https://github.com/jrzaurin/nlp-stuff

A bit of everything about text and nlp [IN PROGRESS]
https://github.com/jrzaurin/nlp-stuff

deep-learning python3 text-classification text-mining text-processing

Last synced: 10 months ago
JSON representation

A bit of everything about text and nlp [IN PROGRESS]

• Host: GitHub
• URL: https://github.com/jrzaurin/nlp-stuff
• Owner: jrzaurin
• Created: 2017-06-05T11:48:42.000Z (about 9 years ago)
• Default Branch: master
• Last Pushed: 2021-11-05T10:19:21.000Z (over 4 years ago)
• Last Synced: 2025-04-14T08:48:15.732Z (about 1 year ago)
• Topics: deep-learning, python3, text-classification, text-mining, text-processing
• Language: Jupyter Notebook
• Homepage:
• Size: 3.02 MB
• Stars: 28
• Watchers: 4
• Forks: 10
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          


  



# NLP stuff

I add here stuff related to NLP.

Within each directory there should be a README file to help guiding you through the code. So far, this is what I have included:

1. `text_classification_DL_battle`

	Amazon Reviews classification (score prediction) using Hierarchical Attention Networks ([Zichao Yang, et al., 2016](https://www.cs.cmu.edu/~./hovy/papers/16HLT-hierarchical-attention-networks.pdf)), BERT models at the [Hugginface's transformer library](https://github.com/huggingface/transformers) and the [Fastai Text API](https://fastai1.fast.ai/text.html).

2. `text_classification_HAN`

	Amazon Reviews classification (score prediction) using Hierarchical Attention Networks ([Zichao Yang, et al., 2016](https://www.cs.cmu.edu/~./hovy/papers/16HLT-hierarchical-attention-networks.pdf)). I have also used a number of Dropout mechanisms from the work *Regularizing and Optimizing LSTM Language Models* ([Stephen Merity, Nitish Shirish Keskar and Richard Socher, 2017](https://arxiv.org/pdf/1708.02182.pdf)). The companion Medium post can be found [here](https://towardsdatascience.com/predicting-amazon-reviews-scores-using-hierarchical-attention-networks-with-pytorch-and-apache-5214edb3df20).

3. `text_classification_without_DL`

	Predicting the review score for Amazon reviews (Shoes, Clothes and jewelery).

	using tf-idf, LDA and [EnsembleTopics](https://github.com/lmcinnes/enstop)

	along with `lightGBM` and `hyperopt` for the final classification and

	hyper-parameter optimization. I placed special emphasis in the text

	preprocessing.

4. `text_classification_EDA`

	Amazon Reviews classification using tf-idf and *EDA: Easy Data Augmentation

	Techniques for Boosting Performance on Text Classification Tasks* ([Jason Wei

	and Kai Zou 2019](https://github.com/jasonwei20/eda_nlp)) along with

	`lightGBM` and `hyperopt` for the final classification and hyper-parameter

	optimization. Following the philosophy of the previous exercise, I placed

	some emphasis in the text preprocessing, in particular in the use of certain

	tokenizers.

5. `rnn_character_tagging`

	Tagging at character level using RNNs with the aim of differentiating for example, different coding languages or writing styles. The code here is based in a [post](http://nadbordrozd.github.io/blog/2017/06/03/python-or-scala/) by [Nadbor](https://www.linkedin.com/in/nadbor-drozd-12316063/).

6. `textrank`

	The simplest text summarization approach using the `Pagerank` algorithm via

	the	[networkx](https://networkx.github.io/documentation/networkx-1.10/index.html)

	package and comparing the results with the

	proper`Textrank` implementation *Variations of the Similarity Function of TextRank for Automated Summarization* ([Federico Barrios et al., 2016](https://github.com/summanlp/textrank)).

7. `text_classification_CNN_with_tf`

	This is a dir with very old Tensorflow code using the 20_newsgroup dataset.

	My aim back then was is simply to illustrate 3 different ways of building

	a Convolutional neural network for text classification using Tensorflow.

	Last time I checked (October 2019) The code still run, but if you run it

	you will get every possible warning to upgrade. This dir is mostly for me

	to keep track of the things I do more than any other thing.

Any comments or suggestions please: jrzaurin@gmail.com or even better open an issue.