Attribute Embedding: Learning Hierarchical Representations of Product Attributes from Consumer Reviews

Area of Science:

• Business Analytics
• Computational Linguistics
• Marketing Science

Background:

• Understanding customer perspectives is crucial for sales, product design, and engineering.
• Existing methods struggle to connect technical product specifications (engineered attributes) with abstract customer benefits (meta-attributes).

Purpose of the Study:

• To develop a methodological framework for extracting a hierarchy of product attributes from consumer reviews.
• To enable flexible sentiment analysis linking engineered attributes to meta-attributes and identify key drivers of customer satisfaction.
• To provide managers with tools for monitoring relevant review content and comparing products across brands.

Main Methods:

• Utilized machine learning and natural language processing (NLP) to analyze consumer reviews.
• Developed a framework to extract a hierarchy of product attributes based on contextual information.
• Applied the framework to the tablet computer category, generating dashboards and perceptual maps.

Main Results:

• The attribute hierarchy successfully links engineered attributes to meta-attributes within a product category.
• Enabled identification of how consumers perceive meta-attributes and which engineered attributes are primary drivers.
• Validated the attribute hierarchy using both primary and secondary data.

Conclusions:

• The framework provides actionable insights for product development and marketing strategies.
• Managers can effectively monitor customer feedback and benchmark products against competitors.
• The methodology facilitates exploration of product evolution and market dynamics, as demonstrated in the tablet computer example.