Rachsuda Jiamthapthaksin

A system for popular Thai slang extraction from social media content with n-gram based tokenization

With increased penetration of smart devices and internet connectivity, many Thais are more readily engaged in social media, online forums, and chat groups. As there is an increased consumption of social media content, there is a shift from the consumption of traditional medias in which formal language are used regularly such as broadcast and traditional print medias. Social media posts are a reflection of the trend, where posts usually made by younger generations usually involve communication in slang and non-formal language which is not typically available in formalized dictionaries. As the Thai population like to follow trends, one of behaviors of that many Thai social media users engage in, is to follow the latest popular social media trends in slang and word usage. As slang are changed and evolved over time, it is usually useful to have an online mining tool in which could capture the trends of emerging and popular slang. This paper proposes an approach that extracts popular Thai slang by comparing social media posts and utilizing tokenization, a dictionary based approach to extract unknown words, before expanding it by using n-gram approach to figure what are currently trending and popular slang words.

A Framework of Incorporating Thai Social Networking Data in Online Marketing Survey

With the introduction of high-speed Internet and smartphones at an affordable price range, many Thai citizens possess smartphones and utilize them as part of their daily life activities. The high mobile phones penetration and social networking usage is conductive to new approaches in performing marketing survey. This research proposes a framework that automatically incorporates Thai social networking data with online marketing survey for marketing analysis. In particular, it provides online marketing survey to a respondent, and automatically associates his/her Facebook data for further analysis. The benefits of the framework includes reducing manpower required in traditional surveys, offers easy accessibility to the respondents, automatically retrieving social networking data, and associating them the online questionnaires of each respondent for further marketing analysis.

Recommender Systems for university elective course recommendation

Recommender Systems are an ongoing research that is applied in various domains. Course recommendation is considered a challenged domain that has not been explored thoroughly. It benefits undergraduate students who need suggestion and also enhances course selection processes during the pre-registration period. This paper introduces a recommendation system for university elective courses, which recommends the courses based on the similarity between the course templates of students. This paper utilizes two popular algorithms: collaborative based recommendation using Pearson Correlation Coefficient and Alternating Least Square (ALS), and compares their performance on a dataset of academic records of university students. The experimental results show that applying ALS in this domain is superior to collaborative based with 86 percent of accuracy.

Integrating Labeled Latent Dirichlet Allocation into sentiment analysis of movie and general domains

Sentiment Analysis is an ongoing research, which involves design and development of various algorithms. The goal of this work is to improve the accuracy of widely used algorithms in sentiment analysis. To achieve it, the work proposes to integrate different preprocessing methods including Labeled Latent Dirichlet Allocation, removing stop words and using adjectives that have a significant impact on the documents sentiment, into three popular text classification algorithms: Support Vector Machine, Naïve Bayes and artificial neural network. By implementing them and using 5 real datasets in general and specific domains, the study evaluates the effectiveness of the proposed preprocessing method in sentiment analysis. The results show that it achieves improvement on both domains.

State-of-the-art Vietnamese word segmentation

Word segmentation is the first step of any tasks in Vietnamese language processing. This paper reviews state-of-the-art approaches and systems for word segmentation in Vietnamese. To have an overview of all stages from building corpora to developing toolkits, we discuss building the corpus stage, approaches applied to solve the word segmentation and existing toolkits to segment words in Vietnamese sentences. In addition, this study shows clearly the motivations on building corpus and implementing machine learning techniques to improve the accuracy for Vietnamese word segmentation. According to our observation, this study also reports a few of achievements and limitations in existing Vietnamese word segmentation systems.

Trending topic discovery of Twitter Tweets using clustering and topic modeling algorithms

There is no previous research that compares the results of k-means, CLOPE clustering and Latent Dirichlet Allocation (LDA) topic modeling algorithms for detecting trending topics on tweets. Since not all tweets contain hashtags, we considered three training data feature sets: hashtags, keywords and keywords + hashtags in this study. Our proposed methodology proved that CLOPE can also be used in a non-transactional database like Twitter data set to answer the trending topic discovery and could provide more topic patterns than k-means and LDA. Using additional feature sets has improved the results of k-means and LDA, thus, keywords + hashtags can identify more meaningful topics.

Parallelized FPA-SVM: Parallelized parameter selection and classification using Flower Pollination Algorithm and Support Vector Machine

Support Vector Machine (SVM) is one of the most popular machine learning algorithm to perform classification tasks and help organizations in different ways to improve their efficiency. A lot of studies have been made to improve SVM including speed, accuracy, and/or scalability. The algorithm possesses parameters that need precision tuning to perform well. This work proposes a novel parallelized parameter selection using Flower Pollination Algorithm (FPA) to quickly find the optimal parameters of SVM. In particular, MapReduce algorithm introduced in big data framework is applied to both FPA and SVM, which forms a fully distributed algorithm to support a large dataset. The experimental results of Parallelized FPA-SVM on real datasets show its outstanding speed in generating optimal models while maintaining high accuracy.

User preferences profiling based on user behaviors on Facebook page categories

User preference profiling is important in both social networking mining and recommender systems. Facebook provides information of page category over two hundred relating to user preferences, but the predefined categories may not fit application well. Explicitly mapping those categories to a desirable set of user preferences is a tedious task. This paper proposes an effective user profiling technique using features constructed from user behaviours on Facebook page in different categories. The models created from three well known classification algorithms: Naïve Bayes (NB), Artificial Neural Network (ANN), and Support Vector Machine (SVM) turn the raw data of user behaviours into a set of user preferences defined by application. The experiments performed on Facebook dataset show that the constructed features are implicitly reflecting user preferences and can be used to tailor the preferences as needed. Among the three algorithms SVM leverages classification performance the most with accuracy over seventy-two percent.

Incorporating Social Network Thai Text Mining with Lifestyle Segmentation Analysis

Due to popularity of social networks, smartphones and mobile Internet usage in Thailand, Thai people, especially young adults have been utilizing social network such as Facebook on the move via their smartphone in daily basis. The study proposes a framework that are attempting to incorporate social network Thai text data with online marketing survey using computer-assisted self-interview questionnaire in order to develop better, information-richer marketing segmentation. The merits of the study are to reduce manpower and errors in traditional offline survey, to offer easier access to the young adults, to retrieve and utilize social network data by combining them with lifestyle factors to develop market segmentation with better understanding of the sophisticated customers.

Clustering analysis on alumni data using abandoned and Reborn Particle Swarm Optimization

Alumni data is one of the most important data that university management uses for developing the learning process decisions. This paper applies the idea of Abandoned and Reborn PSO (AR-PSO) to convert a clustering problem into the optimization form with an objective function to minimize the ugliness of the desired clusters. This algorithm of Clustering using AR-PSO (CAR-PSO) is slightly adapted to the cluster problem domain. The generated clusters need to be examined to decide if they are acceptable. There are three evaluations; the closeness, the separation and the purity. Finally, the experiment results show that the CAR-PSO is comparable with &-means in both types of alumni data while leaving the other two clustering algorithms.