Your browser was unable to load all of the resources. They may have been blocked by your firewall, proxy or browser configuration.
Press Ctrl+F5 or Ctrl+Shift+R to have your browser try again.

uber / ludwig

Ludwig is a toolbox built on top of TensorFlow that allows to train and test deep learning models without the need to write code.

Reproduce COTA using Ludwing #21

Lucas Rolim lucaslrolim ·
#1

Hi,

I'm watching an amazing talk about how Uber handles Customer Care using COTA and I'm wondering if it's possible to reproduce the architecture of this project just using the Ludwig framework. The architecture I'm referencing is in this Uber presentation: https://youtu.be/jIGHY7fz2XA?t=991.

There is any similar example of COTA using Ludwig?

Thank you guys for this incredible framework!

  • replies 2
  • views 1.4K
  • likes 0
Piero Molino w4nderlust ·
#2

Hey Lucas,
funny enough I'm the person giving that talk :)
And actually COTA was the first model developed with Ludwig, so definitely yes, you can reproduce exactly that architecture in Ludwig.
In your model definition you would need to have the input text of the ticket + all the other features about the user and the context you can extract from your data, then in the output feature you can have contact type, action and template as categories. In cota we also added dependencies between the output features (check the ludwig documentation for more details) and we made the prediction of the contact type actually a prediction of a path from root to leaf in the tree of contact types by using a sequence output feature.
Let me know if this is not clear or if you have further questions.
Cheers,
Piero

P.S. for more details on COTA:

COTA: Improving the Speed and Accuracy of Customer Support through Ranking and Deep Networks

For a company looking to provide delightful user experiences, it is of paramount importance to take care of any customer issues. This paper proposes COTA, a system to improve speed and reliability of customer support for end users through automated tick...

arxiv.org

COTA: Improving Uber Customer Care with NLP & Machine Learning

In this article, Uber Engineering introduces our Customer Obsession Ticket Assistant (COTA), a new tool that puts machine learning and natural language processing models in the service of customer care to help agents deliver improved support experiences.

eng.uber.com

Lucas Rolim lucaslrolim · Author
#3

Thank you, Piero! I checked your website and there is even one more funny coincidence. In my master thesis, I'm implementing in DGL a very similar model that you had implemented on Uber Eats using Graph Convolutional Networks :sweat_smile:

The paper and the blog post are very pretty clear and well structured! I still with just one more question. In my case a have a set of input features (text + user features + order features) and I would like to recommend a FAQ post (text with tags) based on the user question. Any suggestions about how to adjust the architecture to this problem and train in an unsupervised way?

Again, thank you for your help.

Best regards,