• Journal of Intelligence and Information Systems
• /
• v.26 no.4
• /
• pp.1-25
• /
• 2020

Sentiment Analysis (SA) is a Natural Language Processing (NLP) task that analyzes the sentiments consumers or the public feel about an arbitrary object from written texts. Furthermore, Aspect-Based Sentiment Analysis (ABSA) is a fine-grained analysis of the sentiments towards each aspect of an object. Since having a more practical value in terms of business, ABSA is drawing attention from both academic and industrial organizations. When there is a review that says "The restaurant is expensive but the food is really fantastic", for example, the general SA evaluates the overall sentiment towards the 'restaurant' as 'positive', while ABSA identifies the restaurant's aspect 'price' as 'negative' and 'food' aspect as 'positive'. Thus, ABSA enables a more specific and effective marketing strategy. In order to perform ABSA, it is necessary to identify what are the aspect terms or aspect categories included in the text, and judge the sentiments towards them. Accordingly, there exist four main areas in ABSA; aspect term extraction, aspect category detection, Aspect Term Sentiment Classification (ATSC), and Aspect Category Sentiment Classification (ACSC). It is usually conducted by extracting aspect terms and then performing ATSC to analyze sentiments for the given aspect terms, or by extracting aspect categories and then performing ACSC to analyze sentiments for the given aspect category. Here, an aspect category is expressed in one or more aspect terms, or indirectly inferred by other words. In the preceding example sentence, 'price' and 'food' are both aspect categories, and the aspect category 'food' is expressed by the aspect term 'food' included in the review. If the review sentence includes 'pasta', 'steak', or 'grilled chicken special', these can all be aspect terms for the aspect category 'food'. As such, an aspect category referred to by one or more specific aspect terms is called an explicit aspect. On the other hand, the aspect category like 'price', which does not have any specific aspect terms but can be indirectly guessed with an emotional word 'expensive,' is called an implicit aspect. So far, the 'aspect category' has been used to avoid confusion about 'aspect term'. From now on, we will consider 'aspect category' and 'aspect' as the same concept and use the word 'aspect' more for convenience. And one thing to note is that ATSC analyzes the sentiment towards given aspect terms, so it deals only with explicit aspects, and ACSC treats not only explicit aspects but also implicit aspects. This study seeks to find answers to the following issues ignored in the previous studies when applying the BERT pre-trained language model to ACSC and derives superior ACSC models. First, is it more effective to reflect the output vector of tokens for aspect categories than to use only the final output vector of [CLS] token as a classification vector? Second, is there any performance difference between QA (Question Answering) and NLI (Natural Language Inference) types in the sentence-pair configuration of input data? Third, is there any performance difference according to the order of sentence including aspect category in the QA or NLI type sentence-pair configuration of input data? To achieve these research objectives, we implemented 12 ACSC models and conducted experiments on 4 English benchmark datasets. As a result, ACSC models that provide performance beyond the existing studies without expanding the training dataset were derived. In addition, it was found that it is more effective to reflect the output vector of the aspect category token than to use only the output vector for the [CLS] token as a classification vector. It was also found that QA type input generally provides better performance than NLI, and the order of the sentence with the aspect category in QA type is irrelevant with performance. There may be some differences depending on the characteristics of the dataset, but when using NLI type sentence-pair input, placing the sentence containing the aspect category second seems to provide better performance. The new methodology for designing the ACSC model used in this study could be similarly applied to other studies such as ATSC.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.3
• /
• pp.9-15
• /
• 2019

Large global online shopping malls, such as Amazon, offer services in English or in the language of a country when their products are sold. Since many customers purchase products based on the product reviews, the shopping malls actively utilize the sentimental analysis technique in judging preference of each product using the large amount of review data that the customer has written. And the result of such analysis can be used for the marketing to look the potential shoppers. However, it is difficult to apply this English-based semantic analysis system to different languages used around the world. In this study, more than 500,000 data from Amazon fine food reviews was used for training a deep learning based system. First, sentiment analysis evaluation experiments were carried out with three models of English test data. Secondly, the same data was translated into seven languages (Korean, Japanese, Chinese, Vietnamese, French, German and English) and then the similar experiments were done. The result suggests that although the accuracy of the sentimental analysis was 2.77% lower than the average of the seven countries (91.59%) compared to the English (94.35%), it is believed that the results of the experiment can be used for practical applications.

