GSAS: Dissertations
Permanent URI for this collection
Browse
Recent Submissions
Now showing 1 - 6 of 172
- ItemA provably group authentication protocol for various LTE networks(National Institute of Development Administration, 2018) Boriphat Kijjabuncha; Pipat HiranvanichakornGroup authentication is beneficial for group work in the Long Term Evolution (LTE) networks because it reduces the traffic of networks. For practical use, members of a group should be able to come from different network providers. In addition, while some group members use a network service, others may use other network services. Although the group members are on different networks, they should be able to work together. To fulfill these needs, we propose a secure group authentication protocol (SEGA) in which each group member uses his/her long-term private key and public key to create shared secret (keys) with network devices, such as Home and mobile management entity (MME). These shared keys are computed by using the DiffieHellman key exchange and are utilized in the authentication process. By using this technique instead of pre-shared keys between mobile devices and network devices, SEGA is flexible and scalable. In SE-GA, only the first member in an MME’s area has to authenticate himself/herself with the Home, while the remaining members in the area can authenticate directly with the MME. Thus the protocol reduces the amount of network usage. In this research, authentication proof is also given using the well-known BAN logic. Security analysis of the proposed protocol is also given and a comparison of our protocol with SE-AKA and GLARM was demonstrated. According to the comparison, we can see that the proposed protocol outperforms the former ones.
- ItemMulti-objective genetic algorithm for supervised clustering(National Institute of Development Administration, 2018) Vipa Thananant; Surapong AuwatanamongkolSupervised clustering organizes data instances into clusters on the basis of similarities between the data instances as well as class labels for the data instances. Supervised clustering seeks to meet multiple objectives, such as compactness of clusters, homogeneity of data in clusters with respect to their class labels, and separateness of clusters. With these objectives in mind, a new supervised clustering algorithm based on a multi-objective crowding genetic algorithm, named SC-MOGA, is proposed in this thesis. The algorithm searches for the optimal clustering solution that simultaneously achieves the three objectives mentioned above. The SC-MOGA performs very well on a small dataset, but for a large dataset it may not be able to converge to an optimal solution or can take a very long running time to converge to a solution. Hence, a data sampling method based on the Bisecting K-Means algorithm is also introduced, to find representatives for supervised clustering. This method groups the data instances of a dataset into small clusters, each containing data instances with the same class label. Data representatives are then randomly selected from each cluster. The experimental results show that SC-MOGA with the proposed data sampling method is very effective. It outperforms three previously proposed supervised clustering algorithms, namely SRIDHCR, LK-Means and SCEC, in terms of four cluster validity indexes. The experimental results show that the proposed data sampling method not only helps to reduce the number of data instances to be clustered by the SC-MOGA, but also enhances the quality of the data clustering results. Moreover, the biased initial approach is proposed in this thesis to find a good initial population to bias. The experimental results show that biased initial population of SCMOGA will improve clustering quality and the more percentage of biased initial population the better clustering quality.
- ItemHuman emotion recognition in Thai short text(National Institute of Development Administration, 2018) Jirawan Charoensuk; Ohm SornilEmotion classification is one of the topics in effective computing applicable in various research areas such as speech synthesis, image processing, and especially, text processing. Emotion classification is aimed at identifying a suitable emotion label for each review. In this research, a hierarchical classification framework to identify emotions (objective opinion and anger, disgust, fear, sadness, happiness, and surprise) is proposed for actual customer reviews written in Thai. The hierarchical classification framework consists of three levels: opinion, sentiment, and emotion. First, the opinion level distinguishes customers’ reviews into two types, namely objective and subjective opinions. Second, the sentiment level is used to categorize the subjective opinions as either positive or negative. Last, in the emotion level, an emotion label is assigned to an opinion as either anger, disgust, fear, happiness, sadness, or surprise. The proposed method consists of three main processes: (1) text preprocessing, (2) feature extraction, and (3) emotion classification. Text preprocessing provides necessary information and normalization of words in the reviews and comprises word segmentation, part-ofspeech (POS) tagging, word replacement, and stop-word elimination. Feature extraction is a process to construct a vector space model (VSM) for opinion classification. Five feature sets for generating the VSM are created by using a corpusand lexicon-based approach: the term frequency-inverse document frequency (Tf-Idf) of unigram words (TUW), bigram words (TBW), unigram POS (TUP), and bigram POS (TBP), and a Thai sentiment lexicon (TSL). Furthermore, a decision tree, multinomial naïve Bayes, and a support vector machine (SVM) are used as classifiers in the emotion classification process. The experimental results show that for the hierarchical approach where the subjectivity of a review is first determined, the polarity of an opinion is identified, and then the emotional label is calculated yielded the highest performance with an accuracy of 69.60%. Overall, TBW was the most effective feature subset used for filtering opinions, determining polarity, and classifying negative emotions. Lexicon resources such as TSL and the POS tag sets in the morphology level improved the accuracy of opinion filtering in two- and three-level hierarchical classification. SVM achieved a high performance in identifying contrasting opinions such as objective versus subjective opinions and positive versus negative sentiment. Meanwhile, multinomial naïve Bayes performed the best when identifying closely related emotions such as happiness versus surprise in positive emotion classification.
