Deep Natural Language Classification

This repository contains Python3 programs to solve natural language classification tasks using deep neural networks.

To train a natural language classifier you only need two files: a training dataset and a model configuration file. Dataset files must conform to a custom format and are parsed automatically. Model configurations are simple JSON files specifying the architecture, inputs and targets of a neural network classifier.

Features

• Create custom deep neural networks by stacking different building blocks like RNNs, CNNs, GCNs and more
• Implemented in Tensorflow which automatically enables GPU acceleration
• Create multi-task models to perform several classification tasks with a single neural network
• Numerical representation of natural language using word vectors, part-of-speech tags and syntactic dependencies
• Websocket server that processes natural language sentences using one or more trained neural networks

Requirements

• Python 3
• Tensorflow (see installation instructions, for better performance GPU use is recommended)

Optional:

• Stanford CoreNLP Server to compute part-of-speech tags and syntactic dependencies (see section Setting up a Stanford CoreNLP server)

Documentation

• Quickstart guide
  A simple example of how to train a neural network for natural language classification.

• Dataset format
  Describes a JSON format for storing datasets containing natural language sentences with different types of annotations.

• Data generation
  Fill natural language sentences containing placeholders with short phrases to generate complete sentences.

• Setting up a Stanford CoreNLP Server for use with this project

• Representing natural language sentences numerically
  Programs to add word vector, POS tag and syntactic dependency annotations to a dataset.

• Model configuration
  Describes how to specify neural network models using a JSON model configuration file.

• Training neural networks
  How to train a neural network with a dataset and model configuration file.

• Starting a query server
  Explains how to start a websocket server that processes natural language sentences using a trained neural network.

License

MIT