency. I shall go through the multiple approaches in which I tried to solve that problem, the obstacles I faced and finally how I came to the final solution.

Outline

1st Slide
AI vs ML vs Deep Learning
2nd Slide
ANN vs Neurons in human brain
3rd Slide
Example problems that can be solved using Deep Learning
4th Slide
Problem Definition for resume classificaiton
5th Slide
Formulation and generalization of a typical Candidate Resume
6th Slide
Intial Data modelling for the solution
- Number of samples - Entity Recognition algortihms - Preprocessing - Working at offset level - Feature encoding 7th Slide
Journey to the final solution
1. Start with a smaller problem - Try identifying only one of the Major entities by way of gaps
2. Expand the input and the problem and change encoding – Identify gaps present in between all the major entities in the document
3. Differentiate the problem and change the encoding back to 98 distinct features
4. 2nd pass on output of the 3rd stage to augment the result of the previous step.
8th Slide
Final solution and final data model
GATE + NN Multinomial(in two passes)
9th Slide
Insights in to the results
1st Pass identifies smaller sized entities more accurately
2nd Pass identifies larger sized entities more accurately
1st Pass introduces lots of explicit ouliers, which are very much not related to general location and size of entities
2nd Pass muddles with the smaller sized entity’s recogniton
10th Slide
Tools and technologies used

Speaker bio

Krishna Bhavsar has spent around 10 years working on natural language processing, social media analytics, and text mining in various industry domains such as hospitality, banking, healthcare, and more. He has worked on analysing social media responses for popular television shows and popular retail brands and products. His first commercial Tech publication was released worldwide in November 2017, titled, Natural Language Processing with Python - Cookbook with Packt publishing house, UK. He has also published a paper on sentiment analysis augmentation techniques in 2010 NAACL. He recently created an NLP pipeline/toolset and open sourced it for public use. Krishna completed his Post Graduate Diploma in Data Sciences and Business Administration from Great Lakes Institute, Chennai in December 2017.

Apart from academics and technology, Krishna has a passion for motorcycles and football. In his free time, he likes to travel and explore. He has gone on pan-India road trips on his motorcycle and backpacking trips across most of the countries in South East Asia and Europe.
LinkedIn profile - https://www.linkedin.com/in/krishna-bhavsar1

Slides

https://www.slideshare.net/secret/uWAzwRSh2caOfX

Preview video

https://youtu.be/zuufyfjh_sM

Comments

  • Vinayak Joglekar (@vinayakj) a year ago

    Would love to hear Krishnakumar and his experience

  • Kodiyarasan Elangovan (@sugukodi) a year ago

    cool stuffs learnt

  • Pradeep Parasuraman (@pparasuraman) a year ago

    Would like to know more on these topics.

  • Krishna Murthy (@kvsrkmds210) a year ago

    Would like to Know more about these topics

  • saurabh agarwal (@saurabh-agl) a year ago

    Hi Krishna,
    The slides are a little unclear on solution and approach. Can you provide a bit more detail on the path you took to arrive at the solution.
    Thanks

    • Krishna Bhavsar (@krishnabhavsar) Proposer a year ago

      Hi Saurabh,

      The slides were a first cut version. I have updated the slideshare link. You can go though them, but really you wouldn’t get much idea of my appraoch from the slides. That’s something I would illustrate using the infographics in the slides. You are welcome to get in touch with me on email or LinkedIn and discuss this further.

      Regards,
      Krishna

Login with Twitter or Google to leave a comment