• Journal of Intelligence and Information Systems
• /
• v.28 no.2
• /
• pp.191-206
• /
• 2022

Recently, it is a de-facto approach to utilize a pre-trained language model(PLM) to achieve the state-of-the-art performance for various natural language tasks(called downstream tasks) such as sentiment analysis and question answering. However, similar to any other machine learning method, PLM tends to depend on the data distribution seen during the training phase and shows worse performance on the unseen (Out-of-Distribution) domain. Due to the aforementioned reason, there have been many efforts to develop domain-specified PLM for various fields such as medical and legal industries. In this paper, we discuss the training of a finance domain-specified PLM for the Korean language and its applications. Our finance domain-specified PLM, KB-BERT, is trained on a carefully curated financial corpus that includes domain-specific documents such as financial reports. We provide extensive performance evaluation results on three natural language tasks, topic classification, sentiment analysis, and question answering. Compared to the state-of-the-art Korean PLM models such as KoELECTRA and KLUE-RoBERTa, KB-BERT shows comparable performance on general datasets based on common corpora like Wikipedia and news articles. Moreover, KB-BERT outperforms compared models on finance domain datasets that require finance-specific knowledge to solve given problems.

• Journal of Intelligence and Information Systems
• /
• v.28 no.1
• /
• pp.131-154
• /
• 2022

Recently, researches on unstructured data analysis have been actively conducted with the development of information and communication technology. In particular, topic modeling is a representative technique for discovering core topics from massive text data. In the early stages of topic modeling, most studies focused only on topic discovery. As the topic modeling field matured, studies on the change of the topic according to the change of time began to be carried out. Accordingly, interest in dynamic topic modeling that handle changes in keywords constituting the topic is also increasing. Dynamic topic modeling identifies major topics from the data of the initial period and manages the change and flow of topics in a way that utilizes topic information of the previous period to derive further topics in subsequent periods. However, it is very difficult to understand and interpret the results of dynamic topic modeling. The results of traditional dynamic topic modeling simply reveal changes in keywords and their rankings. However, this information is insufficient to represent how the meaning of the topic has changed. Therefore, in this study, we propose a method to visualize topics by period by reflecting the meaning of keywords in each topic. In addition, we propose a method that can intuitively interpret changes in topics and relationships between or among topics. The detailed method of visualizing topics by period is as follows. In the first step, dynamic topic modeling is implemented to derive the top keywords of each period and their weight from text data. In the second step, we derive vectors of top keywords of each topic from the pre-trained word embedding model. Then, we perform dimension reduction for the extracted vectors. Then, we formulate a semantic vector of each topic by calculating weight sum of keywords in each vector using topic weight of each keyword. In the third step, we visualize the semantic vector of each topic using matplotlib, and analyze the relationship between or among the topics based on the visualized result. The change of topic can be interpreted in the following manners. From the result of dynamic topic modeling, we identify rising top 5 keywords and descending top 5 keywords for each period to show the change of the topic. Existing many topic visualization studies usually visualize keywords of each topic, but our approach proposed in this study differs from previous studies in that it attempts to visualize each topic itself. To evaluate the practical applicability of the proposed methodology, we performed an experiment on 1,847 abstracts of artificial intelligence-related papers. The experiment was performed by dividing abstracts of artificial intelligence-related papers into three periods (2016-2017, 2018-2019, 2020-2021). We selected seven topics based on the consistency score, and utilized the pre-trained word embedding model of Word2vec trained with 'Wikipedia', an Internet encyclopedia. Based on the proposed methodology, we generated a semantic vector for each topic. Through this, by reflecting the meaning of keywords, we visualized and interpreted the themes by period. Through these experiments, we confirmed that the rising and descending of the topic weight of a keyword can be usefully used to interpret the semantic change of the corresponding topic and to grasp the relationship among topics. In this study, to overcome the limitations of dynamic topic modeling results, we used word embedding and dimension reduction techniques to visualize topics by era. The results of this study are meaningful in that they broadened the scope of topic understanding through the visualization of dynamic topic modeling results. In addition, the academic contribution can be acknowledged in that it laid the foundation for follow-up studies using various word embeddings and dimensionality reduction techniques to improve the performance of the proposed methodology.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.109-125
• /
• 2019

As interest on intelligent search engines increases, various studies have been conducted to extract and utilize the features related to products intelligencely. In particular, when users search for goods in e-commerce search engines, the 'color' of a product is an important feature that describes the product. Therefore, it is necessary to deal with the synonyms of color terms in order to produce accurate results to user's color-related queries. Previous studies have suggested dictionary-based approach to process synonyms for color features. However, the dictionary-based approach has a limitation that it cannot handle unregistered color-related terms in user queries. In order to overcome the limitation of the conventional methods, this research proposes a model which extracts RGB values from an internet search engine in real time, and outputs similar color names based on designated color information. At first, a color term dictionary was constructed which includes color names and R, G, B values of each color from Korean color standard digital palette program and the Wikipedia color list for the basic color search. The dictionary has been made more robust by adding 138 color names converted from English color names to foreign words in Korean, and with corresponding RGB values. Therefore, the fininal color dictionary includes a total of 671 color names and corresponding RGB values. The method proposed in this research starts by searching for a specific color which a user searched for. Then, the presence of the searched color in the built-in color dictionary is checked. If there exists the color in the dictionary, the RGB values of the color in the dictioanry are used as reference values of the retrieved color. If the searched color does not exist in the dictionary, the top-5 Google image search results of the searched color are crawled and average RGB values are extracted in certain middle area of each image. To extract the RGB values in images, a variety of different ways was attempted since there are limits to simply obtain the average of the RGB values of the center area of images. As a result, clustering RGB values in image's certain area and making average value of the cluster with the highest density as the reference values showed the best performance. Based on the reference RGB values of the searched color, the RGB values of all the colors in the color dictionary constructed aforetime are compared. Then a color list is created with colors within the range of ${\pm}50$ for each R value, G value, and B value. Finally, using the Euclidean distance between the above results and the reference RGB values of the searched color, the color with the highest similarity from up to five colors becomes the final outcome. In order to evaluate the usefulness of the proposed method, we performed an experiment. In the experiment, 300 color names and corresponding color RGB values by the questionnaires were obtained. They are used to compare the RGB values obtained from four different methods including the proposed method. The average euclidean distance of CIE-Lab using our method was about 13.85, which showed a relatively low distance compared to 3088 for the case using synonym dictionary only and 30.38 for the case using the dictionary with Korean synonym website WordNet. The case which didn't use clustering method of the proposed method showed 13.88 of average euclidean distance, which implies the DBSCAN clustering of the proposed method can reduce the Euclidean distance. This research suggests a new color synonym processing method based on RGB values that combines the dictionary method with the real time synonym processing method for new color names. This method enables to get rid of the limit of the dictionary-based approach which is a conventional synonym processing method. This research can contribute to improve the intelligence of e-commerce search systems especially on the color searching feature.