SENTENCE SIMILARITY MEASUREMENT BASED ONTHEMATIC ROLE AND SEMANTIC NETWORKTECHNIQUES

Abstract

Automated Short Essay Assessment is a subjective assessment that emphasizes important contents more than writing style. 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 annotation for every significant argument seems able to provide information regarding the relations between the word. 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 set in Malay that comprise of passive, simple, negative, mixed and complex questions. To prove the effectiveness of the techniques, the test result that apply Grammar Pola Technique was used as a benchmark because the study uses the same data set. The average of f-measure test accuracy rate is 93.53% when using Thematic Role and Semantic Network Techniques compared to 82.36% accuracy rate when using Pola Grammar Technique. The result of Thematic Role can be used on research involving Malay linguistic to test sentence structure matching that has verb by considering types of sentence.