A Contextual Relationship Model for Deceptive Opinion Spam Detection.

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• E-commerce Security

Background:

• E-commerce growth has led to increased online business and competition.
• Deceptive opinion spam misleads consumers and negatively impacts purchasing decisions.
• Existing detection methods often rely on traditional machine learning and lack semantic understanding.

Purpose of the Study:

• To develop an advanced model for detecting deceptive opinion spam in e-commerce.
• To address the limitations of current methods by incorporating semantic and contextual information.
• To improve the accuracy and reliability of identifying misleading online reviews.

Main Methods:

• Proposed a novel model combining capsule neural networks, bidirectional long short-term memory, attention mechanisms, and paragraph vector distributed bag of words.
• Focused on extracting representations that preserve the context and inter-relationships between opinions.
• Leveraged contextual relationships between deceptive opinions for enhanced detection.

Main Results:

• The proposed model achieved superior performance in detecting deceptive opinion spam.
• Demonstrated significant improvements over existing state-of-the-art models.
• Effectively captured semantic coherence and contextual nuances of deceptive reviews.

Conclusions:

• The developed model offers a powerful approach to identifying deceptive opinion spam.
• Highlighting the importance of contextual relationships in spam detection.
• Provides a more robust solution for maintaining trust and integrity in e-commerce platforms.