[Paper Review] A New Similarity Measure for Taxonomy Based on Edge Counting

Manjula Shenoy. K, K. C. Shet+1 more|arXiv (Cornell University)|Nov 20, 2012

Advanced Text Analysis Techniques4 citations

TL;DR

This paper proposes a novel taxonomy-based similarity measure that enhances Wu and Palmer's edge-counting method by incorporating the shortest path between concepts and the overall taxonomy depth, introducing a penalization factor to correct inconsistencies where nearby concepts in a hierarchy receive higher similarity scores than those in the same branch. The result is a more semantically accurate and robust similarity measure for applications in information retrieval, ontology matching, and text mining.

ABSTRACT

This paper introduces a new similarity measure based on edge counting in a taxonomy like WorldNet or Ontology. Measurement of similarity between text segments or concepts is very useful for many applications like information retrieval, ontology matching, text mining, and question answering and so on. Several measures have been developed for measuring similarity between two concepts: out of these we see that the measure given by Wu and Palmer [1] is simple, and gives good performance. Our measure is based on their measure but strengthens it. Wu and Palmer [1] measure has a disadvantage that it does not consider how far the concepts are semantically. In our measure we include the shortest path between the concepts and the depth of whole taxonomy together with the distances used in Wu and Palmer [1]. Also the measure has following disadvantage i.e. in some situations, the similarity of two elements of an IS-A ontology contained in the neighborhood exceeds the similarity value of two elements contained in the same hierarchy. Our measure introduces a penalization factor for this case based upon shortest length between the concepts and depth of whole taxonomy.

Motivation & Objective

• Address the limitation in Wu and Palmer's similarity measure, which fails to account for semantic distance between concepts.
• Correct the counterintuitive result where concepts in the same hierarchy receive lower similarity scores than those in neighboring branches.
• Develop a similarity measure that incorporates both the shortest path between concepts and the depth of the entire taxonomy.
• Introduce a penalization factor to ensure that similarity scores reflect true semantic relatedness in IS-A ontologies.
• Improve performance in applications such as information retrieval, ontology matching, and question answering by enhancing concept similarity assessment.

Proposed method

• Extend Wu and Palmer's edge-counting approach by integrating the shortest path length between two concepts in the taxonomy.
• Incorporate the depth of the entire taxonomy as a normalization factor to improve consistency across different hierarchical levels.
• Define a new similarity formula that combines the shortest path distance and the depth of the taxonomy to reduce bias in similarity scores.
• Introduce a penalization factor based on the relative position of concepts and their distance from the root to correct for misleading similarity rankings.
• Use the Lowest Common Subsumer (LCS) concept to determine the point of convergence between two concepts in the taxonomy.
• Normalize similarity scores using a function that accounts for both the depth of the LCS and the total depth of the taxonomy.

Experimental results

Research questions

• RQ1 How can we improve the semantic accuracy of concept similarity measures in IS-A taxonomies?
• RQ2 Why does Wu and Palmer's measure produce inconsistent results when comparing concepts in the same hierarchy versus neighboring branches?
• RQ3 Can incorporating the shortest path and overall taxonomy depth improve similarity score consistency?
• RQ4 What is the impact of a penalization factor on similarity scores for closely related but differently positioned concepts?
• RQ5 How does the proposed measure compare to existing methods in terms of robustness and semantic correctness?

Key findings

• The proposed similarity measure resolves the issue where concepts in the same hierarchy receive lower similarity scores than those in neighboring branches.
• By integrating the shortest path and taxonomy depth, the measure provides more consistent and semantically meaningful similarity scores.
• The penalization factor effectively reduces overestimation of similarity for concepts that are close in the hierarchy but far from the root.
• The method improves upon Wu and Palmer's approach by ensuring that similarity values reflect true conceptual relatedness rather than structural position alone.
• The new measure demonstrates enhanced reliability in applications requiring accurate concept similarity, such as ontology matching and information retrieval.
• Theoretical analysis confirms that the measure maintains desirable properties such as symmetry and proper normalization across different taxonomy depths.