• Journal of Intelligence and Information Systems
• /
• v.26 no.3
• /
• pp.109-125
• /
• 2020

The Ministry of National Defense is pushing for the Defense Acquisition Program to build strong defense capabilities, and it spends more than 10 trillion won annually on defense improvement. As the Defense Acquisition Program is directly related to the security of the nation as well as the lives and property of the people, it must be carried out very transparently and efficiently by experts. However, the excessive diversification of laws and regulations related to the Defense Acquisition Program has made it challenging for many working-level officials to carry out the Defense Acquisition Program smoothly. It is even known that many people realize that there are related regulations that they were unaware of until they push ahead with their work. In addition, the statutory statements related to the Defense Acquisition Program have the tendency to cause serious issues even if only a single expression is wrong within the sentence. Despite this, efforts to establish a sentence comparison system to correct this issue in real time have been minimal. Therefore, this paper tries to propose a "Comparison System between the Statement of Military Reports and Related Laws" implementation plan that uses the Siamese Network-based artificial neural network, a model in the field of natural language processing (NLP), to observe the similarity between sentences that are likely to appear in the Defense Acquisition Program related documents and those from related statutory provisions to determine and classify the risk of illegality and to make users aware of the consequences. Various artificial neural network models (Bi-LSTM, Self-Attention, D_Bi-LSTM) were studied using 3,442 pairs of "Original Sentence"(described in actual statutes) and "Edited Sentence"(edited sentences derived from "Original Sentence"). Among many Defense Acquisition Program related statutes, DEFENSE ACQUISITION PROGRAM ACT, ENFORCEMENT RULE OF THE DEFENSE ACQUISITION PROGRAM ACT, and ENFORCEMENT DECREE OF THE DEFENSE ACQUISITION PROGRAM ACT were selected. Furthermore, "Original Sentence" has the 83 provisions that actually appear in the Act. "Original Sentence" has the main 83 clauses most accessible to working-level officials in their work. "Edited Sentence" is comprised of 30 to 50 similar sentences that are likely to appear modified in the county report for each clause("Original Sentence"). During the creation of the edited sentences, the original sentences were modified using 12 certain rules, and these sentences were produced in proportion to the number of such rules, as it was the case for the original sentences. After conducting 1 : 1 sentence similarity performance evaluation experiments, it was possible to classify each "Edited Sentence" as legal or illegal with considerable accuracy. In addition, the "Edited Sentence" dataset used to train the neural network models contains a variety of actual statutory statements("Original Sentence"), which are characterized by the 12 rules. On the other hand, the models are not able to effectively classify other sentences, which appear in actual military reports, when only the "Original Sentence" and "Edited Sentence" dataset have been fed to them. The dataset is not ample enough for the model to recognize other incoming new sentences. Hence, the performance of the model was reassessed by writing an additional 120 new sentences that have better resemblance to those in the actual military report and still have association with the original sentences. Thereafter, we were able to check that the models' performances surpassed a certain level even when they were trained merely with "Original Sentence" and "Edited Sentence" data. If sufficient model learning is achieved through the improvement and expansion of the full set of learning data with the addition of the actual report appearance sentences, the models will be able to better classify other sentences coming from military reports as legal or illegal. Based on the experimental results, this study confirms the possibility and value of building "Real-Time Automated Comparison System Between Military Documents and Related Laws". The research conducted in this experiment can verify which specific clause, of several that appear in related law clause is most similar to the sentence that appears in the Defense Acquisition Program-related military reports. This helps determine whether the contents in the military report sentences are at the risk of illegality when they are compared with those in the law clauses.

• Journal of the Korean Society for information Management
• /
• v.35 no.4
• /
• pp.141-164
• /
• 2018

Recently, as the amount of academic literature has increased rapidly and complex researches have been actively conducted, researchers have difficulty in analyzing trends in previous research. In order to solve this problem, it is necessary to classify information in units of academic papers. However, in Korea, there is no academic database in which such information is provided. In this paper, we propose an automatic classification system that can classify domestic academic literature into multiple classes. To this end, first, academic documents in the technical science field described in Korean were collected and mapped according to class 600 of the DDC by using K-Means clustering technique to construct a learning set capable of multiple classification. As a result of the construction of the training set, 63,915 documents in the Korean technical science field were established except for the values in which metadata does not exist. Using this training set, we implemented and learned the automatic classification engine of academic documents based on deep learning. Experimental results obtained by hand-built experimental set-up showed 78.32% accuracy and 72.45% F1 performance for multiple classification.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.15 no.1
• /
• pp.56-61
• /
• 2022

