• Asia pacific journal of information systems
• /
• 제29권4호
• /
• pp.771-788
• /
• 2019

Deep learning techniques such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) show superior performance in text classification than traditional approaches such as Support Vector Machines (SVMs) and Naïve Bayesian approaches. When using CNNs for text classification tasks, word embedding or character embedding is a step to transform words or characters to fixed size vectors before feeding them into convolutional layers. In this paper, we propose a parallel word-level and character-level embedding approach in CNNs for text classification. The proposed approach can capture word-level and character-level patterns concurrently in CNNs. To show the usefulness of proposed approach, we perform experiments with two English and three Korean text datasets. The experimental results show that character-level embedding works better in Korean and word-level embedding performs well in English. Also the experimental results reveal that the proposed approach provides better performance than traditional CNNs with word-level embedding or character-level embedding in both Korean and English documents. From more detail investigation, we find that the proposed approach tends to perform better when there is relatively small amount of data comparing to the traditional embedding approaches.

• 한국멀티미디어학회논문지
• /
• 제23권9호
• /
• pp.1181-1190
• /
• 2020

Detection of an anomaly pattern deviating normal data distribution in streaming data is an important technique in many application areas. In this paper, a method for detection of an newly emerging pattern in text streaming data which is an ordered sequence of texts is proposed based on text embedding and anomaly pattern detection. Using text embedding methods such as BOW(Bag Of Words), Word2Vec, and BERT, the detection performance of the proposed method is compared. Experimental results show that anomaly pattern detection using BERT embedding gave an average F1 value of 0.85 and the F1 value of 1 in three cases among five test cases.

• 지식경영연구
• /
• 제21권1호
• /
• pp.99-116
• /
• 2020

Text is the most widely used means of exchanging or expressing knowledge and information in the real world. Recently, researches on structuring unstructured text data for text analysis have been actively performed. One of the most representative document embedding method (i.e. doc2Vec) generates a single vector for each document using the whole corpus included in the document. This causes a limitation that the document vector is affected by not only core words but also other miscellaneous words. Additionally, the traditional document embedding algorithms map each document into only one vector. Therefore, it is not easy to represent a complex document with interdisciplinary subjects into a single vector properly by the traditional approach. In this paper, we introduce a multi-vector document embedding method to overcome these limitations of the traditional document embedding methods. After introducing the previous study on multi-vector document embedding, we visually analyze the effects of the multi-vector document embedding method. Firstly, the new method vectorizes the document using only predefined keywords instead of the entire words. Secondly, the new method decomposes various subjects included in the document and generates multiple vectors for each document. The experiments for about three thousands of academic papers revealed that the single vector-based traditional approach cannot properly map complex documents because of interference among subjects in each vector. With the multi-vector based method, we ascertained that the information and knowledge in complex documents can be represented more accurately by eliminating the interference among subjects.

• 지능정보연구
• /
• 제25권3호
• /
• pp.19-41
• /
• 2019

텍스트 데이터에 대한 다양한 분석을 위해 최근 비정형 텍스트 데이터를 구조화하는 방안에 대한 연구가 활발하게 이루어지고 있다. doc2Vec으로 대표되는 기존 문서 임베딩 방법은 문서가 포함한 모든 단어를 사용하여 벡터를 만들기 때문에, 문서 벡터가 핵심 단어뿐 아니라 주변 단어의 영향도 함께 받는다는 한계가 있다. 또한 기존 문서 임베딩 방법은 하나의 문서가 하나의 벡터로 표현되기 때문에, 다양한 주제를 복합적으로 갖는 복합 문서를 정확하게 사상하기 어렵다는 한계를 갖는다. 본 논문에서는 기존의 문서 임베딩이 갖는 이러한 두 가지 한계를 극복하기 위해 다중 벡터 문서 임베딩 방법론을 새롭게 제안한다. 구체적으로 제안 방법론은 전체 단어가 아닌 핵심 단어만 이용하여 문서를 벡터화하고, 문서가 포함하는 다양한 주제를 분해하여 하나의 문서를 여러 벡터의 집합으로 표현한다. KISS에서 수집한 총 3,147개의 논문에 대한 실험을 통해 복합 문서를 단일 벡터로 표현하는 경우의 벡터 왜곡 현상을 확인하였으며, 복합 문서를 의미적으로 분해하여 다중 벡터로 나타내는 제안 방법론에 의해 이러한 왜곡 현상을 보정하고 각 문서를 더욱 정확하게 임베딩할 수 있음을 확인하였다.

• 스마트미디어저널
• /
• 제7권4호
• /
• pp.24-29
• /
• 2018

딥러닝에서 자연어 처리를 위한 텍스트 분석 기법은 워드 임베딩을 통해 단어를 벡터 형태로 표현한다. 본 논문에서는 워드 임베딩 기법과 딥러닝 기법을 이용하여 SMS 문자 메시지를 문서 벡터로 구성하고 이를 스팸 문자 메시지와 정상적인 문자 메시지로 분류하는 방법을 제안하였다. 유사한 문맥을 가진 단어들은 벡터 공간에서 인접한 벡터 공간에 표현되도록 하기 위해 전처리 과정으로 자동 띄어쓰기를 적용하고 스팸 문자 메시지로 차단되는 것을 피하기 위한 목적으로 음절의 자모를 특수기호로 왜곡하여 맞춤법이 파괴된 상태로 단어 벡터와 문장 벡터를 생성하였다. 또한 문장 벡터 생성 시 CBOW와 skip gram이라는 두 가지 워드 임베딩 알고리즘을 적용하여 문장 벡터를 표현하였으며, 딥러닝을 이용한 스팸 문자 메시지 필터링의 성능 평가를 위해 SVM Light와 정확도를 비교 측정하였다.

