• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.3
• /
• pp.740-753
• /
• 2023

Text recognition in natural scene images is a challenging problem in computer vision. The accurate identification of ship number characters can effectively improve the level of ship traffic management. However, due to the blurring caused by motion and text occlusion, the accuracy of ship number recognition is difficult to meet the actual requirements. To solve these problems, this paper proposes a dual-branch network based on the CRNN identification network. The network couples image restoration and character recognition. The CycleGAN module is used for blur restoration branch, and the Pix2pix module is used for character occlusion branch. The two are coupled to reduce the impact of image blur and occlusion. Input the recovered image into the text recognition branch to improve the recognition accuracy. After a lot of experiments, the model is robust and easy to train. Experiments on CTW datasets and real ship maps illustrate that our method can get more accurate results.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.15 no.3
• /
• pp.153-158
• /
• 2015

Nowadays, automatic keyphrase extraction is considered to be an important task. Most of the previous studies focused only on selecting keyphrases within the body of input documents. These studies overlooked latent keyphrases that did not appear in documents. In addition, a small number of studies on latent keyphrase extraction methods had some structural limitations. Although latent keyphrases do not appear in documents, they can still undertake an important role in text mining because they link meaningful concepts or contents of documents and can be utilized in short articles such as social network service, which rarely have explicit keyphrases. In this paper, we propose a new approach that selects qualified latent keyphrases from input documents and overcomes some structural limitations by using deep belief networks in a supervised manner. The main idea of this approach is to capture the intrinsic representations of documents and extract eligible latent keyphrases by using them. Our experimental results showed that latent keyphrases were successfully extracted using our proposed method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.5 no.3
• /
• pp.560-574
• /
• 2011

In this paper, we propose a hybrid method of Machine Learning (ML) algorithm and a rule-based algorithm to implement a lightweight Named Entity (NE) extraction system for Korean SMS text. NE extraction from Korean SMS text is a challenging theme due to the resource limitation on a mobile phone, corruptions in input text, need for extension to include personal information stored in a mobile phone, and sparsity of training data. The proposed hybrid method retaining the advantages of statistical ML and rule-based algorithms provides fully-automated procedures for the combination of ML approaches and their correction rules using a threshold-based soft decision function. The proposed method is applied to Korean SMS texts to extract person's names as well as location names which are key information in personal appointment management system. Our proposed system achieved 80.53% in F-measure in this domain, superior to those of the conventional ML approaches.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.4
• /
• pp.1409-1425
• /
• 2021

In wide-ranging information society, fast and easy access to information in language of one's choice is indispensable, which may be provided by using various multilingual Natural Language Processing (NLP) applications. Natural language text contains references among different language elements, called anaphoric links. Resolving anaphoric links is a key problem in NLP. Anaphora resolution is an essential part of NLP applications. Anaphoric links need to be properly interpreted for clear understanding of natural languages. For this purpose, a mechanism is desirable for the identification and resolution of these naturally occurring anaphoric links. In this paper, a framework based on Hobbs syntactic approach and a system developed by Lappin & Leass is proposed for resolution of reflexive anaphoric links, present in Urdu text documents. Generally, anaphora resolution process takes three main steps: identification of the anaphor, location of the candidate antecedent(s) and selection of the appropriate antecedent. The proposed framework is based on exploring the syntactic structure of reflexive anaphors to find out various features for constructing heuristic rules to develop an algorithm for resolving these anaphoric references. System takes Urdu text containing reflexive anaphors as input, and outputs Urdu text with resolved reflexive anaphoric links. Despite having scarcity of Urdu resources, our results are encouraging. The proposed framework can be utilized in multilingual NLP (m-NLP) applications.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.8
• /
• pp.2278-2291
• /
• 2023

Imported ornamental fish should be quarantined because they can have dangerous diseases depending on their habitat. The quarantine requires a lot of time because quarantine officers collect various information on the imported ornamental fish. Inefficient quarantine processes reduce its work efficiency and accuracy. Also, long-time quarantine causes the death of environmentally sensitive ornamental fish and huge financial losses. To improve existing quarantine systems, information on ornamental fish was collected and structured, and a server was established to develop quarantine performance support software equipped with a text search engine. However, the long names of ornamental fish in general can cause many typos and time bottlenecks when we type search words for the target fish information. Therefore, we need a technique that can correct typos. Typical typo character calibration compares input text with all characters in a calibrated candidate text dictionary. However, this approach requires computational power proportional to the number of typos, resulting in slow processing time and low calibration accuracy performance. Therefore, to improve the calibration accuracy of characters, we propose a fusion system of simple Artificial Neural Network (ANN) models and character preprocessing methods that accelerate the process by minimizing the computation of the models. We also propose a typo character generation method used for training the ANN models. Simulation results show that the proposed typo character correction system is about 6 times faster than the conventional method and has 10% higher accuracy.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.8
• /
• pp.3473-3487
• /
• 2020

In this paper, we present a new approach to creating speech animation with emotional expressions using a small set of example models. To generate realistic facial animation, two example models called key visemes and expressions are used for lip-synchronization and facial expressions, respectively. The key visemes represent lip shapes of phonemes such as vowels and consonants while the key expressions represent basic emotions of a face. Our approach utilizes a text-to-speech (TTS) system to create a phonetic transcript for the speech animation. Based on a phonetic transcript, a sequence of speech animation is synthesized by interpolating the corresponding sequence of key visemes. Using an input parameter vector, the key expressions are blended by a method of scattered data interpolation. During the synthesizing process, an importance-based scheme is introduced to combine both lip-synchronization and facial expressions into one animation sequence in real time (over 120Hz). The proposed approach can be applied to diverse types of digital content and applications that use facial animation with high accuracy (over 90%) in speech recognition.

• 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.

• Journal of Information Processing Systems
• /
• v.19 no.6
• /
• pp.830-841
• /
• 2023

Tokenization is the process of segmenting the input text into smaller units of text, and it is a preprocessing task that is mainly performed to improve the efficiency of the machine learning process. Various tokenization methods have been proposed for application in the field of natural language processing, but studies have primarily focused on efficiently segmenting text. Few studies have been conducted on the Korean language to explore what tokenization methods are suitable for document classification task. In this paper, an exploratory study was performed to find the most suitable tokenization method to improve the performance of a representative spatio-temporal document classifier in Korean. For the experiment, a convolutional neural network model was used, and for the final performance comparison, tasks were selected for document classification where performance largely depends on the tokenization method. As a tokenization method for comparative experiments, commonly used Jamo, Character, and Word units were adopted. As a result of the experiment, it was confirmed that the tokenization of word units showed excellent performance in the case of representative spatio-temporal document classification task where the semantic embedding ability of the token itself is important.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.11
• /
• pp.3991-4010
• /
• 2021

The Question Similarity Measurement of Chinese Crop Diseases and Insect Pests (QSM-CCD&IP) aims to judge the user's tendency to ask questions regarding input problems. The measurement is the basis of the Agricultural Knowledge Question and Answering (Q & A) system, information retrieval, and other tasks. However, the corpus and measurement methods available in this field have some deficiencies. In addition, error propagation may occur when the word boundary features and local context information are ignored when the general method embeds sentences. Hence, these factors make the task challenging. To solve the above problems and tackle the Question Similarity Measurement task in this work, a corpus on Chinese crop diseases and insect pests(CCDIP), which contains 13 categories, was established. Then, taking the CCDIP as the research object, this study proposes a Chinese agricultural text similarity matching model, namely, the AgrCQS. This model is based on mixed information extraction. Specifically, the hybrid embedding layer can enrich character information and improve the recognition ability of the model on the word boundary. The multi-scale local information can be extracted by multi-core convolutional neural network based on multi-weight (MM-CNN). The self-attention mechanism can enhance the fusion ability of the model on global information. In this research, the performance of the AgrCQS on the CCDIP is verified, and three benchmark datasets, namely, AFQMC, LCQMC, and BQ, are used. The accuracy rates are 93.92%, 74.42%, 86.35%, and 83.05%, respectively, which are higher than that of baseline systems without using any external knowledge. Additionally, the proposed method module can be extracted separately and applied to other models, thus providing reference for related research.

• Journal of Korea Society of Industrial Information Systems
• /
• v.28 no.5
• /
• pp.15-30
• /
• 2023

Recently, generative models based on the Transformer architecture, such as ChatGPT, have been gaining significant attention. The Transformer architecture has been applied to various neural network models, including Google's BERT(Bidirectional Encoder Representations from Transformers) sentence generation model. In this paper, a method is proposed to create a text binary classification model for determining whether a comment on Korean movie review is positive or negative. To accomplish this, a pre-trained multilingual BERT sentence generation model is fine-tuned and transfer learned using a new Korean training dataset. To achieve this, a pre-trained BERT-Base model for multilingual sentence generation with 104 languages, 12 layers, 768 hidden, 12 attention heads, and 110M parameters is used. To change the pre-trained BERT-Base model into a text classification model, the input and output layers were fine-tuned, resulting in the creation of a new model with 178 million parameters. Using the fine-tuned model, with a maximum word count of 128, a batch size of 16, and 5 epochs, transfer learning is conducted with 10,000 training data and 5,000 testing data. A text sentiment binary classification model for Korean movie review with an accuracy of 0.9582, a loss of 0.1177, and an F1 score of 0.81 has been created. As a result of performing transfer learning with a dataset five times larger, a model with an accuracy of 0.9562, a loss of 0.1202, and an F1 score of 0.86 has been generated.