Youngsub Han

A Method for Extracting Lexicon for Sentiment Analysis Based on Morphological Sentence Patterns

In these days, people share their emotions, opinions, and experiences of products or services using online review services on their comments, and the people concern the reviews to make decision when buying products or services. Sentiment analysis is one of the solution to observe and summarize emotional opinions from the data. In spite of high demands for developing sentiment analysis, the development of the sentiment analysis faces some challenges to analyze the data, because the data is unstructured, unlabeled, and noisy. The aspect-based sentiment analysis approach helps for more in-depth analysis, however building aspect and emotional expression is one of the challenge for the aspect-based sentiment analysis approach. Accordingly, we propose an unsupervised system for building aspect-expressions to minimize human-coding efforts. The proposed method uses morphological sentence patterns through an aspect-expression pattern recognizer. It guarantees relatively higher accuracy. As well as, we found some characteristics for selecting patterns to extracting aspect-expressions accurately. The greatest advantage of our system is performing without any human coded train-set.

Sentiment analysis on social media using morphological sentence pattern model

Social media became popular than ever as people are willing to share their emotions and opinions or to participate in social networking. Accordingly, the understanding of social media usage became important. The sentiment analysis is emerged as one of useful methods to analyze emotional stats expressed in textual data including social media data. However, this method still presents some limitations, particularly with an accuracy issue. For example, our previous sentiment analysis used a probability model and needed to adopt human-coded train-sets to maintain an acceptable accuracy level (89%). To overcome and improve this weakness, we propose an automated sentiment analysis in this paper by using the morphological sentence pattern model. We found that this new approach presented in this paper allowed us to achieve a higher level of accuracy (91.2%). The movie reviews were used for this analysis from IMDb, Rotten Tomatoes, Metacritic, YouTube and Twitter.

A real-time knowledge extracting system from social big data using distributed architecture

A huge amount of data is being generated by social media in real time. Accordingly, demands for extracting meaningful information from the social data have been dramatically increased. However, most of the previous research encompasses potential problems with data processing, management and analysis in real time. In this paper, we propose a distributed system architecture for generating meaningful information from text-based social data. The system collects data from multi-source channels, such as Twitter, YouTube, and The New York Times. Also, the system extracts terms and sentiment from each document using data mining technologies. In addition, the system uses HDFS, Map-reduce, and message service to handle the huge data. By analyzing keywords in texts and user account information, the system generates a summary of results including terms, sentiments and data variations for further analysis, including reputation, social trends, and customer reactions. The experiment results show that our approach is able to effectively process the social data in real time.

A Research for Finding Relationship Between Mass Media and Social Media Based on Agenda Setting Theory

We are living in a flood of information. We hear about lots of social issues such as politics and economies in every day from the mass media. Before the appearance social media, it is difficult to interact people’s opinions with the others about the social issues. However, we can analyze important social issues using big data generated from social media. We tried to apply the relationship between agenda setting theory and social media because we have received social issues from official accounts like news using social media, and then users shared social issues to other users, so we choose tweets of Baltimore Riot to analyze. We collected tweets related with Baltimore Riot, and then we extracted term keywords using text mining technologies such as TF-IDF. Actually, we analyzed tweets of 04-27-2015 Based on detected important words, we analyzed tweets at 5-min intervals, and we extracted tweets of mass media and others. Based on user’s profiles, we found relationship of mass media and social issues. About initial phase of the social issues as it happened, local mass media leaded about incidents, and tweets exchanged and shared in local area. After writing an influence Twitter user, social issues of Baltimore Riot spread to other areas. As a result, we could detect agenda setting theory in social media using big data technology. It implies that the local mass media led the social issues, the Baltimore Riot became one of major social issues to people at the end.

A semi-supervised tweet classification method using news articles

As the Internet and social media became more popular, the demand for Online Reputation Management (ORM) has increased. For ORM in social media, it is a priority to recognize whether a content of a social message means the named entity. Since Twitter exchanges 500 million tweets per day, it is impossible for a human to analyze this enormous amount of tweet. Because of this, an automated system for tweet disambiguation is necessary. In this paper, we propose a semi-supervised and automated system for the named entity disambiguation in Twitter based on the news articles. A classifier is built with keywords from news articles which are related to a company. The keywords are extracted based on the document frequency. For the keywords that help to discriminate company tweets, a proper threshold was found in a heuristic way. From the experiment, we found that news articles can be used to disambiguate a named entity on tweets as an external source and we verified our system performed well in some cases.