Tools for loading and analyzing large text corpora.
Project description
CorpusSearch
A tool to load and search in text corpora.
The tool provides routines to search in large corpora in pandas dataframe format, where rows contain textual information on the level of sentences or paragraphs. Dataframes can be single or multilevel indexed and loaded from url, doi or local files. Accepted formats are pickle, excel, json and csv.
This package is designed to work with Jupyter Notebooks as well as in the IPython command line. If used in a Notebook, the search can be done through a GUI.
Installation
The package can be installed via pip:
pip install corpussearch
Since the package is under active development, the most recent version is always on Github, and can be installed by
pip install git+https://github.com/TOPOI-DH/corpussearch.git
Basic usage
Import the package
from corpussearch.base import CorpusTextSearch
The class is instantiated by providing the path to the source file. Excepted formats are pickled dataframes, CSV, JSON or Excel files.
Standard parameters assume pickled, multiindexed dataframes, where the main text is contained in a column ‘text’. For other sources change parameters accordingly.
search = CorpusTextSearch('./path/to/dataframe/file.pickle')
A reduction to a specific part and page number is obtained by chaining the desired reductions .reduce(key,value), where key can be either a level of the multiindex, or a column name. To obtain the resulting dataframe, .results() is added.
result = search.reduce('part','part_name').reduce('page','page_number').results()
GUI usage
Attention: Work in progress
Import the gui part of the package into a Jupyter Notebook
from corpussearch.gui import CorpusGUI
Instantiate with path to source file, as above.
gui = CorpusGUI('./path/to/dataframe/file.pickle')
and display the interface
gui.displayGUI()
A basic word search returns all results where the searchword is contained in the main column, e.g. ‘text’.
Results are displayed in the sentence output boxes, where the right box contains metainformation derived from the non-main columns or multiindex levels.
To navigate between results use the ‘previous’ and ‘next’ buttons.
Additional search logic
To chain search terms, use the ‘more’-button. This opens additional search fields. Possible logic operations are ‘AND’, ‘OR’, and ‘NOT’. Each logic operation is between two consecutive search pairs (part,value). The logic operates in a linear fashion, from the first triple downwards, e.g. for the search ((‘text’,’NAME’) & (‘part’,’PART1’) | (‘page’,’PAGE4’)) each tuple (key,value) yields a boolean vector v, such that the search becomes (v1 & v2 | v3). Evaluation continues for the pair vtemp = (v1 & v2), and finally vres= (vtemp | v3). The resulting boolean vector is used to reduce the full data to the dataframe containing the search result.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for corpussearch-0.0.7-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | b371b4ffd951c4e7716c2e37129a30e6aaaa6f6bf74b1928da05eee7560ea84c |
|
MD5 | 7d4de8715f9d2f1c44608bd4149b45c1 |
|
BLAKE2b-256 | 3777bc96c32d154a0db58e2f7bc573f91c7ee8f8c41201ca3525464cac9c95ea |