sentida

Sentida

The new state-of-the-art Danish sentiment analysis tool further developed from the previous state-of-the-art Sentida and shows significant improvement in classifying sentiment in text compared to Sentida (p < 0.01) in three different validation datasets (TP, TP2, Emma).

Implementation of the previous state-of-the-art Danish SA in R, Sentida, for python along with Sentida and programmed loosely from the VADER sentiment analysis python implementation.

• Authors and Citation
• Installation
• Documentation
• Context
• Notes on the current state of the program
• References

Authors and Citation

Created by Søren Orm and Esben Kran.

Emma: Danish Computational Analysis of Emotion in Text (by S. Orm and E. Kran)

For questions and commercial use, please contact:

• Esben Kran C.
  • [e-mail]
  • Aarhus University, CINeMa
• Søren Orm H.
  • [e-mail]
  • Aarhus University, CINeMa

Installation

You can install Sentida through pip with the following command:

pip install sentida

Documentation and examples

The function:

from sentida import Sentida
Sentida().sentida(
                    text,
                    output = ["mean", "total", "by_sentence_mean", "by_sentence_total"],
                    normal = True,
                    speed = ["normal", "fast"]
                    )
# Speed parameter does not have an effect in version <0.2.1

WARNING: Setting speed to fast drastically reduces sentiment precision in complex sentences but speeds up the process by 180% (10,000 iteration test).

Usage examples:

# Define the class:
SV = Sentida()
_____________________________

SV.sentida(
        text = 'Lad der blive fred.',
        output = 'mean',
        normal = False)

Example of usage:
Lad der bliver fred
Sentiment =  2.0
_____________________________

SV.sentida(
        text = 'Lad der blive fred!',
        output = 'mean',
        normal = False)

With exclamation mark:
Lad der blive fred!
Sentiment =  3.13713
_____________________________

SV.sentida(
        text = 'Lad der blive fred!!!',
        output = 'mean',
        normal = False)

With several exclamation mark:
Lad der blive fred!!!
Sentiment =  3.7896530399999997
_____________________________

SV.sentida(
        text = 'Lad der BLIVE FRED',
        output = 'mean',
        normal = False)

Uppercase:
lad der BLIVE FRED
Sentiment =  3.466
_____________________________

SV.sentida(
        text = 'Det går dårligt.',
        output = 'mean',
        normal = False)

Negative sentence:
Det går dårligt
Sentiment =  -1.8333333333333335
_____________________________

SV.sentida(
        text = 'Det går ikke dårligt.',
        output = 'mean',
        normal = False)

Negation in sentence:
Det går ikke dårligt
Sentiment =  1.8333333333333335
_____________________________

SV.sentida(
        text = 'Lad der blive fred, men det går dårligt.',
        output = 'mean',
        normal = False)

'Men' ('but'):
Lad der blive fred, men det går dårligt
Sentiment =  -1.5
_____________________________

SV.sentida(
        text = 'Lad der blive fred.',
        output = 'mean',
        normal = True)

Normalized:
Lad der blive fred
Sentiment =  0.4
_____________________________

SV.sentida(
        text = 'Lad der bliver fred. Det går dårligt!',
        output = 'by_sentence_mean',
        normal = False)

Multiple sentences mean:
Lad der bliver fred. Det går dårligt!
Sentiments = [2.0, -2.8757025]
_____________________________

SV.sentida(
        text = 'Lad der bliver fred. Det går dårligt!',
        output = 'by_sentence_total',
        normal = False)

Multiple sentences total:
Lad der bliver fred. Det går dårligt!
Sentiments = [2.0, -5.751405]
_____________________________

Acknowledgements and Context

Thanks to CINeMa (https://inema.webflow.io), the Sentida team, jry, VADER, AFINN, and last but not least Formula T., for inspiration and encouragement. For license information, see LICENSE.TXT

The Sentida sentiment analysis tool is freely available for research purposes (please cite Lauridsen et al., 2019). If you want to use the tool for commercial purposes, please contact: - [e-mail] - [e-mail] Or the Sentida team: - [e-mail] - [e-mail] - [e-mail]

SENTIDA Aarhus University, Cognitive Science. 2019 - Cognition & Communication. @authors: sorenorm & esbenkc.

This script was developed along with other tools in an attempt to improve danish sentiment analysis. The tool will be updated as more data is collected and new methods for more optimally accessing sentiment is developed.

Notes

VADER BASIS VALUES

Multiplication values: 0.291, 0.215, and 0.208 for !, !!, and !!! respectively empirically tested by one sentence compared to the three conditions 0.733 for uppercase empirically tested from single control sentence to uppercase version 0.293 for degree modifications from adverbs empirically tested with "extremely"

SENTIDA BASIS VALUES

Currently using VADER basis values Question mark is: XXX Degree modifications for other words are implemented in intensitifer list - Need implementation of larger intensifier list based on sentences

FUTURE IMPROVEMENTS

Still missing: common phrases, adjusted values for exclamation marks, Adjusted values for men-sentences, adjusted values for uppercase, More rated words, more intensifiers/mitigators, better solution than snowball stemmer, Synonym/antonym dictionary. Social media orientated: emoticons, using multiple letters - i.e. suuuuuper. Normalization with respect to sub-(-1) and super-(1) output values

References

Lauridsen, G. A., Dalsgaard, J. A., & Svendsen, L. K. B. (2019). SENTIDA: A New Tool for Sentiment Analysis in Danish. Journal of Language Works - Sprogvidenskabeligt Studentertidsskrift, 4(1), 38–53.

Hutto, C. J., & Gilbert, E. (2014, May 16). VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text. Eighth International AAAI Conference on Weblogs and Social Media. Eighth International AAAI Conference on Weblogs and Social Media. https://www.aaai.org/ocs/index.php/ICWSM/ICWSM14/paper/view/8109