Sentiment classification is an instrument of natural language processing tasks in text analysis to measure customer feedback from given documents such as product reviews, news, and texts. This research aims to experiment with Thai financial news sentiment classification and evaluate sentiment classification performance. In this research, we show financial news sentiment classification experimental results when comparing supervised and semi-supervised methods. In the research methodology, we use PyThaiNLP to tokenize and remove stopwords and split datasets into 85% of the training set and 15% of the testing set. Next, we classify sentiment using machine learning and deep learning approaches with feature extraction such as bag-of-words, term frequency–inverse document frequency, and word embedding (Word2Vec and Bidirectional Encoder Representations from Transformers (BERT)) in given texts. The results show that support vector machine with the BERT model yields the best performance at 83.38%; in contrast, the random forest classifier with bag-of-words yields the worst performance at 54.10% in the machine learning approach. Another experiment reveals that long short-term memory with the BERT model yields the best performance at 84.07% in contrast to the convolutional neural network with bag-of-words, which yields the worst performance at 69.80% in the deep learning approach. The results imply that support vector machine, convolutional neural network, and long short-term memory are suitable for classifying sentiment in complex structure language. From this study, we observe the importance of sentiment classification tools between supervised and semi-supervised learning, and we look forward to furthering this work.

Inductive learning of first-order theory based on examples has serious bottleneck in the enormous hypothesis search space needed, making existing learning approaches perform poorly when compared to the propositional approach. Moreover, in order to choose the appropiate candidates, all Inductive Logic Programming (ILP) systems only use quantitive information, e.g. number of examples covered and length of rules, which is insufficient for search space having many similar candidates. This paper introduces a novel approach to improve ILP by incorporating the qualitative information into the search heuristics by focusing only on a kind of data where one instance consists of several parts, as well as relations among parts. This approach aims to find the hypothesis describing each class by using both individual and relational characteristics of parts of examples. This kind of data can be found in various domains, especially in representing chemical compound structure. Each compound is composed of atoms as parts, and bonds as relations between two atoms. We apply the proposed approach for discovering rules describing the activity of compounds from their structures from two real-world datasets: mutagenicity in nitroaromatic compounds and dopamine antagonist compounds. The results were compared to the existing method using ten-fold cross validation, and we found that the proposed method significantly produced more accurate results in prediction.