Senin, 07 Oktober 2019

Orange3 Menambahkan Sentiment Analysis Bahasa Indonesia

Bagi pengguna aplikasi Text Mining Orange3, tentu saja akan mengalami kesulitan saat akan melakukan penghitungan Sentiment Analysis , dkarenakan Orange3 hanya menyediakan dua bahasa dalam proses Sentimen Analysis yaitu bahasa Inggris dan Slovenia dalam method Liu Hiu .

Anda dapat menambahkan Bahasa Indonesia dalam metode Liu Hiu ini dengan sedikit modifikasi dan penambahan script python pada proses Sentiment Analysisnya yaitu dengan menambahkan file yang berisi kumpulan kata yang memiliki makna sentimen negatif dan sentimen positif dalam bahasa Indonesia.

Untuk menambahkan kata tersebut  adalah sebagai berikut :
  1. Buka folder /usr/local/lib/python3.7/site-packages/orangecontrib/text/sentiment/resources , dalam folder tersebut terdapat dua file yaitu negatif_words_Slolex.txt yang berisi kata negatif dalam bahasa slovenian dan positive_words_Slolex.txt . Selanjutnya copy negatif_words_Slolex.txt menjadi negatif_words_Ina.txt dan selanjutnya file negatif_words_Ina.txt diedit dengan menghapuskan seluruh isi kata dari bahasa slovenia dan menambahkan kata dengan bahasa indonesai yang memiliki nilai negatif , demikain juga dengan file positive_words_Slolex.txt di copy menjadi file positive_words_Ina.txt, sehingga terdapat 4 file pada folder tersebut.
  2. Selanjutnya  buka folder /usr/local/lib/python3.7/site-packages/orangecontrib/text/sentiment , copy  file opinion_lexicon_lso.py menjadi file  opinion_lexicon_ina.py selanjutnya edit sperti script berikut ini:

    import os
    class opinion_lexicon_ina:
        resources_folder = os.path.dirname(__file__)
        @classmethod
        def positive(cls):
            with open(os.path.join(cls.resources_folder,
                                   'resources/positive_words_Ina.txt'),
                      'r') as f:
                return f.read().split('\n')

        @classmethod
        def negative(cls):
            with open(os.path.join(cls.resources_folder,
                                   'resources/negative_words_Ina.txt'),
                      'r') as f:
                return f.read().split('\n')

  3.  Selanjutnya edit file __init__.py :

    import numpy as np
    from nltk.corpus import opinion_lexicon
    from nltk.sentiment import SentimentIntensityAnalyzer

    from orangecontrib.text import Corpus
    from orangecontrib.text.misc import wait_nltk_data
    from orangecontrib.text.preprocess import WordPunctTokenizer
    from orangecontrib.text.vectorization.base import SharedTransform, \
        VectorizationComputeValue
    from orangecontrib.text.sentiment.opinion_lexicon_ina import opinion_lexicon_ina


    class Liu_Hu_Sentiment:
        sentiments = ('sentiment',)
        name = 'Liu Hu'

        methods = {'English': opinion_lexicon,
                   'Indonesia': opinion_lexicon_ina
    }

        @wait_nltk_data
        def __init__(self, language):
            self.language = language
            self.positive = set(self.methods[language].positive())
            self.negative = set(self.methods[language].negative())

        def transform(self, corpus, copy=True):
            scores = []
            tokenizer = WordPunctTokenizer()
            tokens = tokenizer(corpus.documents)

            for doc in tokens:
                pos_words = sum(word in self.positive for word in doc)
                neg_words = sum(word in self.negative for word in doc)
                scores.append([100*(pos_words - neg_words)/max(len(doc), 1)])
            X = np.array(scores).reshape((-1, len(self.sentiments)))

            # set  compute values
            shared_cv = SharedTransform(self)
            cv = [VectorizationComputeValue(shared_cv, col)
                  for col in self.sentiments]

            if copy:
                corpus = corpus.copy()
            corpus.extend_attributes(X, self.sentiments, compute_values=cv)
            return corpus


    class Vader_Sentiment:
        sentiments = ('pos', 'neg', 'neu', 'compound')
        name = 'Vader'

