• Journal of information and communication convergence engineering
• /
• v.21 no.1
• /
• pp.9-16
• /
• 2023

The natural language on social network platforms has a certain front-to-back dependency in structure, and the direct conversion of Chinese text into a vector makes the dimensionality very high, thereby resulting in the low accuracy of existing text classification methods. To this end, this study establishes a deep learning model that combines a big data ultra-deep convolutional neural network (UDCNN) and long short-term memory network (LSTM). The deep structure of UDCNN is used to extract the features of text vector classification. The LSTM stores historical information to extract the context dependency of long texts, and word embedding is introduced to convert the text into low-dimensional vectors. Experiments are conducted on the social network platforms Sogou corpus and the University HowNet Chinese corpus. The research results show that compared with CNN + rand, LSTM, and other models, the neural network deep learning hybrid model can effectively improve the accuracy of text classification.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.15 no.2
• /
• pp.111-120
• /
• 2015

Protein named entity identification is one of the most essential and fundamental predecessor for extracting information about protein-protein interactions from biomedical literature. In this paper, we explore the use of abstracts of biomedical literature in MEDLINE for protein name identification and present the results of the conducted experiments. We present a robust and effective approach to classify biomedical named entities into protein and non-protein classes, based on a rich set of features: orthographic, keyword, morphological and newly introduced Protein-Score features. Our procedure shows significant performance in the experiments on GENIA corpus using Random Forest, achieving the highest values of precision 92.7%, recall 91.7%, and F-measure 92.2% for protein identification, while reducing the training and testing time significantly.

• Journal of Computing Science and Engineering
• /
• v.10 no.4
• /
• pp.103-110
• /
• 2016

Emotion recognition is an important component of affective computing, and is significant in the implementation of natural and friendly human-computer interaction. An effective approach to recognizing emotion from text is based on a machine learning technique, which deals with emotion recognition as a classification problem. However, in emotion recognition, the texts involved are usually very short, leaving a very large, sparse feature space, which decreases the performance of emotion classification. This paper proposes to resolve the problem of feature sparseness, and largely improve the emotion recognition performance from short texts by doing the following: representing short texts with word cluster features, offering a novel word clustering algorithm, and using a new feature weighting scheme. Emotion classification experiments were performed with different features and weighting schemes on a publicly available dataset. The experimental results suggest that the word cluster features and the proposed weighting scheme can partly resolve problems with feature sparseness and emotion recognition performance.

• Journal of the Korea Society for Simulation
• /
• v.11 no.3
• /
• pp.35-48
• /
• 2002

In the work presented here, we describe a method to extract TrueType features for supporting letter recognition. Even if variously existing document processing techniques have been challenged, almost few methods are capable of recognize a letter associated with its TrueType features supporting OCR free, which boost up fast processing time for image text retrieval. By reviewing the mechanism generating digital fonts and birth of TrueType, we realize that each TrueType is drawn by its contour of the glyph table. Hence, we are capable of deriving the segment with density for a letter with a specific TrueType, defined by the number of occurrence over a segment width. A certain number of occurrence appears frequently often due to the fixed segment width. We utilize letter recognition by comparing TrueType feature library of a letter with that from input word images. Experiments have been carried out to justify robustness of the proposed method showing acceptable results.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.7
• /
• pp.2407-2424
• /
• 2022

Recently, Transformer has made great progress in video retrieval tasks due to its high representation capability. For the structure of a Transformer, the cascaded self-attention modules are capable of capturing long-distance feature dependencies. However, the local feature details are likely to have deteriorated. In addition, increasing the depth of the structure is likely to produce learning bias in the learned features. In this paper, an improved Transformer structure named TransDCS (Transformer with Dynamic Convolution and Shortcut) is proposed. A Multi-head Conv-Self-Attention module is introduced to model the local dependencies and improve the efficiency of local features extraction. Meanwhile, the augmented shortcuts module based on a dual identity matrix is applied to enhance the conduction of input features, and mitigate the learning bias. The proposed model is tested on MSRVTT, LSMDC and Activity-Net benchmarks, and it surpasses all previous solutions for the video-text retrieval task. For example, on the LSMDC benchmark, a gain of about 2.3% MdR and 6.1% MnR is obtained over recently proposed multimodal-based methods.

• MALSORI
• /
• no.53
• /
• pp.75-92
• /
• 2005

Prosody Generation Based on C-ToBI Representation for Text-to-SpeechSeungwon Kim, Yu Zheng, Gary Geunbae Lee, Byeongchang KimProsody modeling is critical in developing text-to-speech (TTS) systems where speech synthesis is used to automatically generate natural speech. In this paper, we present a prosody generation architecture based on Chinese Tone and Break Index (C-ToBI) representation. ToBI is a multi-tier representation system based on linguistic knowledge to transcribe events in an utterance. The TTS system which adopts ToBI as an intermediate representation is known to exhibit higher flexibility, modularity and domain/task portability compared with the direct prosody generation TTS systems. However, the cost of corpus preparation is very expensive for practical-level performance because the ToBI labeled corpus has been manually constructed by many prosody experts and normally requires a large amount of data for accurate statistical prosody modeling. This paper proposes a new method which transcribes the C-ToBI labels automatically in Chinese speech. We model Chinese prosody generation as a classification problem and apply conditional Maximum Entropy (ME) classification to this problem. We empirically verify the usefulness of various natural language and phonology features to make well-integrated features for ME framework.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.4
• /
• pp.2162-2179
• /
• 2019

The performance of text classification is highly related to the feature selection methods. Usually, two tasks are performed when a feature selection method is applied to construct a feature set; 1) assign score to each feature and 2) select the top-N features. The selection of top-N features in the existing filter-based feature selection methods is biased by their discriminative power and the empirical process which is followed to determine the value of N. In order to improve the text classification performance by presenting a more illustrative feature set, we present an approach via a potent representation learning technique, namely DBN (Deep Belief Network). This algorithm learns via the semantic illustration of documents and uses feature vectors for their formulation. The nodes, iteration, and a number of hidden layers are the main parameters of DBN, which can tune to improve the classifier's performance. The results of experiments indicate the effectiveness of the proposed method to increase the classification performance and aid developers to make effective decisions in certain domains.

• Journal of Information Processing Systems
• /
• v.19 no.4
• /
• pp.427-438
• /
• 2023

Plenty of works have indicated that single image super-resolution (SISR) models relying on synthetic datasets are difficult to be applied to real scene text image super-resolution (STISR) for its more complex degradation. The up-to-date dataset for realistic STISR is called TextZoom, while the current methods trained on this dataset have not considered the effect of multi-scale features of text images. In this paper, a multi-scale and attention fusion model for realistic STISR is proposed. The multi-scale learning mechanism is introduced to acquire sophisticated feature representations of text images; The spatial and channel attentions are introduced to capture the local information and inter-channel interaction information of text images; At last, this paper designs a multi-scale residual attention module by skillfully fusing multi-scale learning and attention mechanisms. The experiments on TextZoom demonstrate that the model proposed increases scene text recognition's (ASTER) average recognition accuracy by 1.2% compared to text super-resolution network.

• Journal of KIISE:Software and Applications
• /
• v.31 no.4
• /
• pp.420-438
• /
• 2004

Artificially or naturally contained texts in the natural images have significant and detailed information about the scenes. If we develop a method that can extract and recognize those texts in real-time, the method can be applied to many important applications. In this paper, we suggest a new method that extracts the text areas in the natural images using the low-level image features of color continuity. gray-level variation and color valiance and that verifies the extracted candidate regions by using the high-level text feature such as stroke. And the two level features are combined hierarchically. The color continuity is used since most of the characters in the same text lesion have the same color, and the gray-level variation is used since the text strokes are distinctive in their gray-values to the background. Also, the color variance is used since the text strokes are distinctive in their gray-values to the background, and this value is more sensitive than the gray-level variations. The text level stroke features are extracted using a multi-resolution wavelet transforms on the local image areas and the feature vectors are input to a SVM(Support Vector Machine) classifier for the verification. We have tested the proposed method using various kinds of the natural images and have confirmed that the extraction rates are very high even in complex background images.

• Journal of information and communication convergence engineering
• /
• v.4 no.4
• /
• pp.174-179
• /
• 2006

In this paper we present a text detection method inspired by wavelet packet analysis and improved fuzzy clustering algorithm(IAFC).This approach assumes that the text and non-text regions are considered as two different texture regions. The text detection is achieved by using wavelet packet analysis as a feature analysis. The wavelet packet analysis is a method of wavelet decomposition that offers a richer range of possibilities for document image. From these multi scale features, we adapt the improved fuzzy clustering algorithm based on the unsupervised learning rule. The results show that our text detection method is effective for document images scanned from newspapers and journals.