Python package for text mining of time-series data
Text data is often recorded as a time series with significant variability over time. Some examples of time-series text data include social media conversations, product reviews, research metadata, central banker communication, and newspaper headlines. Arabica makes exploratory analysis of these datasets simple by providing:
- Descriptive n-gram analysis: n-gram frequencies
- Time-series n-gram analysis: n-gram frequencies over a period
- Text visualization: n-gram heatmap, line plot, word cloud
- Sentiment analysis: VADER sentiment classifier
- Financial sentiment analysis: with FinVADER
- Structural breaks identification: Jenks Optimization Method
It automatically cleans data from punctuation on input. It can also apply all or a selected combination of the following cleaning operations:
- Remove digits from the text
- Remove the standard list(s) of stopwords
- Remove an additional list of stop words
Arabica works with texts of languages based on the Latin alphabet, uses cleantext for punctuation cleaning, and enables stop words removal for languages in the NLTK corpus of stopwords.
It reads dates in:
- US-style: MM/DD/YYYY (2013-12-31, Feb-09-2009, 2013-12-31 11:46:17, etc.)
- European-style: DD/MM/YYYY (2013-31-12, 09-Feb-2009, 2013-31-12 11:46:17, etc.) date and datetime formats.
Arabica requires Python 3.8 - 3.10, NLTK - stop words removal, cleantext - text cleaning, wordcloud - word cloud visualization, plotnine - heatmaps and line graphs, matplotlib - word clouds and graphical operations, vaderSentiment - sentiment analysis, finvader - financial sentiment analysis, and jenskpy for breakpoint identification.
To install using pip, use:
pip install arabica
- Import the library:
from arabica import arabica_freq
from arabica import cappuccino
from arabica import coffee_break - Choose a method:
arabica_freq enables a specific set of cleaning operations (lower casing, numbers, common stop words, and additional stop words removal) and returns a dataframe with aggregated unigrams, bigrams, and trigrams frequencies over a period.
def arabica_freq(text: str, # Text
time: str, # Time
date_format: str, # Date format: 'eur' - European, 'us' - American
time_freq: str, # Aggregation period: 'Y'/'M'/'D', if no aggregation: 'ungroup'
max_words: int, # Maximum of most frequent n-grams displayed for each period
stopwords: [], # Languages for stop words
stopwords_ext: [], # Languages for extended stop words list, currently provided lists: 'english'
skip: [], # Remove additional strings. Cuts the characters out without tokenization, useful for specific or rare characters. Be careful not to bias the dataset.
numbers = True, # Remove numbers
lower_case = True) # Lowercase text
numbers: bool = False, # Remove numbers
lower_case: bool = False # Lowercase text
) cappuccino enables cleaning operations (lower casing, numbers, common stop words, and additional stop words removal) and provides plots for descriptive (word cloud) and time-series (heatmap, line plot) visualization.
def cappuccino(text: str, # Text
time: str, # Time
date_format: str, # Date format: 'eur' - European, 'us' - American
plot: str, # Chart type: 'wordcloud'/'heatmap'/'line'
ngram: int, # N-gram size, 1 = unigram, 2 = bigram, 3 = trigram
time_freq: str, # Aggregation period: 'Y'/'M', if no aggregation: 'ungroup'
max_words int, # Maximum of most frequent n-grams displayed for each period
stopwords: [], # Languages for stop words
stopwords_ext: [], # Languages for extended stop words list, currently provided lists: 'english'
skip: [], # Remove additional strings. Cuts the characters out without tokenization, useful for specific or rare characters. Be careful not to bias the dataset.
numbers: bool = False, # Remove numbers
lower_case: bool = False # Lowercase text
)coffee_break provides sentiment analysis and breakpoint identification in aggregated time series of sentiment. The implemented models are:
-
VADER is a lexicon and rule-based sentiment classifier attuned explicitly to general language expressed in social media
-
FinVADER improves VADER's classification accuracy on financial texts, including two financial lexicons
Break points in the time series are identified with the Fisher-Jenks algorithm (Jenks, 1977. Optimal data classification for choropleth maps).
def coffee_break(text: str, # Text
time: str, # Time
date_format: str, # Date format: 'eur' - European, 'us' - American
model: str, # Sentiment classifier, 'vader' - general language, 'finvader' - financial text
skip: [], # Remove additional strings. Cuts the characters out without tokenization, useful for specific or rare characters. Be careful not to bias the dataset.
preprocess: bool = False, # Clean data from numbers and punctuation
time_freq: str, # Aggregation period: 'Y'/'M'
n_breaks: int # Number of breakpoints: min. 2
)- Read the documentation
For more examples of coding, read these tutorials:
General use:
- Sentiment Analysis and Structural Breaks in Time-Series Text Data here
- Visualization Module in Arabica Speeds Up Text Data Exploration here
- Text as Time Series: Arabica 1.0 Brings New Features for Exploratory Text Data Analysis here
Applications:
- Business Intelligence: Customer Satisfaction Measurement with N-gram and Sentiment Analysis here
- Research meta-data analysis: Research Article Meta-data Description Made Quick and Easy here
- Media coverage text mining
- Social media analysis
💬 Please visit here for any questions, issues, bugs, and suggestions.
Using arabica in a paper or thesis? Please cite this paper:
@article{Koráb:2024,
author = {{Koráb}, P., and {Poměnková}, J.},
title = {Arabica: A Python package for exploratory analysis of text data},
journal = {Journal of Open Source Software},
volume = {97},
number = {9},
pages = {6186},
year = {2024},
doi = {doi.org/10.21105/joss.06186},
}