_                   _             
 __ __| |_  _ _  __ _  __ | |_  ___  _ _ 
 \ \ /|  _|| '_|/ _` |/ _||  _|/ _ \| '_|
 /_\_\ \__||_|  \__,_|\__| \__|\___/|_|  
                                         
xtractor
Topic extractor with the idea of generating labels using genism.n_similarity
by Peter Nagy

Overview

xtractor is little package which aims to label text automatically harnessing the power of pre-trained word vectors.
The idea is the following:

• You must provide one or more genism compatible pre-trained word vectors
• You must define categories with keywords
• You must provide a tokenized text features you want to label
• Run the extractor to label input text
• The extractor digests the cosine distance of each word (vector) in the sentence and each keyword (vector)
• Then it chooses the most "similar" category as label

Installation

$ pip install xtractor

Usage

See example.py for a more detailed example.

from xtractor import TopicExtractor as te
extractor = te.TopicExtractor(models=models, categories=categories)
labels = extractor.extract(pandas_data_frame)

Parameters

TopicExtractor(models=models, categories=categories)

models

• list of genism compatible models

categories

• list of categories Format:

extract(X=pandas_dataframe)

• input pandas data frame or python list
• in case X is a pandas dataframe, it must have only one column (the feature column)
• X can be a regular python list
• the features are expected to be tokenized string (e.g. following format: ['Tokenized', 'string'])
• the return value is a regular list containing the category names (labels) for each input row respectively (e.g. in case of a 2 row input ['economy', 'sport'])

Precision

It really depends on the quality of you pre-trained word vector and on the quality of your intuitively defined category keywords. In my use case I have used these vectors and played with several iterations of keywords.
I have reached around 69% precision which is not bad. With more carefully picked keywords it can be enhanced.

F.A.Q.

• Q: Why did you make this? A: Because I looked for a way to automatically label huge amount of (hungarian) text and I found no simple way.

Author

• peter nagy | nagypeterjob@gmail.com