Information Types in Product Reviews

2025-02-20

AI Summary

This paper addresses the limitations of coarse-grained and poorly interpretable semantic parsing in product reviews. We propose an information typology framework grounded in communicative goals, systematically defining 24 fine-grained information types—the first such taxonomy in this domain. Leveraging this typology, we develop a zero-shot multi-label classifier for large-scale automatic review annotation. Subsequently, we map combinations of information types to three key analytical dimensions: review helpfulness, sentiment polarity, and rhetorical structure, thereby establishing an interpretable analytical paradigm. Experiments demonstrate statistically significant predictive performance for review helpfulness (p < 0.01) and sentiment orientation, alongside fine-grained, attribution-based explanations. The framework further enables cross-category intent identification and rhetorical effectiveness evaluation. By unifying theoretical rigor with practical applicability, our approach advances both the understanding and utilization of user-generated reviews.

Technology Category

• Natural Language Processing: Sentiment Analysis, Stylistic Analysis, and Argument Mining
• Knowledge Representation and Reasoning: Preferences
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Web Mining and Content Analysis: Sentiment analysis and opinion mining

Abstract

Information in text is communicated in a way that supports a goal for its reader. Product reviews, for example, contain opinions, tips, product descriptions, and many other types of information that provide both direct insights, as well as unexpected signals for downstream applications. We devise a typology of 24 communicative goals in sentences from the product review domain, and employ a zero-shot multi-label classifier that facilitates large-scale analyses of review data. In our experiments, we find that the combination of classes in the typology forecasts helpfulness and sentiment of reviews, while supplying explanations for these decisions. In addition, our typology enables analysis of review intent, effectiveness and rhetorical structure. Characterizing the types of information in reviews unlocks many opportunities for more effective consumption of this genre.

Problem

• Identify communicative goals in product reviews
• Classify review data for large-scale analysis
• Analyze review intent and rhetorical structure

Innovation

• Zero-shot multi-label classifier
• Typology of communicative goals
• Analysis of review intent