- ItemA penalty function in binary logistic regression(National Institute of Development Administration, 2018) Piyada Phrueksawatnon; Jirawan JitthavechAn algorithm is proposed to determine the logistic ridge parameter minimizing the MSE of the estimated parameter estimators, together with a theorem on the upperbound of the optimal logistic ridge parameter to facilitate the nonlinear optimization. A simulation is used to evaluate the relative efficiencies of the proposed estimator and other six well-known ridge estimators with respect to the maximum likelihood estimator. The simulation results confirm that the relative efficiency of the proposed estimator is highest among other well-known estimators. Finally, a real-life data set is used to repeat the evaluation and the conclusion is the same as in the simulation
- ItemAnalyzing social media content to gain competitive intelligence(National Institute of Development Administration, 2017) Jitrlada Rojratanavijit; Preecha VichitthamarosThe emergence of social media in Thailand has given millions of users a platform to express and share their opinions about products and services, among other subjects, and so Twitter is considered to be a rich source of information for companies to understand their customers by extracting and analyzing sentiment from Tweets. The main goal was to investigate the possible uses of Twitter information for businesses in Thailand to take advantage of and to solve any associated limitations caused by the semantics of the Thai language. Hence, social media content, specifically Tweets were utilized to generate Competitive Intelligence (CI). A new method for Twitter sentiment analysis called ASTS was proposed by using both supervised learning and lexiconbased techniques. Experiments were conducted using Tweets on three mobile network operator companies: AIS, DTAC, and TRUEMOVEH obtained using the Twitter search API focused on Tweets in Thai. A total of 72,661 were collected over a period of six months (from October 1, 2014 to March 31, 2015). ASTS consists of three modules: (1) data collection, (2) data pre-processing, and (3) classification and evaluation. A collection program was developed to search for keywords in the Twitter feed using the Twitter Search API and setting the language parameter “lang=th” and excluding reTweets. The process for the data pre-processing module was divided into three steps: (1) Text extraction from the Tweets, (2) Text preprocessing, and (3) Thai word segmentation. For the classification and evaluation module, the main intention was to identify opinion polarity, positive, negative, and neutral. The classification process was divided into two sub-modules: opinion filtering using supervised learning techniques and opinion polarity identification using lexiconbased techniques. Experimental results showed that the proposed method overcomes previous limitations from other studies and was very effective in most cases. The average accuracy is 84.80% with 82.42% precision, 83.88% recall and 82.97% Fmeasure. In particular, this clearly shows that opinion filtering helped to analyze Tweets more accurately. A case study approach for CI in social media aptly demonstrated the use of ASTS. Out of a total of 20,269 Tweets, 9,631 mentioned AIS (47.52%), 7,099 mentioned DTAC (35.02%), and 3,539 mentioned TRUEMOVEH (17.46%). The sentiment scores from the analysis results of using ASTS showed the overall customer sentiment for the companies. The sentiment score for TRUEMOVEH (-0.27) was slightly better than AIS (-0.38) and DTAC (-0.45). Benchmarking against competitors is essential information for CI. Strength and weakness analysis on the companies was derived using radar charts of the benchmarking of sentiment scores on the top five keyword mentions (net, wifi, promotion, switching, and employee). Furthermore, examples of using CI in terms of monitoring, opportunity events, and early warning alerts were presented. Opportunity events can be advantageous in response to negative sentiment Tweets on competing companies and can help a company to entice customers away from competing companies. Early warning alerts are based on negative sentiment Tweets on a company from which it should quickly identify customer dissatisfaction and then correct the associated problem. The results of this study show the usefulness of the proposed method for theoretical reference and as a practical guide. The findings from this analysis prove that CI extracted from social media content can help businesses to understand their customers’ opinions and compare them with those of their competitors. As a result, this research illustrates that CI from analyzing social media content has great potential to produce useful information, actionable knowledge, and critical insights for companies to enhance competitiveness and solve business problems.
- ItemThe upper bounds of the ruin probability for an insurance discrete-time risk model with proportional reinsurance and investment(National Institute of Development Administration, 2018) Apichart Luesamai; Samruam ChongcharoenIn this study, the two upper bounds of the ruin probability for discrete time risk model derived by adding two controlled factors to the classical discrete time risk model: proportional reinsurance and investment are proposed. These upper bounds are derived using an inductive method and rely on a recursive form of the finite time and/or an integral equation of ultimate (infinite time) of ruin probability which is also derived in this study. Both of the upper bounds are formulated by the assumption that the retention level of reinsurance and the amount of stock investment during each time period are controlled as constant values. The first upper bound can be used with the finite time ruin probability and the ultimate ruin probability under the condition that the value of the adjustment coefficient can be found. The second upper bound is formulated by a using new worse than used distribution. This upper bound can only be used with the finite time ruin probability, and its value can be found even though the value of the adjustment coefficient does not exist. However, this upper bound has limitations on the total claims amount which the total claims amount in each time period must come from the summation of independent and identically distributed (i.i.d.) claim amounts, and the number of claims is also i.i.d. in each time period.