In this paper, we classify and present AI algorithms and natural language processing methods used in chatbots. A framework that can be used to implement a chatbot is also described. A chatbot is a system with a structure that interprets the input string by constructing the user interface in a conversational manner and selects an appropriate answer to the input string from the learned data and outputs it. However, training is required to generate an appropriate set of answers to a question and hardware with considerable computational power is required. Therefore, there is a limit to the practice of not only developing companies but also students learning AI development. Currently, chatbots are replacing the existing traditional tasks, and a practice course to understand and implement the system is required. RNN and Char-CNN are used to increase the accuracy of answering questions by learning unstructured data by applying technologies such as deep learning beyond the level of responding only to standardized data. In order to implement a chatbot, it is necessary to understand such a theory. In addition, the students presented examples of implementation of the entire system by utilizing the methods that can be used for coding education and the platform where existing developers and students can implement chatbots.

• Journal of Internet Computing and Services
• /
• v.25 no.4
• /
• pp.47-56
• /
• 2024

This paper proposes a method for fine-tuning BERT-based natural language processing models to perform sentiment analysis on Korean review data. By varying the input sequence length during this process and comparing the performance, we aim to explore the optimal performance according to the input sequence length. For this purpose, text review data collected from the clothing shopping platform M was utilized. Through web scraping, review data was collected. During the data preprocessing stage, positive and negative satisfaction scores were recalibrated to improve the accuracy of the analysis. Specifically, the GPT-4 API was used to reset the labels to reflect the actual sentiment of the review texts, and data imbalance issues were addressed by adjusting the data to 6:4 ratio. The reviews on the clothing shopping platform averaged about 12 tokens in length, and to provide the optimal model suitable for this, five BERT-based pre-trained models were used in the modeling stage, focusing on input sequence length and memory usage for performance comparison. The experimental results indicated that an input sequence length of 64 generally exhibited the most appropriate performance and memory usage. In particular, the KcELECTRA model showed optimal performance and memory usage at an input sequence length of 64, achieving higher than 92% accuracy and reliability in sentiment analysis of Korean review data. Furthermore, by utilizing BERTopic, we provide a Korean review sentiment analysis process that classifies new incoming review data by category and extracts sentiment scores for each category using the final constructed model.

• Journal of Intelligence and Information Systems
• /
• v.20 no.4
• /
• pp.89-105
• /
• 2014

After emergence of Internet, social media with highly interactive Web 2.0 applications has provided very user friendly means for consumers and companies to communicate with each other. Users have routinely published contents involving their opinions and interests in social media such as blogs, forums, chatting rooms, and discussion boards, and the contents are released real-time in the Internet. For that reason, many researchers and marketers regard social media contents as the source of information for business analytics to develop business insights, and many studies have reported results on mining business intelligence from Social media content. In particular, opinion mining and sentiment analysis, as a technique to extract, classify, understand, and assess the opinions implicit in text contents, are frequently applied into social media content analysis because it emphasizes determining sentiment polarity and extracting authors' opinions. A number of frameworks, methods, techniques and tools have been presented by these researchers. However, we have found some weaknesses from their methods which are often technically complicated and are not sufficiently user-friendly for helping business decisions and planning. In this study, we attempted to formulate a more comprehensive and practical approach to conduct opinion mining with visual deliverables. First, we described the entire cycle of practical opinion mining using Social media content from the initial data gathering stage to the final presentation session. Our proposed approach to opinion mining consists of four phases: collecting, qualifying, analyzing, and visualizing. In the first phase, analysts have to choose target social media. Each target media requires different ways for analysts to gain access. There are open-API, searching tools, DB2DB interface, purchasing contents, and so son. Second phase is pre-processing to generate useful materials for meaningful analysis. If we do not remove garbage data, results of social media analysis will not provide meaningful and useful business insights. To clean social media data, natural language processing techniques should be applied. The next step is the opinion mining phase where the cleansed social media content set is to be analyzed. The qualified data set includes not only user-generated contents but also content identification information such as creation date, author name, user id, content id, hit counts, review or reply, favorite, etc. Depending on the purpose of the analysis, researchers or data analysts can select a suitable mining tool. Topic extraction and buzz analysis are usually related to market trends analysis, while sentiment analysis is utilized to conduct reputation analysis. There are also various applications, such as stock prediction, product recommendation, sales forecasting, and so on. The last phase is visualization and presentation of analysis results. The major focus and purpose of this phase are to explain results of analysis and help users to comprehend its meaning. Therefore, to the extent possible, deliverables from this phase should be made simple, clear and easy to understand, rather than complex and flashy. To illustrate our approach, we conducted a case study on a leading Korean instant noodle company. We targeted the leading company, NS Food, with 66.5% of market share; the firm has kept No. 1 position in the Korean "Ramen" business for several decades. We collected a total of 11,869 pieces of contents including blogs, forum contents and news articles. After collecting social media content data, we generated instant noodle business specific language resources for data manipulation and analysis using natural language processing. In addition, we tried to classify contents in more detail categories such as marketing features, environment, reputation, etc. In those phase, we used free ware software programs such as TM, KoNLP, ggplot2 and plyr packages in R project. As the result, we presented several useful visualization outputs like domain specific lexicons, volume and sentiment graphs, topic word cloud, heat maps, valence tree map, and other visualized images to provide vivid, full-colored examples using open library software packages of the R project. Business actors can quickly detect areas by a swift glance that are weak, strong, positive, negative, quiet or loud. Heat map is able to explain movement of sentiment or volume in categories and time matrix which shows density of color on time periods. Valence tree map, one of the most comprehensive and holistic visualization models, should be very helpful for analysts and decision makers to quickly understand the "big picture" business situation with a hierarchical structure since tree-map can present buzz volume and sentiment with a visualized result in a certain period. This case study offers real-world business insights from market sensing which would demonstrate to practical-minded business users how they can use these types of results for timely decision making in response to on-going changes in the market. We believe our approach can provide practical and reliable guide to opinion mining with visualized results that are immediately useful, not just in food industry but in other industries as well.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.183-203
• /
• 2018

