Statistical Methods Using Sentiment Lexicons

2023.12.19

Machine Learning Ontology Digital Transformation Artificial Intelligence Probabilistic Generative Model Clojure Python Natural Language Processing Image Processing Speech Recognition  Navigation of this blog

Statistical Methods Using Sentiment Lexicons

Sentiment Lexicons (Sentiment Polarity Lexicons) will be used to indicate how positive or negative a word or phrase is. The following methods can be used to perform sentiment analysis using this statistical method.

1. simple count-based methods:

  • If a word or phrase is included in the sentiment dictionary, identify whether it is positive or negative.
  • Count the number of positive and negative words in the text and estimate the overall sentiment polarity from the results.

2. weighting method:

  • For each word or phrase, use the polarity score in the sentiment dictionary. Positive words have a positive score and negative words have a negative score.
  • The scores for each word in the text are added together to compute the overall emotional polarity.

3. combined TF-IDF method:

  • To account for word importance, TF-IDF (Term Frequency-Inverse Document Frequency) is used. This is a weighting method that takes into account the frequency of occurrence of a word and the frequency of occurrence of the word in the document as a whole.
  • A TF-IDF score is assigned to each word, which is then combined with the polarity score from the sentiment dictionary to estimate the sentiment polarity of the entire document.

4. machine learning approach:

These methods range from simple and intuitive to complex machine learning methods. Which method is best depends on the specific task and data. Methods using sentiment dictionaries are suitable for simple sentiment analysis tasks because of their low computational cost. However, when context and nuances of language need to be taken into account, machine learning models can provide better results.

These methods are described below.

Simple count-based approach in statistical methods using Sentiment Lexicons

<Overview>

A simple count-based approach will be the basic method for estimating the sentiment polarity of text using Sentiment Lexicons. The following is a step-by-step outline of this approach.

1. obtaining Sentiment Lexicons:

Sentiment Lexicons are dictionaries that indicate whether a word or phrase is positive or negative. Sentiment Lexicons in general contain polarity scores for each word or phrase, and this information is captured and put into a usable format.

2. text preprocessing:

Preprocess the text to be analyzed. This includes lowercasing the text, removing punctuation and numbers, removing stop words (common, meaningless words), etc. Preprocessing is done to remove extraneous information while keeping the text consistent.

3. sentiment word counting:

Count how many words and phrases in Sentiment Lexicons appear in the text, counting positive and negative words, respectively. This allows us to determine which way the overall emotional polarity of the text leans.

4. calculation of the emotion score:

Sum the Sentiment Lexicons polarity scores corresponding to the counted positive and negative words. The score for each word or phrase is determined based on the lexicon, and by subtracting the sum of the negative word scores from the sum of the positive word scores, an overall sentiment score for the document can be obtained.

5. emotion polarity determination:

If the calculated sentiment score is positive, the document is determined to have a positive sentiment. Conversely, if it is negative, the document is judged to have a negative sentiment.

<Example Implementation>

To implement a simple count-based approach, the following steps are performed using Python. Here, we use the NLTK (Natural Language Toolkit) library.

First, if NLTK is not installed, install it as follows

pip install nltk

Next, the following will be an example of implementing a simple count-based approach using NLTK.

import nltk
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords
from collections import Counter

# Download NLTK's Stopword
nltk.download('stopwords')

# Use NLTK's positive and negative word lists
positive_lexicon = set(nltk.corpus.opinion_lexicon.positive())
negative_lexicon = set(nltk.corpus.opinion_lexicon.negative())

def simple_count_based_sentiment(text):
    # Text preprocessing
    text = text.lower()
    words = word_tokenize(text)
    words = [word for word in words if word.isalnum()]  # Select words containing only letters and numbers
    words = [word for word in words if word not in stopwords.words('english')]  # Delete stop word

    # Positive and negative word counts
    positive_count = sum(1 for word in words if word in positive_lexicon)
    negative_count = sum(1 for word in words if word in negative_lexicon)

    # Calculation of Emotion Score
    sentiment_score = positive_count - negative_count

    # Determination of Emotional Polarity
    if sentiment_score > 0:
        sentiment = 'Positive'
    elif sentiment_score < 0:
        sentiment = 'Negative'
    else:
        sentiment = 'Neutral'

    return sentiment, sentiment_score

# Text Example
sample_text = "I love this product. It's amazing!"

# Performing an Emotional Analysis
result, score = simple_count_based_sentiment(sample_text)

# Display Results
print(f"Sentiment: {result}")
print(f"Sentiment Score: {score}")

In this example, the positive and negative word lists are obtained using NLTK’s opinion_lexicon, and the number of occurrences of these words in a given text is counted. Finally, an emotion score is calculated by subtracting the number of negative word occurrences from the number of positive word occurrences, and the emotion polarity is determined based on this score.

