Sentiment analysis

The sentiment classes tokenize text and score a list of tokens. A typical workflow is:

from ekonlpy.sentiment import MPKO

analyzer = MPKO(kind=1)
tokens = analyzer.tokenize("금리 인상이 필요하다")
result = analyzer.get_score(tokens)
print(result)

get_score returns Positive, Negative, Polarity, and Subjectivity. Pass return_breakdown=True to include one record per input term. For MPKO and EUKO, by_count=False uses the polarity values in their lexicons:

from ekonlpy.sentiment import MPKO

analyzer = MPKO(kind=1)
tokens = analyzer.tokenize("금리 인상이 필요하다")
print(analyzer.get_score(tokens, return_breakdown=True))
print(analyzer.get_phrase_breakdown(tokens))

HIV4 and LM classify terms by occurrence. Their bundled implementations do not populate _poldict, so by_count=False produces zero weighted scores for those classes; use the default by_count=True mode.

Exported analyzers

Class	Text and lexicon	Notes
MPKO	Korean monetary policy	kind selects the bundled policy lexicon. Supported kinds are 0, 1, 3, and 7. Positive means hawkish and negative means dovish.
EUKO	Korean economic uncertainty	Scores terms from the bundled uncertainty lexicon.
KSA	Korean general sentiment	Bundled KOSAC polarity dictionary with KoNLPy's Kkma tokenizer. Requires the optional Java backed KoNLPy integration.
HIV4	English general sentiment	Harvard IV-4 terms are stemmed by the default tokenizer.
LM	English financial sentiment	Loughran and McDonald terms are stemmed by the default tokenizer.
MPCK	Korean monetary policy classifier	Uses the bundled NLTK Naive Bayes model and MeCab tokenization.

All six are importable from ekonlpy.sentiment:

from ekonlpy.sentiment import EUKO, HIV4, KSA, LM, MPCK, MPKO

Korean and English dictionary analyzers

EUKO scores Korean economic uncertainty terms. It supports kind=0 and kind=1:

from ekonlpy.sentiment import EUKO

analyzer = EUKO(kind=1)
tokens = analyzer.tokenize("금리 인상이 필요하다")
print(tokens)
print(analyzer.get_score(tokens))

HIV4 and LM use an English tokenizer with Porter stemming:

from ekonlpy.sentiment import HIV4, LM

for analyzer in (HIV4(), LM()):
    tokens = analyzer.tokenize("abundance and abandon")
    print(type(analyzer).__name__, tokens, analyzer.get_score(tokens))

MPCK classification

MPCK uses its bundled classifier by default:

from ekonlpy.sentiment import MPCK

classifier = MPCK()
tokens = classifier.tokenize("금리 인상이 필요하다")
features = tokens + classifier.ngramize(tokens)
print(classifier.classify(features, intensity_cutoff=1.3))

The result contains Polarity, Intensity, Pos score, and Neg score. load_classifier and save_classifier read and write Python pickle files. Only load a model file you trust: unpickling can execute arbitrary code. The bundled model is loaded automatically; custom model files are an advanced, trusted-input workflow.

Optional KSA and KOSAC analyzers

KSA uses the bundled KOSAC polarity dictionary with KoNLPy's Java backed Kkma tokenizer. KOSAC is a separate analyzer that also uses Kkma. Install and configure the optional KoNLPy and Java dependencies before constructing either class:

from ekonlpy.sentiment import KSA

analyzer = KSA()
tokens = analyzer.tokenize("좋은 정책")
print(analyzer.get_score(tokens))

This optional example is not executed in the project's default development environment because KoNLPy and Java are not installed there.