mat-analyzer

Text analyzer for data for Media Analytics Tool project.

Development

Requirements

• Python 3
• gcc and g++ libraries

Installation & test run

git clone git@github.com:lvtffuk/mat-analyzer.git
cd mat-analyzer
pip install -r requirements.txt
python ./

Creating new analyzer

New analyzer is created when the module is defined in src/analyzers. The class must extend BaseAnalyzer and at least define get_name and _analyze methods.

# test.py
from src.analyzers.base import BaseAnalyzer

class Test(BaseAnalyzer):

	def get_name(self) -> str:
		return "Test"

	def _analyze(self) -> None:
		# Do some stuff

Analyzers are loaded as modules by the name defined in the env vars.

Settings

The settings are set with environment variables.

Variable	Description	Required	Default value
ANALYZER	The analyzer to use. The app can be run with different analyzers on same input data.	✔️
INPUT_FILE	The filepath of the csv file with input file texts.	✔️
OUTPUT_DIR	The directory where the output is stored.	✔️
DATA_KEY	The column in the input csv file where are the texts.	✔️
DOC_ID_KEY	The column in the input csv file where the document ID is.	✔️
CSV_SEPARATOR	The separator of the input csv files.	❌	;
CONFIG_FILE	The path to the config yml file of the analyzer.	❌	None
LANGUAGE	The language for the udpipe analysis.	❌	cs
CLEAR	Indicates if the output dir should be cleared before the run. All downloads are starting again.	❌	0
STOP_WORDS	The path to non-header csv file including stop words.	❌	None

Input file

The input csv file must contain at least two columns. One with document ids and one with the texts to analyze.

Example

Example

"doc_id";"text";"additional_field"
"1";"Some text";"foo"
"2";"Some other text";"bar"

First line must be a header.

Analyzers

BTM

Creates Biterm Topic Model from lemmatized texts. The model is stored as btm-model.pkl in the output directory.

Configuration

iterations: 20
seed: 12321
T: 8
M: 20
alpha: 6.25
beta: 0.01

LDA

Creates LDA model from lemmatized texts. The model is stored as lda.model, lda.model.expElogbeta.npy, lda.model.id2word and lda.model.state in the output directory. For another analyzes the corpus is stored in the output directory as lda-corpus.json.

Configuration

num_topics: 10

LSI

Creates LSI model from lemmatized texts. The model is stored as lsi.model and lsi.model.projection in the output directory.

VT

Creates XML file for Voyant tools. The file is stored as voyant-tools.xml in the output directory.

Configuration

author_key: 'author'
published_key: 'published'

author_key

The column in the input file representing the author. The field is required.

published_key

The column in the input file representing the publication time of the text. The field is required.

Import

The file should be uploaded in the Voyant tools but before that the xpath of the fields must be specified:

Field	Xpath
Documents	//items
Content	//item/content
Author	//item/author
Publication Date	//item/published

For more information check the docs.

Word2Vec

Creates Word2Vec model from lemmatized texts. The model is stored as word2vec.model file in the output directory.

Configuration

vector_size: 100
window: 5
min_count: 1
workers: 4

Output

The output directory contains models mentioned above and additional files.

File	Description
corpus	The directory containing NER corpus (It will be deleted in the future).
udpipe-data.csv	All of the lemmatized tokens data.
udpipe.csv	The original texts with lemmatized words.
udpipe.md5	MD5 checksum of the input file. If the checksum matches the lemmatization is not done again.

Docker

The image is stored in GitHub packages registry and the app can be run in the docker environment.

docker pull ghcr.io/lvtffuk/mat-analyzer:latest

docker run \
--name=mat-analyzer \
-e 'ANALYZER=BTM|LSI|Word2Vec|VT' \
-e 'INPUT_FILE=./input/tweets.csv' \
-e 'OUTPUT_DIR=./output' \
-e 'DATA_KEY=tweet_id' \
-e 'DOC_ID_KEY=tweet' \
-v '/absolute/path/to/output/dir:/usr/src/app/output' \
-v '/absolute/path/to/input/dir:/usr/src/app/input' \
ghcr.io/lvtffuk/mat-analyzer:latest  

The volumes must be set for accessing input and output data.

This work was supported by the European Regional Development Fund-Project “Creativity and Adaptability as Conditions of the Success of Europe in an Interrelated World” (No. CZ.02.1.01/0.0/0.0/16_019/0000734)."