Text-based LSTM Networks for Automatic Thai Love Quotes Generation on Twitter

Abstract

In this article, we propose the model for automatic Thai love quotes generation on Twitter using Text-based LSTM (Long Short-Term Memory) Network. This model is designed to learn the relationship of words in sentences from Twitter tweets with the hashtag “love quotes” and love songs 3,097 sentences 28,749 words. This approach differs from other previous research about sentences generation. The process of model training, we compare Loss from 2 input formats including with 1) Integer value 2) word2vec. The experimental has 4 approaches including 1) LSTM+Integer value with 2 words input 2) LSTM+Integer value with 3 words input 3) LSTM+Word2Vec with 2 words input and 4) LSTM+Word2Vec with 3 words input. The LSTM+word2vec showed the lowest Loss. The evaluation using Human-targeted Translation Edit Rate (HTER). The average of HTER rate of LSTM+Word2vec model with 2 input words is 0.29 and for 3 input words is 0.26