Improving the Polarity of Text through word2vec Embedding for Primary Classical Arabic Sentiment Analysis

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Sentiment analysis research has grown significantly, but classical Arabic literature remains under-explored.
• Existing Arabic text feature extraction methods often rely on word frequency, neglecting semantic relationships.
• No dedicated dataset or advanced feature extraction for sentiment analysis of classical Arabic books was available.

Purpose of the Study:

• To develop a novel classical Arabic dataset (CASAD) for sentiment analysis from art books.
• To implement advanced word embedding techniques for extracting deep semantic features from formal Arabic text.
• To evaluate the effectiveness of various machine learning models for sentiment classification on this dataset.

Main Methods:

• Creation of the CASAD dataset by collecting and human-expert labeling sentences from classical Arabic art books.
• Application of word embedding techniques (similar to Word2Vec) to extract semantic features.
• Evaluation of features using machine learning algorithms: Support Vector Machines (SVM), Logistic Regression (LR), Naive Bayes (NB), K-Nearest Neighbors (KNN), Latent Dirichlet Allocation (LDA), and Classification And Regression Trees (CART).
• Statistical validation and reliability testing of dataset labels.
• Tenfold cross-validation to assess classification rates for two and three classes.

Main Results:

• The Logistic Regression model combined with Word2Vec embeddings demonstrated superior performance in sentiment classification.
• The developed CASAD dataset and feature extraction methods provide a robust foundation for Arabic sentiment analysis.
• Comparative analysis showed varying classification accuracies across different machine learning algorithms.

Conclusions:

• The Logistic Regression with Word2Vec approach is highly effective for sentiment analysis in classical Arabic.
• The CASAD dataset and proposed feature extraction method advance the field of Arabic Natural Language Processing.
• This research provides a benchmark for future sentiment analysis studies on classical Arabic texts.