End-to-End Memory Networks for Goal-Oriented Conversational Agents

A Keras implementation with TensorFlow back end, tested on the Dialog bAbI dataset

---

This repository contains my unofficial Keras implementation of an End-to-End Memory Network [1] applied to Goal-Oriented dialog systems as described in Bordes et al [2]. The model can be trained on either of the five Dialog bAbI dataset [3]: set in the context of restaurant reservation, its tasks required manipulating sentences and symbols, so as to properly conduct conversations, issue API calls and use the information provided by the outputs of such calls.

Requirements

Mandatory:

• Python 3.6.10
• packages:
  • setuptools==45.1.*
  • tensorflow==2.1.*

If you want to draw the neural network schema:

• graphviz installed system-wide
• packages:
  • pydot==1.4.*
  • pydot-ng==2.0.*

Mathematical models

Figure 1: Model of the End-to-End Memory Network (single-hop).

Figure 2: Three-hops network, high level representation.

---

References

[1] Sainbayar Sukhbaatar, Arthur Szlam, Jason Weston, Rob Fergus, "End-To-End Memory Networks"
http://arxiv.org/abs/1503.08895

[2] Antoine Bordes, Y-Lan Boureau, Jason Weston, "Learning End-to-End Goal-Oriented Dialog"
https://arxiv.org/abs/1605.07683

[3] The bAbI project by Facebook AI research
https://research.fb.com/downloads/babi/

Credits

I used some of the voicy-ai's code inside its public repository in order to implement the data utilities.