News articles are the most suitable medium for examining the events occurring at home and abroad. Especially, as the development of information and communication technology has brought various kinds of online news media, the news about the events occurring in society has increased greatly. So automatically summarizing key events from massive amounts of news data will help users to look at many of the events at a glance. In addition, if we build and provide an event network based on the relevance of events, it will be able to greatly help the reader in understanding the current events. In this study, we propose a method for extracting event networks from large news text data. To this end, we first collected Korean political and social articles from March 2016 to March 2017, and integrated the synonyms by leaving only meaningful words through preprocessing using NPMI and Word2Vec. Latent Dirichlet allocation (LDA) topic modeling was used to calculate the subject distribution by date and to find the peak of the subject distribution and to detect the event. A total of 32 topics were extracted from the topic modeling, and the point of occurrence of the event was deduced by looking at the point at which each subject distribution surged. As a result, a total of 85 events were detected, but the final 16 events were filtered and presented using the Gaussian smoothing technique. We also calculated the relevance score between events detected to construct the event network. Using the cosine coefficient between the co-occurred events, we calculated the relevance between the events and connected the events to construct the event network. Finally, we set up the event network by setting each event to each vertex and the relevance score between events to the vertices connecting the vertices. The event network constructed in our methods helped us to sort out major events in the political and social fields in Korea that occurred in the last one year in chronological order and at the same time identify which events are related to certain events. Our approach differs from existing event detection methods in that LDA topic modeling makes it possible to easily analyze large amounts of data and to identify the relevance of events that were difficult to detect in existing event detection. We applied various text mining techniques and Word2vec technique in the text preprocessing to improve the accuracy of the extraction of proper nouns and synthetic nouns, which have been difficult in analyzing existing Korean texts, can be found. In this study, the detection and network configuration techniques of the event have the following advantages in practical application. First, LDA topic modeling, which is unsupervised learning, can easily analyze subject and topic words and distribution from huge amount of data. Also, by using the date information of the collected news articles, it is possible to express the distribution by topic in a time series. Second, we can find out the connection of events in the form of present and summarized form by calculating relevance score and constructing event network by using simultaneous occurrence of topics that are difficult to grasp in existing event detection. It can be seen from the fact that the inter-event relevance-based event network proposed in this study was actually constructed in order of occurrence time. It is also possible to identify what happened as a starting point for a series of events through the event network. The limitation of this study is that the characteristics of LDA topic modeling have different results according to the initial parameters and the number of subjects, and the subject and event name of the analysis result should be given by the subjective judgment of the researcher. Also, since each topic is assumed to be exclusive and independent, it does not take into account the relevance between themes. Subsequent studies need to calculate the relevance between events that are not covered in this study or those that belong to the same subject.