Approach to weighting with statistical methods using Sentiment Lexicons

The weighting approach in the statistical method using Sentiment Lexicons shall calculate a sentiment score for each word or phrase by considering the polarity score within Sentiment Lexicons. The following are the basic steps of the weighting approach

1. Acquisition of Sentiment Lexicons:

  • Obtain Sentiment Lexicons and extract the polarity score for each word or phrase.

2. Text preprocessing:

  • Preprocess the text, including lowercasing, removing punctuation, removing stop words, etc.

3. Polarity Score Calculation:

  • Refer to the polarity score in Sentiment Lexicons for each word or phrase in the text.
  • The polarity scores for each word or phrase are summed to compute the sentiment score for the entire document.

4. Determination of Emotion Polarity:

  • Based on the calculated sentiment score, determine the sentiment polarity of the document.

The following is a simple example of implementing this approach using Python.

from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords

# Dummy Sentiment Lexicons (dictionaries containing words and their polarity scores)
sentiment_lexicons = {
    'good': 0.5,
    'bad': -0.5,
    'excellent': 0.8,
    'poor': -0.7
}

def weighted_sentiment_analysis(text, lexicons):
    # Text preprocessing
    text = text.lower()
    words = word_tokenize(text)
    words = [word for word in words if word.isalnum()]  # Select words containing only letters and numbers
    words = [word for word in words if word not in stopwords.words('english')]  # Delete stop word
    # Calculation of Emotion Score
    sentiment_score = sum(lexicons.get(word, 0) for word in words)

    # Determination of Emotional Polarity
    if sentiment_score > 0:
        sentiment = 'Positive'
    elif sentiment_score < 0:
        sentiment = 'Negative'
    else:
        sentiment = 'Neutral'

    return sentiment, sentiment_score

# Text Example
sample_text = "The movie was excellent, but the service was poor."

# Performing an Emotional Analysis
result, score = weighted_sentiment_analysis(sample_text, sentiment_lexicons)

# Display Results
print(f"Sentiment: {result}")
print(f"Sentiment Score: {score}")

In this example, dummy sentiment Lexicons are used as sentiment_lexicons.

Approach of combining TF-IDF with statistical methods using Sentiment Lexicons

An approach that combines TF-IDF with a statistical method using Sentiment Lexicons performs sentiment analysis by considering the importance of words.” TF-IDF (Term Frequency-Inverse Document Frequency), described in “Overview of tfidf and its Implementation in Clojure” is a method that combines word occurrence frequency and inverse document frequency to calculate word weights. The basic steps of this approach are described below.

1. Obtaining Sentiment Lexicons:

  • Obtain Sentiment Lexicons and extract a polarity score for each word or phrase.
    Text preprocessing.
  • Preprocess the text, e.g., lowercasing, removing punctuation, removing stop words, etc.

2 TF-IDF computation:

  • Calculate TF (Term Frequency) and IDF (Inverse Document Frequency) for each word in the text.
  • The TF-IDF is a multiplication of the TF and IDF and represents the weight of the word.

3. Computing Polarity Scores:

  • The TF-IDF for each word is combined with the polarity score in Sentiment Lexicons to compute a sentiment score for each word.
  • The sentiment scores for each word are summed to obtain the sentiment score for the entire document.

4. Determination of Emotion Polarity:

  • Determine the emotional polarity of the document based on the calculated emotional score.

The following is a simple example of sentiment analysis combining TF-IDF and Sentiment Lexicons using Python.

from sklearn.feature_extraction.text import TfidfVectorizer
from nltk.tokenize import word_tokenize
from nltk.corpus import stopwords

# Dummy Sentiment Lexicons (dictionaries containing words and their polarity scores)
sentiment_lexicons = {
    'good': 0.5,
    'bad': -0.5,
    'excellent': 0.8,
    'poor': -0.7
}

def tfidf_sentiment_analysis(text, lexicons):
    # Text preprocessing
    text = text.lower()
    words = word_tokenize(text)
    words = [word for word in words if word.isalnum()]  # アルファベットと数字のみを含む単語を選択
    words = [word for word in words if word not in stopwords.words('english')]  # ストップワードの削除

    # Calculation of TF-IDF
    vectorizer = TfidfVectorizer()
    tfidf_matrix = vectorizer.fit_transform([' '.join(words)])

    # Combines TF-IDF and polarity scores for each word to compute an emotion score
    sentiment_score = sum(tfidf_matrix[0, vectorizer.vocabulary_.get(word, 0)] * lexicons.get(word, 0) for word in words)

    # Determination of Emotional Polarity
    if sentiment_score > 0:
        sentiment = 'Positive'
    elif sentiment_score < 0:
        sentiment = 'Negative'
    else:
        sentiment = 'Neutral'

    return sentiment, sentiment_score

# Text Example
sample_text = "The movie was excellent, but the service was poor."

# Performing an Emotional Analysis
result, score = tfidf_sentiment_analysis(sample_text, sentiment_lexicons)

# Display Results
print(f"Sentiment: {result}")
print(f"Sentiment Score: {score}")

In this example, the TfidfVectorizer is used to compute the TF-IDF of the words in the text, and the weights are combined with the polarity score of the Sentiment Lexicons to compute the sentiment score.

