Effects of an Algorithm with a Recommendation Tree for Indirect Speech Acts

Abstract

For context-based recommendation systems, it is important to determine intentions from indirect speech acts.An algorithm of deriving intentions from indirect speech acts has been proposed, but the algorithm included unclear portions and there were no important experimental results for kinds of speech acts. Therefore, this paper proposes an improved algorithm and two experimental observations are discussed for accuracies and kinds of answers in indirect speech acts.Logical formulas are rewritten to if-else statements and the number of conditions is reduced from 24 to 8 in the algorithm. From experimental results, it is verified that the correct rate of the proposed method is 48.2 points higher than the one of the traditional method in indirect speech acts. Answers of "what" most include indirect speech acts and the accuracy of the proposed method is 53.8 points higher than the traditional one in them.