Digital Library[ Search Result ]
Sentence Similarity Prediction based on Siamese CNN-Bidirectional LSTM with Self-attention
Mintae Kim, Yeongtaek Oh, Wooju Kim
http://doi.org/10.5626/JOK.2019.46.3.241
A deep learning model for semantic similarity between sentences was presented. In general, most of the models for measuring similarity word use level or morpheme level embedding. However, the attempt to apply either word use or morpheme level embedding results in higher complexity of the model due to the large size of the dictionary. To solve this problem, a Siamese CNN-Bidirectional LSTM model that utilizes phonemes instead of words or morphemes and combines long short term memory (LSTM) with 1D convolution neural networks with various window lengths that bind phonemes is proposed. For evaluation, we compared our model with Manhattan LSTM (MaLSTM) which shows good performance in measuring similarity between similar questions in the Naver Q&A dataset (similar to Kaggle Quora Question Pair).
Korean Movie-review Sentiment Analysis Using Parallel Stacked Bidirectional LSTM Model
Yeongtaek Oh, Mintae Kim, Wooju Kim
http://doi.org/10.5626/JOK.2019.46.1.45
The sentiment analysis is a field of document classification that classifies the sensitivity of text documents. The sentiment analysis methodology that employs the use of deep learning can be divided into a process of tokenizing a document, obtaining a sentence vector through embedding and classifying a vectorized document. We reviewed the methods of various existing studies and found out the appropriate methodology focusing on embedding methods and deep learning models for the Korean documents through comparative experiments. The document pre-processing method compares documents to words, syllables and phonemes. Additionally, a comparative experiment was conducted on the Naver movie review data set nsmc (naver sentiment movie corpus) from the CNN to the LSTM, bi-LSTM, stacked bi-LSTM and the newly proposed Parallel Stacked Bidirectional LSTM model. The results showed that the performance of the proposed model was higher than that of the existing simple deep learning model. Moreover, itachieved the best classification performance of approximately 88.95% through the ensemble among the models learned through other pre-processing.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr