Search

This paper proposes to develop a model to detect the text semantic similarity in Thai by using a Siamese neural network with MaLSTM. As the text's intent is varied and hard to analyze, thus making the comparison between two sentences for the semantics similarity challenging. Our project wants to find similar intent of two questions by comparing the question between frequently asked questions (FAQs) and input questions from customers via Facebook Messenger of computer engineering at the King Mongkut's University of Technology Thonburi (CPE KMUTT). The data gather manually from the Pantip, CPE KMUTT FAQ, and Wikipedia, which create a dataset and data corpus. Although, the overall score performance of the model is not as we expected. Our model shows promising results, as it can detect between two questions' intent similarity. Results suggest that with the increase of the data, the model should have great potential for finding semantic similarity.

Word segmentation is an important preprocessing step in natural language processing applications, particularly in languages with no demarcation indicators including Thai. A simple method like dictionary-based segmentation does not consider the context of the sentence. This paper proposes an attention-based deep learning approach for Thai word segmentation. With the help of attention, the model can learn character correlations across the entire sentence without gradient vanishing or gradient explode problems and tokenize them into word vectors. The goal of this research is to test three different types of attention mechanisms to determine the effectiveness of word tokenization. The visualization of attention for each attention mechanism is also included as an outcome.

Latent Dirichlet Allocation(LDA) is one of the powerful techniques in extracting topics from a document. The original LDA takes the Bag-of-Word representation as the input and produces topic distributions in documents as output. The drawback of Bag-of-Word is that it represents each word with a plain one-hot encoding which does not encode the word level information. Later research in Natural Language Processing(NLP) demonstrate that word embeddings technique such as Skipgram model can provide a good representation in capturing the relationship and semantic information between words. In recent studies, many NLP tasks could gain better performance by applying the word embedding as the representation of words. In this paper, we propose Deep Word-Topic Latent Dirichlet Allocation(DWT-LDA), a new process for training LDA with word embedding. A neural network with word embedding is applied to the Collapsed Gibbs Sampling process as another choice for word topic assignment. To quantitatively evaluate our model, the topic coherence framework and topic diversity are the metrics used to compare between our approach and the original LDA. The experimental result shows that our method generates more coherent and diverse topics.

COVID-19 is undeniably one of the worst incidents in the 21st century. There are a wide variety of factors that impact the spreading of COVID-19. This paper presents the study of relations of how public policy implementation might affect the onset and the spread of COVID-19 cases. Cluster analysis was employed to identify data patterns associating with the policy implementation profiles. The results suggest that the effectiveness of policy adoption relates to the onset spreading of COVID-19. This also indicates that the decision of public administrators was critical in the latter stage of the pandemic situation management.

This paper presents a computational study of Thai legal documents using text mining and network analytic approach. Thai legal systems rely much on the existing judicial rulings. Thus, legal documents contain complex relationships and require careful examination. The objective of this study is to use text mining to model relationships between these legal documents and draw useful insights. A structure of document relationship was found as a result of the study in forms of a network that is related to the meaningful relations of legal documents. This can potentially be developed further into a document retrieval system based on how documents are related in the network.

Prosody is an essential aspect of speech, as it carries both lexical and non-lexical information. A conventional approach for studying speech prosody is to collect and analyze F0 data based on certain hypotheses and then develop a theory based on the observation as the final conclusion of the study. This process is however far from complete, as the developed theory has not been actually tested for its ability to predict actual acoustic data. This paper presents PENTATrainer2, a prosody modeling tool based on the parallel encoding and target approximation framework. PENTATrainer2 can facilitate prosody studies in testing hypotheses and theories against speech data by using an automatic analysis-by-synthesis and stochastic learning algorithm. Users can flexibly design the annotation scheme based on their own hypotheses and determine whether the hypothesized categories can lead to accurate synthetic F0 contours. PENTATrainer2 consists of three main components: multi-layer annotation, target approximation, and stochastic optimization. First, acoustic data are annotated in parallel layers, each of which corresponds to a functional category that may affect F0 contours. These layers are then compiled into unique functional combinations. The combinations represent underlying invariant representations of communicative functions and their interaction with each other. Target approximation parameters of each combination are then learned through analysis-by-synthesis and stochastic optimization. Pilot tests of PENTATrainer 2 have been conducted on Thai, Mandarin and English. The results demonstrate not only high accuracy of the synthesized F0 contours but also distinctive contrasts in the distribution of pitch target parameters. This indicates the effectiveness of PENTATrainer2 in modeling speech prosody.

This paper presents an acoustic-to-articulatory mapping of a three-dimensional theoretical vocal tract model using deep learning methods. Prominent deep learning-based network structures are explored and evaluated for their suitability in capturing the relationship between acoustic and articulatory-oriented vocal tract parameters. The dataset was synthesized from VocalTractLab, a three-dimensional theoretical articulatory synthesizer, in forms of the pairs of acoustic, represented by Mel-frequency cepstral coefficients (MFCCs), and articulatory signals, represented by 23 vocal tract parameters. The sentence structure used in the dataset generation were both monosyllabic and disyllabic vowel articulations. Models were evaluated using the root-mean-square error (RMSE) and R-squared (R2). The deep artificial neural network architecture (DNN), regulating using batch normalization, achieves the best performance for both inversion tasks, RMSE of 0.015 and R2 of 0.970 for monosyllabic vowels and RMSE of 0.015and R2 of 0.975 for disyllabic vowels. The comparison, between a formant of a sound from inverted articulatory parameters and the original synthesized sound, demonstrates that there is no statistically different between original and estimated parameters. The results indicate that the deep learning-based model is effectively estimated articulatory parameters in a three-dimensional space of a vocal tract model.

Named Entity Recognition (NER) is a handy tool for many natural language processing tasks to identify and extract a unique entity such as person, location, organization and time. In English and Chinese, NER has been thoroughly researched and is able to be applied in more practical settings. Its development in Thai is still limited because of rare resources and language difficulties such as the lack of boundary indicator for words, phrases and sentences. In this paper, we present an application of Bi-LSTM-CRF with word/character level representation, to solve this problem. Firstly, we prepared texts by tokenizing a sentence to a bunch of words. We then prepared word representation and Bi-LSTM character representation. In the end, we built a recurrent neural network combined with CRF to learn the sequence of text and extract the knowledge to build NER recognition to overcome this problem. Our model was evaluated by the NER opensource corpus from a Facebook group ThaiNLP. The results of our model yielded precision, recall, and F1 at 91.79%, 91.51% and 91.65% respectively.

Image processing and computer vision algorithms are very computationally intensive and could not be implemented on low power microcontroller used in embedded and small IoT devices. Performing computation on the cloud is not practical for always-on real-time operations and deploying a high-end microcontroller or a microprocessor consume too much power and are generally too expensive to fit into these small systems. In this paper, we propose a new programming workflow using reusable hardware modules and a graphical programming interface for implementing a complete image processing system on an FPGAs which overcomes the steep learning curve of tradition FPGAs design tool. The design can be deployed onto the low-ended FPGAs in a similar price point as a mid-range microcontroller. To demonstrate our proposed framework, we have implemented a number of image transformation operations on an FPGAs development board with Lattice iCE40 Ultra Plus FPGA and a tiny camera module. Results have shown that our designs can fit in a low-ended FPGAs while performing 5.5-345x faster and consuming 2.4-4x less power compared to current state-of-the-art microcontroller used in small embedded and IoT devices.

The Artificial Bee Colony algorithm is originally designed for solving numerical optimization problems, whereas the Travelling Salesman Problem is classified as a combinatorial optimization one. This paper proposes a combinatorial variant of the Artificial Bee Colony algorithm by using reinforcement learning update. Reinforcement values are introduced, and positive reinforcement is given to the selected cities when a better solution is found by an employed bee. On the contrary, negative reinforcement is assigned to the selected cities when a worse solution is found. Onlooker bees then select cities to update their solutions according to these reinforcement values. The performance of the proposed algorithm is tested on six benchmark problems. The results show that the algorithm with the reinforcement learning update provides better solutions than the algorithm without the reinforcement learning update in five out of six benchmark problems. The convergence rates of the algorithm with the reinforcement learning update are also faster than those of the algorithm without the reinforcement learning update.

This paper presents an overview of the Parallel Encoding and Target Approximation (PENTA) model of speech prosody, in response to an extensive critique by Arvaniti & Ladd (2009). PENTA is a framework for conceptually and computationally linking communicative meanings to fine-grained prosodic details, based on an articulatory-functional view of speech. Target Approximation simulates the articulatory realisation of underlying pitch targets – the prosodic primitives in the framework. Parallel Encoding provides an operational scheme that enables simultaneous encoding of multiple communicative functions. We also outline how PENTA can be computationally tested with a set of software tools. With the help of one of the tools, we offer a PENTA-based hypothetical account of the Greek intonational patterns reported by Arvaniti & Ladd, showing how it is possible to predict the prosodic shapes of an utterance based on the lexical and postlexical meanings it conveys.

