Anthill Inside 2017

On theory and concepts in Machine Learning, Deep Learning and Artificial Intelligence. Formerly Deep Learning Conf.

Deep learning for feature extraction from incident data

Submitted by Arthi Venkataraman (@arthi) on Wednesday, 12 April 2017

videocam
Preview video

Technical level

Intermediate

Section

Crisp talk

Status

Submitted

Vote on this proposal

Login to vote

Total votes:  +6

Abstract

Lots of incident data is available in large corporate. However it is Noisy and inaccurate. Classification directly using TFIDF vectorization and machine learning models gives low accuracy. Lots of effort is spent in hand curation of data. Objective is to automatically extract features using deep learning techniques to get a higher lever representation of the text in the incidents. Downstream tasks like classification and clustering will be performed on this space. Presentation will cover the deep learning network architecture of such a system and the results obtained.

Outline

 Problem statement

 Objective

 Tasks  Overview of Relevant Reference Literature  What are we trying to do?  Prior work  View of data  Challenge  Proposed approach  Network architecture  Results  Future work  Conclusions

• What is Representation Learning? 1 min • Representation Learning in NLP and it’s practical applications? 2 mins • Introduction to specific problem we are trying to solve – 1 min • Deep Learning Architecture - 4 mins o Our NLP pre-processing pipeline
o Deep Learning network architecture we have used and Parameters trained with
• Our Results , Conclusions and Further work – Rest of time

• OUTCOMES/CONCLUSION The participants will be exposed to the concept of Representation learning. They will understand the application of Representation learning to the feature extraction space. They will understand how to apply deep learning techniques for natural language processing. Specifically they will be exposed to the deep learning network architecture used, applicability of deep learning in the field of natural language processing, the algorithm used, how the weights are learnt, lessons learnt and results obtained. The unique contribution of this talk is the methodology of application of deep learning for automatically extracting features from the textual incident management domain data.

Requirements

None

Speaker bio

Arthi Venkataraman has 20+ years of experience in the design, development and testing of projects in different domains. She is currently a Senior Member in the Distinguished Members of Technical Staff cadre at Wipro Technologies. Her current role involves development of solutions which involve application of deep learning techniques to natural language processing with intent to better perform in different tasks like classification, clustering, question answering, summarization, etc.
Previously she has been involved in Bot development for different business problems spanning the area of Natural Language Processing, Machine Learning and Semantics Technologies She has a B.E Degree in Computer Science from University Visvesvariah College of Engineering, Bangalore and an MBA (PGDSM) from IIM, Bangalore.
She has previously presented papers and spoken at other international conferences with maximum audience sizes of many hundreds. This presentation is based on Arthi’s experience in feature extraction from natural language using deep learning techniques.

Links

Preview video

https://youtu.be/j3T0X4i4bXg

Comments

  • 1
    Zainab Bawa (@zainabbawa) Reviewer a year ago

    Thanks for sharing the draft slides and the preview video. The proposal currently surveys existing literature. Do you intend to talk about / show applications?

    • 1
      Arthi Venkataraman (@arthi) Proposer a year ago

      Hi Zainab,
      Thanks for comment. The subject of the talk is the Application of Deep learning feature extraction for the Incident data. The talk will focus on the network architecture , the parameters we used and the results we got
      on applying the Deep learning feature extraction techniques for textual data specially the incident / problem tickets we have got.
      The literature survey is only to set the background /context of speech. It is not the main focus of the speech and can be removed if required.

      Rgds.
      Arthi

  • 1
    Zainab Bawa (@zainabbawa) Reviewer a year ago

    Thanks for the clarification, Arthi. Please submit your draft slide deck asap, no later than 11 June, detailing the applications you intend to cover, and takeaways for the audience.

    • 1
      Arthi Venkataraman (@arthi) Proposer a year ago

      Hi Zainab,
      Have shared the current draft of the slides. There could be few more edits before actual presentation.

      Thanks,
      Arthi

Login with Twitter or Google to leave a comment