        @wait_nltk_data
        def __init__(self):
            self.vader = SentimentIntensityAnalyzer()

        def transform(self, corpus, copy=True):
            scores = []
            for text in corpus.documents:
                pol_sc = self.vader.polarity_scores(text)
                scores.append([pol_sc[x] for x in self.sentiments])
            X = np.array(scores).reshape((-1, len(self.sentiments)))
            # set  compute values
            shared_cv = SharedTransform(self)
            cv = [VectorizationComputeValue(shared_cv, col)
                  for col in self.sentiments]
            if copy:
                corpus = corpus.copy()
            corpus.extend_attributes(X, self.sentiments, compute_values=cv)
            return corpus

    if __name__ == "__main__":
        corpus = Corpus.from_file('deerwester')
        liu = Liu_Hu_Sentiment('Indonesia')
        corpus2 = liu.transform(corpus[:5])

  4. Agar Widget Sentiment Analysis terdapat pilihan bahasa Indonesia , selanjutnya edit file /usr/local/lib/python3.7/dist-packages/orangecontrib/text/widgets/owsentimentanalysis.py

    from AnyQt.QtCore import Qt
    from AnyQt.QtWidgets import QApplication, QGridLayout, QLabel

    from Orange.widgets import gui, settings
    from Orange.widgets.utils.signals import Input, Output
    from Orange.widgets.widget import OWWidget
    from orangecontrib.text import Corpus
    from orangecontrib.text.sentiment import Vader_Sentiment, Liu_Hu_Sentiment

    class OWSentimentAnalysis(OWWidget):
        name = "Sentiment Analysis"
        description = "Predict sentiment from text."
        icon = "icons/SentimentAnalysis.svg"
        priority = 320

        class Inputs:
            corpus = Input("Corpus", Corpus)

        class Outputs:
            corpus = Output("Corpus", Corpus)

        method_idx = settings.Setting(1)
        autocommit = settings.Setting(True)
        language = settings.Setting('English')
        want_main_area = False
        resizing_enabled = False

        METHODS = [
            Liu_Hu_Sentiment,
            Vader_Sentiment
        ]
        LANG = ['English', 'Indonesia']

        def __init__(self):
            super().__init__()
            self.corpus = None

            form = QGridLayout()
            self.method_box = box = gui.radioButtonsInBox(
                self.controlArea, self, "method_idx", [], box="Method",
                orientation=form, callback=self._method_changed)
            self.liu_hu = gui.appendRadioButton(box, "Liu Hu", addToLayout=False)
            self.liu_lang = gui.comboBox(None, self, 'language',
                                         sendSelectedValue=True,
                                         items=self.LANG,
                                         callback=self._method_changed)
            self.vader = gui.appendRadioButton(box, "Vader", addToLayout=False)

            form.addWidget(self.liu_hu, 0, 0, Qt.AlignLeft)
            form.addWidget(QLabel("Language:"), 0, 1, Qt.AlignRight)
            form.addWidget(self.liu_lang, 0, 2, Qt.AlignRight)
            form.addWidget(self.vader, 1, 0, Qt.AlignLeft)

            ac = gui.auto_commit(self.controlArea, self, 'autocommit', 'Commit',
                                 'Autocommit is on')
            ac.layout().insertSpacing(1, 8)

        @Inputs.corpus
        def set_corpus(self, data=None):
            self.corpus = data
            self.commit()

        def _method_changed(self):
            self.commit()

        def commit(self):
            if self.corpus is not None:
                method = self.METHODS[self.method_idx]
                if self.method_idx == 0:
                    out = method(language=self.language).transform(self.corpus)
                else:
                    out = method().transform(self.corpus)
                self.Outputs.corpus.send(out)
            else:
                self.Outputs.corpus.send(None)

        def send_report(self):
            self.report_items((
                ('Method', self.METHODS[self.method_idx].name),
            ))

    def main():
        app = QApplication([])
        widget = OWSentimentAnalysis()
        corpus = Corpus.from_file('book-excerpts')
        corpus = corpus[:3]
        widget.set_corpus(corpus)
        widget.show()
        app.exec()

    if __name__ == '__main__':
        main()

Selanjutnya anda kini sudah bisa menggunakan Sentiment Analysis dengan menggunakan analisis NLTK dengan menggunakan bahasa indonesia

Tidak ada komentar:

Posting Komentar

Python Flask Server Dan ESP32

  Catatan kecil ini adalah hasil oprekan mikrokontroler ESP32 dengan menggunakan Server Flask Python. Berawal dari penggunaan MQTT Server pa...