This paper presents a performance evaluation study for small-scale deep learning neural network for Thai speech recognition task. Convolutional neural network and long short term memory networks were built with a relatively small size dataset and small constructs. The aim of this study is to determine which method would be suitable for a small-scale deep learning study. Relatively small speech corpus was used to build deep-learning neural networks with two different architectures, including convolutional neural network (CNN) model and long short term memory (LSTM) model. Models were evaluated using cross validation technique and compare to one another. The result shows that CNN outperformed LSTM for a small-scale deep learning. This suggests that with the limited dataset and small-scale architecture CNN is a more suitable choice in the speech recognition study.

The innovative existence of social network has revolutionized the way we communicate, perceive, and interact with others. Internet has become a medium that connects our lives together and attracts many people to spend time with it, and to engage in different activities from casual to business. However, not everyone is trustworthy. One of the main issues in online social networks is the information overload. Some user accounts may be created with hidden agenda and not represent actual people. Our aim in this paper is to develop and test a method for evaluate the credibility of online users based on their interactions with others. We propose an analysis method that uses several analytic measurements for ranking users in Social Networks based on the reputation models such as PageRank, SocialRank, TwitterRank and MetricRank. We also utilize content classification to improve the ranking effectiveness and integrate them to evaluate the consistency of interaction dialogues.

The necessary operation in every sector of organization is finding a talented or experienced individual to participate in a team that will handle specific tasks. Having the right people in the team would increase the team performance and allow the team to be more effective in delivery the result. Research and development play important roles in the development of science and technology. Peer-reviewed publication is one of the main methods used for evaluating research. Normally, editor of academic journal will find researchers for reviewing research based on mainly the research network. However, there are various and interdisciplinary types of thesis nowadays. They made more difficult results for finding reviewers. This paper aims to develop the system for automatically suggest relevant experts given the task of reviewing the scientific paper. This inevitably would assure the quality of the research process and provide an effective way to manage the publication cycle. The system will automatically find related reviewer for reviewing papers based on the co-author networks, reference network and the reputation model.

This paper proposes a new hybrid method for Thai word segmentation using crowd-sourced dictionary integrated with word bi-gram model. The main dictionary is extracted into basic and compound word dictionaries to improve dictionary based algorithm performance. The word segmentation process begins with heuristic exhaustive matching algorithm using basic word dictionary to generate all possible basic word sequence candidates from an input string. Then, the best candidate is selected by word bi-gram model to solve ambiguity problem. Finally, the sequence of basic words is combined into compound words with compound word dictionary. Another part of this work is applying crowdsourcing paradigm. We implemented a web application for training bi-gram model and dictionary updates from user feedbacks. This process improves the lexical knowledge of the platform over the time. The algorithm was evaluated with two corpora. With InterBEST 2009 corpus, the proposed algorithm yields average precision, recall and f-measure at 97.52%, 97.70%, and 97.63%. With social network corpus, the proposed method yields average precision, recall and f-measure at 98.47%, 98.59%, and 98.54% respectively.

The advent of high-throughput technology has made it possible to measure genome-wide expression profiles, thus providing a new basis for microarray-based diagnosis of disease states. Numerous methods have been proposed to identify biomarkers that can accurately discriminate between case and control classes. Many of the methods used only a subset of ranked genes in the pathway and may not be able to fully represent the classification boundaries for the two disease classes. The use of negatively correlated feature sets (NCFS) to obtain more relevant features in form of phenotype-correlated genes (PCOGs) and inferring pathway activities is proposed in this study. The two pathway activity inference schemes that use NCFS significantly improved the power of pathway markers to discriminate between two phenotypes classes in microarray expression datasets of breast cancer. In particular, the NCFS-i method provided better contrasting features for classification purposes. The improvement is consistent for all cases of pathways used, using both within- and across-dataset validations. The results show that the two proposed methods that use NCFS clearly outperformed other pathway-based classifiers in terms of both ROC area and discriminative score. That is, the identification of PCOGs within each pathway, especially NCFS-i method, helps to reduce noisy or variable measurements, leading to a high performance and more robust classifier. In summary, we have demonstrated that effective incorporation of pathway information into expression-based disease diagnosis and using NCFS can provide better discriminative and more robust models.

