andreas-gompos / bbc-topic-classification

Topic classification using GloVe embeddings and a bidirectional LSTM neural network trained on BBC articles

Topic Classifier

Description

In this project a topic classification model was created. More specifically an LSTM network was trained to distinguish between 5 different categories of articles (business, entertainment, politics, sport, tech). The dataset used, for training the network, was the BBC articles dataset, which consists of 2225 documents, from the BBC news website corresponding to stories from 2004-2005. The model is deployed on Kubernetes on GKE and can used at datagusto.com.

Train the Model

#!/bin/bash
pip install -r requirements.txt
python -m nltk.downloader stopwords punkt wordnet
python ./src/train.py --num_epochs={{ num_epochs }}
                      --top_words={{ top_words }}
                      --max_sequence_length={{ max_sequence_length }}
                      --batch_size={{ batch_size }}
                      --polyaxon_env={{ polyaxon_env }}

Parameter	Description	Valid Values	Default
num_epochs	Number of epochs to be used for training	int	40
batch_size	Batch size to be used for training	int	256
top_words	Number of most common words to be used for training (rare words will be dropped)	int	35000
max_sequence_length	Fixed length of each input text. The text will be padded or trimmed down to that length	int	500
polyaxon_env	Indicate if running in Polyaxon	0, 1	0

Any feedback is welcome! :) LinkedIn