Statistical approach to machine learning using Sentiment Lexicons

There are several steps in the machine learning approach with statistical methods using Sentiment Lexicons. The basic steps are as follows

1 Dataset Preparation:

To train a machine learning model, a labeled sentiment analysis dataset is required. Each text must be labeled with its emotional polarity (positive, negative, neutral, etc.).

2 Feature Extraction:

Extract features from each text using Sentiment Lexicons. This is the Sentiment Lexicons polarity score for the words and phrases in the text.

3 Data Vectorization:

To convert the features into numerical data, the text data is vectorized. This can be done using TF-IDF, Word Embeddings (Word2Vec, GloVe, etc.), document embedding (Doc2Vec, etc.), etc.

4. Select a machine learning model:

Select the machine learning algorithm to be used. Common choices include support vector machines, random forests, decision trees, neural networks, etc.

5. Train the model:

Apply the selected model to the training data to train the sentiment analysis model. The training data consists of features extracted from Sentiment Lexicons and the corresponding emotional polarity labels.

6 Model evaluation:

Once training is complete, the model is evaluated using a test data set. This allows us to evaluate the performance and generalization capabilities of the model.

Below is a simple example of a machine learning approach incorporating Sentiment Lexicons using a simple support vector machine (SVM) with Scikit-learn.

from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.svm import SVC
from sklearn.metrics import accuracy_score

# Dummy Sentiment Lexicons (dictionaries containing words and their polarity scores)
sentiment_lexicons = {
    'good': 0.5,
    'bad': -0.5,
    'excellent': 0.8,
    'poor': -0.7
}

# Dummy labeled data set
texts = ["I love this product. It's amazing!",
         "The service was terrible. I won't come back.",
         "The movie was good, but the ending was disappointing."]

labels = ['Positive', 'Negative', 'Neutral']

# Feature Extraction and Vectorization
vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(texts)

# Adding features using Sentiment Lexicons
lexicon_features = [[sentiment_lexicons.get(word, 0) for word in vectorizer.get_feature_names()] for _ in range(X.shape[0])]
X_with_lexicons = scipy.sparse.hstack((X, np.array(lexicon_features)))

# Data set partitioning
X_train, X_test, y_train, y_test = train_test_split(X_with_lexicons, labels, test_size=0.2, random_state=42)

# Create a model of a support vector machine (SVM)
svm_model = SVC(kernel='linear')

# Model Training
svm_model.fit(X_train, y_train)

# Prediction on test data
predictions = svm_model.predict(X_test)

# Accuracy Rating
accuracy = accuracy_score(y_test, predictions)
print(f"Accuracy: {accuracy}")

In this example, the text is vectorized using TF-IDF, polarity scores for each word are added as features using Sentiment Lexicons, and then support vector machine (SVM) models are trained and evaluated for accuracy.

Challenges and Remedies for Statistical Methods Using Sentiment Lexicons

Several challenges exist in statistical methods using Sentiment Lexicons. These challenges and measures to address them are described below.

1. lack of vocabulary and difficulty in updating:

Challenge: When Sentiment Lexicons contain only a limited vocabulary, they cannot respond to new words and expressions as they appear.
Solution: Regular updating of Lexicons and introduction of dynamic learning methods are possible. It is important to expand Lexicons from time to time to accommodate new words and phrases.

2. ignoring word context:

Challenge: Simple count-based methods ignore word context and consider only the number of occurrences of individual words. This may limit the understanding of context-dependent emotions.
Solution: More complex models could be employed, such as methods that take into account the context before and after a word, or n-gram models that take into account word combinations.

3. word polysemy:

Challenge: A word can have different emotional polarity in different contexts and usages, and simple dictionary-based methods have difficulty in dealing with this polysemy.
Solution: To more accurately capture the meaning and context of a word, more sophisticated feature representations could be introduced, such as using Word Embeddings.

4. coping with contextual changes:

Challenge: Even if a word previously had a particular sentiment polarity, that polarity may change as the context changes. Static Sentiment Lexicons cannot deal with changes in context.
Solution: Dynamic Sentiment Lexicons could be used, or models could be built that can quickly adapt to contextual changes. This would require online learning and real-time updating.

Reference Information and Reference Books

For more information on natural language processing in general, see “Natural Language Processing Technology” and “Overview of Natural Language Processing and Examples of Various Implementations.

Reference books include “Natural language processing (NLP): Unleashing the Power of Human Communication through Machine Intelligence“.

“Practical Natural Language Processing: A Comprehensive Guide to Building Real-World NLP Systems“

“Natural Language Processing With Transformers: Building Language Applications With Hugging Face“

コメント

タイトルとURLをコピーしました