Sentence Similarity Measurement for Smart Education Based on Thematic Role and Semantic Network Techniques

Abstract

In smart education, Automated Short Essay Assessment is a subjective assessment that emphasizes on contents. Word Order Technique and Syntactic-Semantic Knowledge Technique have been used in previous researches. However, it cannot differentiate sentence pair that is not similar semantically and only proven to produce excellent result on short sentence. Thematic Role for every significant argument seems able to provide information regarding the relations between the word. The lack of Malay lexical semantics database has caused constraints in solving ambiguity and backward tracking problems. Wordnet Semantics Network calculates semantic similarities of two synsets (token) by taking into account the depth of semantic relations. This study is conducted on Compiler course in Malay. The average of f-measure test accuracy rate is 93.53% using Thematic Role and Semantic Network Techniques compared to 82.36% using Pola Grammar Technique. The findings will contribute in achieving smart education as one of the elements in smart city.