Product Aspect Clustering by Incorporating Background Knowledge for Opinion Mining

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Data Mining

Background:

• Product aspect recognition is crucial for fine-grained opinion mining.
• Current methods often overlook clustering synonymous aspects.
• Clustering similar aspects with different representations poses a significant challenge.

Purpose of the Study:

• To address the challenge of product aspect clustering.
• To develop a method for grouping synonymous aspects effectively.
• To improve the accuracy of fine-grained opinion mining.

Main Methods:

• Learning two types of background knowledge: relevant and irrelevant aspect relations.
• Enriching background knowledge using web data.
• Designing a hierarchical clustering algorithm based on aspect similarity computed from background knowledge sets.

Main Results:

• The proposed method outperforms baseline approaches in product aspect clustering.
• Experimental results show superior performance in camera and mobile phone domains.
• Web-based enrichment of background knowledge is demonstrated as feasible and effective.

Conclusions:

• The developed product aspect clustering method, leveraging background knowledge from aspect relations, significantly enhances clustering accuracy.
• Utilizing web data to expand background knowledge is a viable strategy.
• The approach offers a promising solution for more accurate fine-grained opinion mining.