Efficiently Semantic-Aware Pairwise Similarity: an Applicable Use-Case

Abstract

Pairwise similarity is an essential operation in multidisciplinary fields of study. However, its operation is expensive due to its complexity as well as the evolution of big data. Besides, similarity scores without semantics leads to neither accurate nor practical results. In this paper, we study the problem of pairwise similarity in terms of semantics, efficiency, and increment. More concretely, we build our own synonym database and organize it in a reflexive synonym index to support us in semantic-aware pairwise similarity. In addition, we employ cache to perform incremental similarity computing in the process of pair-by-pair similarity search. Moreover, we design our feature-based hierarchy to eliminate irrelevant candidate objects before doing pairwise similarity. Furthermore, we analyze a practical case and apply our method into such social network-based applications in order to demonstrate its feasibility and efficiency in reality.