Information technology (IT) plays a vital role in healthcare administration and management. IT facilitates doctors, nurses or physicians to improve the speed and performance of the medical system as a whole. This results in accumulation of healthcare data since various types of software which generate their own data types have been widely used. Since different software were used in different medical institutions in Thailand, sharing data among institution is difficult. This research therefore proposed data exchange protocol for healthcare service in Thailand as a cloud-based platform. The system is designed to be consistent and appropriate to hospital information system (HIS) in Thailand. For healthcare data, it defines a standard for exchanging electronic healthcare information between the different software systems from different hospitals. This system also considers security and privacy under the standard of healthcare security (HIPAA) and regulation in Thailand. Furthermore, it can be applied to use for referring patient cases by developing and improving from the referral system in Thailand to make it more effective.

This paper presents the development of DOM, a mobile big data analytics engine for mining Thai public opinions. The engine takes in data from multiple well-known social network sources, and then processes them using MapReduce, a keyword-based sentiment analysis technique, and an influencer analysis algorithm to determine public opinions and sentiments of certain topics. The system was evaluated its sentiment prediction accuracy by matching the predicted result with the human sentiment and tested on various case studies. The effectiveness of the approach demonstrates the practical applications of the engine.

Accuracy of disease classification has always been a challenging goal of bioinformatics research. Microarray-based classification of disease states relies on the use of gene expression profiles of patients to identify those that have profiles differing from the control group. A number of methods have been proposed to identify diagnostic markers that can accurately discriminate between different classes of a disease. Pathway-based microarray analysis for disease classification can help improving the classification accuracy. The experimental results showed that the use of pathway activities inferred by the negatively correlated feature sets (NCFS) based methods achieved higher accuracy in disease classification than other different pathway-based feature selection methods for two breast cancer metastasis datasets.

This paper proposes an effective method for identifying functional modules of the weighted gene co-expression network using a minimum spanning tree (MST) approach coupled with network neighborhood connectivity. The MST-based gene co-expression network was reconstructed to serve as the backbone of gene co-expression network. Highly connected hub genes were identified based on the connectivity of the backbone network. All sub-networks were extracted by expanding from the hub genes to their neighborhood genes. Finally, functional modules were identified by integrating sub-networks with similar gene expression profiles. We tested the method with both simulated and autism spectrum disorder microarray data sets. The results show that our approach is better in highlighting the hub genes and can effectively identify functional modules with highly enriched pathways.

Feature selection (FS) plays a crucial role in machine learning to build a robust model for either learning or classification from a large amount of data. Among feature selection techniques, the Relief algorithm is one of the most common due to its simplicity and effectiveness. The performance of the Relief algorithm, however, could be dramatically affected by the consistency of the data patterns. For instance, Relief-F could become less accurate in the presence of noise. The accuracy would decrease further if an outlier sample was included in the dataset. Therefore, it is very important to select the samples to be included in the dataset carefully. This paper presents an effort to improve the effectiveness of Relief algorithm by filtering samples before selecting features. This method is termed Sample Filtering Relief Algorithm (SFRA). The main idea of this method is to discriminate outlier samples out of the main pattern using self organizing map (SOM) and then proceed with feature selection using the Relief algorithm. We have tested SFRA with a gene expression dataset of interferon-?(IFN-?) response of Hepatitis B patients that contains outlier data. SFRA could successfully remove outlier samples that have been verified by visual inspection by experts. Also, it has better accuracy in separating the relevant and irrelevant features than other feature selection methods considered.

This paper introduces the Common Prosody Platform (CPP), a computational platform that implements major theories and models of prosody. CPP aims at a) adapting theory-specific assumptions into computational algorithms that can generate surface prosodic forms, and b) making all the models trainable through global optimization based on automatic analysis-bysynthesis learning. CPP allows examination of prosody in much finer detail than has been previously done and provides a means for speech scientists to directly compare theories and their models. So far, four theories have been included in the platform, the Command-Response model, the Autosegmental-Metrical theory, the Task Dynamic model, and the Parallel Encoding and Target Approximation model. Preliminary tests show that all the implemented models can achieve good local contour fitting with low errors and high correlations.