• 대한임베디드공학회논문지
• /
• 제17권6호
• /
• pp.347-353
• /
• 2022

In this paper, we propose a method that performs a text-video retrieval model by replacing video properties using captions. In general, the exisiting embedding-based models consist of both joint embedding space construction and the CNN-based video encoding process, which requires a lot of computation in the training as well as the inference process. To overcome this problem, we introduce a video-captioning module to replace the visual property of video with captions generated by the video-captioning module. To be specific, we adopt the caption generator that converts candidate videos into captions in the inference process, thereby enabling direct comparison between the text given as a query and candidate videos without joint embedding space. Through the experiment, the proposed model successfully reduces the amount of computation and inference time by skipping the visual processing process and joint embedding space construction on two benchmark dataset, MSR-VTT and VATEX.

• International Journal of Computer Science & Network Security
• /
• 제22권8호
• /
• pp.399-403
• /
• 2022

Natural texts are analyzed to obtain their intended meaning to be classified depending on the problem under study. One way to represent words is by generating vectors of real values to encode the meaning; this is called word embedding. Similarities between word representations are measured to identify text class. Word embeddings can be created using word2vec technique. However, recently fastText was implemented to provide better results when it is used with classifiers. In this paper, we will study the performance of well-known classifiers when using both techniques for word embedding with Arabic dataset. We applied them to real data collected from Wikipedia, and we found that both word2vec and fastText had similar accuracy with all used classifiers.

• 한국음향학회지
• /
• 제40권5호
• /
• pp.496-502
• /
• 2021

딥러닝 기반의 심층 화자 임베딩 방식은 최근 문장 독립 화자 검증 연구에 널리 사용되고 있으며, 기존의 i-vector 방식에 비해 더 좋은 성능을 보이고 있다. 본 연구에서는 심층 화자 임베딩 방식을 발전시키기 위하여, 화자의 그룹 정보를 도입한 그룹기반 화자 임베딩을 제안한다. 훈련 데이터 내에 존재하는 전체 화자들을 정해진 개수의 그룹으로 비지도 클러스터링 하며, 고정된 길이의 그룹 임베딩 벡터가 각각의 그룹을 대표한다. 그룹 결정 네트워크가 각 그룹에 대응되는 그룹 가중치를 출력하며, 이를 이용한 그룹 임베딩 벡터들의 가중 합을 통해 집합 그룹 임베딩을 추출한다. 최종적으로 집합 그룹 임베딩을 심층 화자 임베딩에 더해주어 그룹기반 화자 임베딩을 생성한다. 이러한 방식을 통해 그룹 정보를 심층 화자 임베딩에 도입함으로써, 화자 임베딩이 나타낼 수 있는 전체 화자의 검색 공간을 줄일 수 있고, 이를 통해 화자 임베딩은 많은 수의 화자를 유연하게 표현할 수 있다. VoxCeleb1 데이터베이스를 이용하여 본 연구에서 제안하는 방식이 기존의 방식을 개선시킨다는 것을 확인하였다.

• ETRI Journal
• /
• 제46권1호
• /
• pp.35-47
• /
• 2024

This study introduces CR-M-SpanBERT, a coreference resolution (CR) model that utilizes multiple embedding-based span bidirectional encoder representations from transformers, for antecedent recognition in natural language (NL) text. Information extraction studies aimed to extract knowledge from NL text autonomously and cost-effectively. However, the extracted information may not represent knowledge accurately owing to the presence of ambiguous entities. Therefore, we propose a CR model that identifies mentions referring to the same entity in NL text. In the case of CR, it is necessary to understand both the syntax and semantics of the NL text simultaneously. Therefore, multiple embeddings are generated for CR, which can include syntactic and semantic information for each word. We evaluate the effectiveness of CR-M-SpanBERT by comparing it to a model that uses SpanBERT as the language model in CR studies. The results demonstrate that our proposed deep neural network model achieves high-recognition accuracy for extracting antecedents from NL text. Additionally, it requires fewer epochs to achieve an average F1 accuracy greater than 75% compared with the conventional SpanBERT approach.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권1호
• /
• pp.170-191
• /
• 2024

The transmission of confidential data using cover media is called steganography. The three requirements of any effective steganography system are high embedding capacity, security, and imperceptibility. The text file's structure, which makes syntax and grammar more visually obvious than in other media, contributes to its poor imperceptibility. Text steganography is regarded as the most challenging carrier to hide secret data because of its insufficient redundant data compared to other digital objects. Unicode characters, especially non-printing or invisible, are employed for hiding data by mapping a specific amount of secret data bits in each character and inserting the character into cover text spaces. These characters are known with limited spaces to embed secret data. Current studies that used Unicode characters in text steganography focused on increasing the data hiding capacity with insufficient redundant data in a text file. A sequential embedding pattern is often selected and included in all available positions in the cover text. This embedding pattern negatively affects the text steganography system's imperceptibility and security. Thus, this study attempts to solve these limitations using the Part-of-speech (POS) tagging technique combined with the randomization concept in data hiding. Combining these two techniques allows inserting the Unicode characters in randomized patterns with specific positions in the cover text to increase data hiding capacity with minimum effects on imperceptibility and security. Format-preserving encryption (FPE) is also used to encrypt a secret message without changing its size before the embedding processes. By comparing the proposed technique to already existing ones, the results demonstrate that it fulfils the cover file's capacity, imperceptibility, and security requirements.