• Journal of the Korean Institute of Intelligent Systems
• /
• v.17 no.5
• /
• pp.696-700
• /
• 2007

A new method that improves the performance of the phonetic transcription based speech recognition system is presented with the speaker-independent phonetic recognizer. Since SI phoneme HMM based speech recognition system uses only the phoneme transcription of the input sentence, the storage space could be reduced greatly. However, the performance of the system is worse than that of the speaker dependent system due to the phoneme recognition errors generated from using SI models. A new training method that iteratively estimates the phonetic transcription and transformation vectors is presented to reduce the mismatch between the training utterances and a set of SI models using speaker adaptation techniques. For speaker adaptation the stochastic matching methods are used to estimate the transformation vectors. The experiments performed over actual telephone line shows that a reduction of about 45% in the error rates could be achieved as compared to the conventional method.

• Journal of Intelligence and Information Systems
• /
• v.25 no.4
• /
• pp.105-122
• /
• 2019

Dimensionality reduction is one of the methods to handle big data in text mining. For dimensionality reduction, we should consider the density of data, which has a significant influence on the performance of sentence classification. It requires lots of computations for data of higher dimensions. Eventually, it can cause lots of computational cost and overfitting in the model. Thus, the dimension reduction process is necessary to improve the performance of the model. Diverse methods have been proposed from only lessening the noise of data like misspelling or informal text to including semantic and syntactic information. On top of it, the expression and selection of the text features have impacts on the performance of the classifier for sentence classification, which is one of the fields of Natural Language Processing. The common goal of dimension reduction is to find latent space that is representative of raw data from observation space. Existing methods utilize various algorithms for dimensionality reduction, such as feature extraction and feature selection. In addition to these algorithms, word embeddings, learning low-dimensional vector space representations of words, that can capture semantic and syntactic information from data are also utilized. For improving performance, recent studies have suggested methods that the word dictionary is modified according to the positive and negative score of pre-defined words. The basic idea of this study is that similar words have similar vector representations. Once the feature selection algorithm selects the words that are not important, we thought the words that are similar to the selected words also have no impacts on sentence classification. This study proposes two ways to achieve more accurate classification that conduct selective word elimination under specific regulations and construct word embedding based on Word2Vec embedding. To select words having low importance from the text, we use information gain algorithm to measure the importance and cosine similarity to search for similar words. First, we eliminate words that have comparatively low information gain values from the raw text and form word embedding. Second, we select words additionally that are similar to the words that have a low level of information gain values and make word embedding. In the end, these filtered text and word embedding apply to the deep learning models; Convolutional Neural Network and Attention-Based Bidirectional LSTM. This study uses customer reviews on Kindle in Amazon.com, IMDB, and Yelp as datasets, and classify each data using the deep learning models. The reviews got more than five helpful votes, and the ratio of helpful votes was over 70% classified as helpful reviews. Also, Yelp only shows the number of helpful votes. We extracted 100,000 reviews which got more than five helpful votes using a random sampling method among 750,000 reviews. The minimal preprocessing was executed to each dataset, such as removing numbers and special characters from text data. To evaluate the proposed methods, we compared the performances of Word2Vec and GloVe word embeddings, which used all the words. We showed that one of the proposed methods is better than the embeddings with all the words. By removing unimportant words, we can get better performance. However, if we removed too many words, it showed that the performance was lowered. For future research, it is required to consider diverse ways of preprocessing and the in-depth analysis for the co-occurrence of words to measure similarity values among words. Also, we only applied the proposed method with Word2Vec. Other embedding methods such as GloVe, fastText, ELMo can be applied with the proposed methods, and it is possible to identify the possible combinations between word embedding methods and elimination methods.

• Journal of KIISE:Software and Applications
• /
• v.37 no.4
• /
• pp.317-322
• /
• 2010

This paper proposes how to improve performance of the Korean document sentiment-classification system using semantic properties of the sentiment words. A sentiment word means a word with sentiment, and sentiment features are defined by a set of the sentiment words which are important lexical resource for the sentiment classification. Sentiment feature represents different sentiment intensity in general field and in specific domain. In general field, we can estimate the sentiment intensity using a snippet from a search engine, while in specific domain, training data can be used for this estimation. When the sentiment intensity of the sentiment features are estimated, it is called semantic orientation and is used to estimate the sentiment intensity of the sentences in the text documents. After estimating sentiment intensity of the sentences, we apply that to the weights of sentiment features. In this paper, we evaluate our system in three different cases such as general, domain-specific, and general/domain-specific semantic orientation using support vector machine. Our experimental results show the improved performance in all cases, and, especially in general/domain-specific semantic orientation, our proposed method performs 3.1% better than a baseline system indexed by only content words.

• Science of Emotion and Sensibility
• /
• v.24 no.1
• /
• pp.91-104
• /
• 2021

This study explored users' emotional states identified from the wake-up words -"Hi, KIA!"- using a machine learning algorithm considering the user interface of passenger cars' voice. We targeted four emotional states, namely, excited, angry, desperate, and neutral, and created a total of 12 emotional scenarios in the context of car driving. Nine college students participated and recorded sentences as guided in the visualized scenario. The wake-up words were extracted from whole sentences, resulting in two data sets. We used the soundgen package and svmRadial method of caret package in open source-based R code to collect acoustic features of the recorded voices and performed machine learning-based analysis to determine the predictability of the modeled algorithm. We compared the accuracy of wake-up words (60.19%: 22%~81%) with that of whole sentences (41.51%) for all nine participants in relation to the four emotional categories. Accuracy and sensitivity performance of individual differences were noticeable, while the selected features were relatively constant. This study provides empirical evidence regarding the potential application of the wake-up words in the practice of emotion-driven user experience in communication between users and the artificial intelligence system.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.1
• /
• pp.1-8
• /
• 2021

In the political field of news articles, there are polarized and biased characteristics such as conservative and liberal, which is called political bias. We constructed keyword-based dataset to classify bias of news articles. Most embedding researches represent a sentence with sequence of morphemes. In our work, we expect that the number of unknown tokens will be reduced if the sentences are constituted by subwords that are segmented by the language model. We propose a document embedding model with subword tokenization and apply this model to SVM and feedforward neural network structure to classify the political bias. As a result of comparing the performance of the document embedding model with morphological analysis, the document embedding model with subwords showed the highest accuracy at 78.22%. It was confirmed that the number of unknown tokens was reduced by subword tokenization. Using the best performance embedding model in our bias classification task, we extract the keywords based on politicians. The bias of keywords was verified by the average similarity with the vector of politicians from each political tendency.

• Journal of Intelligence and Information Systems
• /
• v.24 no.2
• /
• pp.59-83
• /
• 2018

With the increasing importance of sentiment analysis to grasp the needs of customers and the public, various types of deep learning models have been actively applied to English texts. In the sentiment analysis of English texts by deep learning, natural language sentences included in training and test datasets are usually converted into sequences of word vectors before being entered into the deep learning models. In this case, word vectors generally refer to vector representations of words obtained through splitting a sentence by space characters. There are several ways to derive word vectors, one of which is Word2Vec used for producing the 300 dimensional Google word vectors from about 100 billion words of Google News data. They have been widely used in the studies of sentiment analysis of reviews from various fields such as restaurants, movies, laptops, cameras, etc. Unlike English, morpheme plays an essential role in sentiment analysis and sentence structure analysis in Korean, which is a typical agglutinative language with developed postpositions and endings. A morpheme can be defined as the smallest meaningful unit of a language, and a word consists of one or more morphemes. For example, for a word '예쁘고', the morphemes are '예쁘(= adjective)' and '고(=connective ending)'. Reflecting the significance of Korean morphemes, it seems reasonable to adopt the morphemes as a basic unit in Korean sentiment analysis. Therefore, in this study, we use 'morpheme vector' as an input to a deep learning model rather than 'word vector' which is mainly used in English text. The morpheme vector refers to a vector representation for the morpheme and can be derived by applying an existent word vector derivation mechanism to the sentences divided into constituent morphemes. By the way, here come some questions as follows. What is the desirable range of POS(Part-Of-Speech) tags when deriving morpheme vectors for improving the classification accuracy of a deep learning model? Is it proper to apply a typical word vector model which primarily relies on the form of words to Korean with a high homonym ratio? Will the text preprocessing such as correcting spelling or spacing errors affect the classification accuracy, especially when drawing morpheme vectors from Korean product reviews with a lot of grammatical mistakes and variations? We seek to find empirical answers to these fundamental issues, which may be encountered first when applying various deep learning models to Korean texts. As a starting point, we summarized these issues as three central research questions as follows. First, which is better effective, to use morpheme vectors from grammatically correct texts of other domain than the analysis target, or to use morpheme vectors from considerably ungrammatical texts of the same domain, as the initial input of a deep learning model? Second, what is an appropriate morpheme vector derivation method for Korean regarding the range of POS tags, homonym, text preprocessing, minimum frequency? Third, can we get a satisfactory level of classification accuracy when applying deep learning to Korean sentiment analysis? As an approach to these research questions, we generate various types of morpheme vectors reflecting the research questions and then compare the classification accuracy through a non-static CNN(Convolutional Neural Network) model taking in the morpheme vectors. As for training and test datasets, Naver Shopping's 17,260 cosmetics product reviews are used. To derive morpheme vectors, we use data from the same domain as the target one and data from other domain; Naver shopping's about 2 million cosmetics product reviews and 520,000 Naver News data arguably corresponding to Google's News data. The six primary sets of morpheme vectors constructed in this study differ in terms of the following three criteria. First, they come from two types of data source; Naver news of high grammatical correctness and Naver shopping's cosmetics product reviews of low grammatical correctness. Second, they are distinguished in the degree of data preprocessing, namely, only splitting sentences or up to additional spelling and spacing corrections after sentence separation. Third, they vary concerning the form of input fed into a word vector model; whether the morphemes themselves are entered into a word vector model or with their POS tags attached. The morpheme vectors further vary depending on the consideration range of POS tags, the minimum frequency of morphemes included, and the random initialization range. All morpheme vectors are derived through CBOW(Continuous Bag-Of-Words) model with the context window 5 and the vector dimension 300. It seems that utilizing the same domain text even with a lower degree of grammatical correctness, performing spelling and spacing corrections as well as sentence splitting, and incorporating morphemes of any POS tags including incomprehensible category lead to the better classification accuracy. The POS tag attachment, which is devised for the high proportion of homonyms in Korean, and the minimum frequency standard for the morpheme to be included seem not to have any definite influence on the classification accuracy.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.101-104
• /
• 1999

본 논문에서는 음성 신호를 이용해서 화자의 감정을 인식하기 위해 3가지 시스템을 구축하고 이들의 성능을 비교해 보았다. 인식 대상으로 하는 감정은 기쁨, 슬픔, 화남, 두려움, 지루함, 평상시의 감정이고, 각 감정에 대한 감정 음성 데이터베이스를 직접 구축하였다. 피치와 에너지 정보를 감성 인식의 특징으로 이용하였고, 인식 알고리듬은 MLB(Maximum-Likelihood Bayes)분류기, NN(Nearest Neighbor)분류기 및 HMM(Hidden Markov Model)분류기를 이용하였다. 이 중 MLB 분류기와 NN 분류기에서는 특징벡터로 피치와 에너지의 평균과 표준편차, 최대값 등 통계적인 정보를 이용하였고, TMM 분류기에서는 각 프레임에서의 델타 피치와 델타델타 피치, 델타 에너지와 델타델타 에너지 등 시간적 정보를 이용하였다. 실험은 화자종속, 문장독립형 방식으로 하였고, 인식 실험 결과는 MLB를 이용해서 $68.9\%, NN을 이용해서 $66.7\%를 얻었고, HMM 분류기를 이용해서 $89.30\%를 얻었다.

• Journal of the Korea Society of Computer and Information
• /
• v.12 no.5
• /
• pp.121-129
• /
• 2007

Scalarization is a process that a parallel construct like an array statement of Fortran 90 or FORALL of HPF is converted into sequential loops that maintain the correct semantics. Most compilers of HPF, recognized as a standard data parallel language, convert a HPF program into a Fortran 77 program inserted message passing primitives. During scalariztion, a parallel construct FORALL should be translated into Fortran 77 DO loops maintaining the semantics of FORALL. In this paper, we propose a scalarization algorithm which converts a FORALL construct into a DO loop with improved performance. For this, we define and use a relation distance vector to keep necessary dependence informations. Then we evaluate execution times of the codes generated by our method and by PARADIGM compiler method for various array sizes.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2000.12a
• /
• pp.181-184
• /
• 2000

The new method of feature extraction is proposed, considering the statistic feature of human voice, unlike the conventional methods of voice extraction. PCA(principal Component Analysis) is applied to this new method. PCA removes the repeating of data after finding the axis direction which has the greatest variance in input dimension. Then the new method is applied to real voice recognition to assess performance. When results of the number recognition in this paper and the conventional Mel-Cepstrum of voice feature parameter are compared, there is 0.5% difference of recognition rate. Better recognition rate is expected than word or sentence recognition in that less convergence time than the conventional method in extracting voice feature. Also, better recognition tate is expected when the optimum vector is used by statistic feature of data.

• The Journal of Society for e-Business Studies
• /
• v.19 no.3
• /
• pp.107-123
• /
• 2014

Current text mining techniques suffer from the problem that the conventional text representation models cannot express the semantic or conceptual information for the textual documents written with natural languages. The conventional text models represent the textual documents as bag of words, which include vector space model, Boolean model, statistical model, and tensor space model. These models express documents only with the term literals for indexing and the frequency-based weights for their corresponding terms; that is, they ignore semantical information, sequential order information, and structural information of terms. Most of the text mining techniques have been developed assuming that the given documents are represented as 'bag-of-words' based text models. However, currently, confronting the big data era, a new paradigm of text representation model is required which can analyse huge amounts of textual documents more precisely. Our text model regards the 'concept' as an independent space equated with the 'term' and 'document' spaces used in the vector space model, and it expresses the relatedness among the three spaces. To develop the concept space, we use Wikipedia data, each of which defines a single concept. Consequently, a document collection is represented as a 3-order tensor with semantic information, and then the proposed model is called text cuboid model in our paper. Through experiments using the popular 20NewsGroup document corpus, we prove the superiority of the proposed text model in terms of document clustering and concept clustering.