The Poet Identification Using Convolutional Neural Networks

摘要

In this article, we propose to identify the amateur poet by using a Convolutional Neural Networks (CNNs). The poets were selected from the composing of Thai poem Klon-Suphap. The poems content are classified into 7 groups including with (1) royal, (2) parents and teachers, (3) fall in love, (4) broken, (5) festival, (6) advise, (7) depressed and there are poems of each poet in every groups. To identify the poet, input of model represented by the vector (Word2Vec) which had generated from Thai-Text corpus 5.9 Million words. The training data is Thai poem 900 units (baat) and testing data is Thai poem 96 units. CNNs showed the accuracy of 2 poets identification is 100%, 3 poets identification is 80.55%, 4 poets identification is 72.92% and 5 poets identification is 55.25%. In additional, we used 5 participants to read the poems of 2 poets and has predicted in testing data. The average of accuracy is 57.32% which less than the proposed model.