Improving Text-to-SQL with a Hybrid Decoding Method.

Text-to-SQL is a task that converts natural language questions into SQL queries. Recent text-to-SQL models employ two decoding methods: sketch-based and generation-based, but each has its own shortcomings. The sketch-based method has limitations in performance as it does not reflect the relevance between SQL elements, while the generation-based method may increase inference time and cause syntactic errors. Therefore, we propose a novel decoding method, Hybrid decoder, which combines both methods. This reflects inter-SQL element information and defines elements that can be generated, enabling the generation of syntactically accurate SQL queries. Additionally, we introduce a Value prediction module for predicting values in the WHERE clause. It simplifies the decoding process and reduces the size of vocabulary by predicting values at once, regardless of the number of conditions. The results of evaluating the significance of Hybrid decoder indicate that it improves performance by effectively incorporating mutual information among SQL elements, compared to the sketch-based method. It also efficiently generates SQL queries by simplifying the decoding process in the generation-based method. In addition, we design a new evaluation measure to evaluate if it generates syntactically correct SQL queries. The result demonstrates that the proposed model generates syntactically accurate SQL queries.

Improving Document-Level Sentiment Classification Using Importance of Sentences.

Previous researchers have considered sentiment analysis as a document classification task, in which input documents are classified into predefined sentiment classes. Although there are sentences in a document that support important evidences for sentiment analysis and sentences that do not, they have treated the document as a bag of sentences. In other words, they have not considered the importance of each sentence in the document. To effectively determine polarity of a document, each sentence in the document should be dealt with different degrees of importance. To address this problem, we propose a document-level sentence classification model based on deep neural networks, in which the importance degrees of sentences in documents are automatically determined through gate mechanisms. To verify our new sentiment analysis model, we conducted experiments using the sentiment datasets in the four different domains such as movie reviews, hotel reviews, restaurant reviews, and music reviews. In the experiments, the proposed model outperformed previous state-of-the-art models that do not consider importance differences of sentences in a document. The experimental results show that the importance of sentences should be considered in a document-level sentiment classification task.

A neural network model with feature selection for Korean speech act classification.

A speech act is a linguistic action intended by a speaker. Speech act classification is an essential part of a dialogue understanding system because the speech act of an utterance is closely tied with the user's intention in the utterance. We propose a neural network model for Korean speech act classification. In addition, we propose a method that extracts morphological features from surface utterances and selects effective ones among the morphological features. Using the feature selection method, the proposed neural network can partially increase precision and decrease training time. In the experiment, the proposed neural network showed better results than other models using comparatively high-level linguistic features. Based on the experimental result, we believe that the proposed neural network model is suitable for real field applications because it is easy to expand the neural network model into other domains. Moreover, we found that neural networks can be useful in speech act classification if we can convert surface sentences into vectors with fixed dimensions by using an effective feature selection method.