Weighted Joint Sentiment-Topic Model for Sentiment Analysis Compared to ALGA: Adaptive Lexicon Learning Using Genetic...

Area of Science:

• Natural Language Processing
• Machine Learning
• Unsupervised Learning

Background:

• Latent Dirichlet Allocation (LDA) is an unsupervised learning method for topic modeling.
• Joint Sentiment-Topic (JST) extends LDA to include sentiment analysis but has limitations in its emotion parameter.
• Existing models often overlook the contextual influence of words on each other within documents.

Purpose of the Study:

• To introduce two novel topic models, Weighted Joint Sentiment-Topic (WJST) and Weighted Joint Sentiment-Topic 1 (WJST1), to enhance JST.
• To incorporate new parameters that account for word-to-word influence and sentiment dictionary generation.
• To improve the accuracy and performance of sentiment-topic models in both single-domain and multidomain scenarios.

Main Methods:

• Development of WJST and WJST1 models, extending JST by introducing parameters that capture inter-word effects.
• Implementation considering that each word in a document influences its neighbors, and vice versa.
• Evaluation using thirteen diverse datasets, employing perplexity, document-level opinion mining, and topic coherence metrics.

Main Results:

• The new parameters significantly enhance model accuracy, with proposed methods outperforming SVM and logistic regression.
• WJST and WJST1 demonstrate high accuracy, exceeding 80% on most datasets, with WJST1 achieving 97% on the Movie dataset.
• The models show superior performance compared to the Adaptive Lexicon learning using Genetic Algorithm (ALGA).

Conclusions:

• The proposed WJST and WJST1 models effectively improve sentiment-topic modeling by considering word interactions.
• These models offer robust solutions for both single-domain (WJST1) and multidomain (WJST) text analysis.
• The methods provide accurate, unsupervised approaches to emotion detection and topic analysis, surpassing baseline methods.