In this paper, we first introduce a new text representation method to convert a textual document into a tensor space model named TextCuboid, which can preserve various meanings of polysemy. Based upon the new model, we propose two novel data augmentation techniques (called Boolean augmentation and CuboidGAN) that can be directly applied to the TextCuboid model for text classification tasks. Boolean augmentation includes three simple keyword modifications: synonym replacement, synonym insertion, and random deletion. CuboidGAN is composed of two key components, style encoding, and residual regression, and it is trained in two phases to generate unambiguous and plausible concept vectors. Through intensive experiments using five commonly used datasets, we prove that our proposed methods perform better data augmentation than other conventional methods. We also show that each augmentation method component significantly contributes to text classification through ablation studies.

This paper aims to automatically augment numerical tabular data by using the variational autoencoder model. For this, we try to solve the problem of class imbalance in numerical data and to improve the performance of the classification model by augmenting the training data. In this paper, we propose a new augmentation technique called ‘D-VAE’ which performs data augmentation through variational autoencoder with discretization for numerical columuns; D-VAE artificially increases the number of records and the number of columns for a given tabular data. The main features of the proposed technique are to kperform discretization and feature selection in the preprocessing process. For the discretization process, we use k-means algorithm, through which records within a given table are grouped, and then converted into one-hot vectors according to the clustering results. In addition, for memory efficiency, we reduced the number of parameters of the VAE model by using a relatively small number of features through feature selection called REFCV. To evaluate the performance of the proposed technique, we conducted various experiments by numerical data augmentation ratio using four open datasets.

Intraoperative hypotension (IOH) negatively affects the prognosis after surgery. Therefore, in recent years, various studies for IOH prediction based on bio-signal data have been carried out. This paper aims to develop an overfitting-resistant prediction model to forecast 5-minute prior to IOH by domain knowledge-based loss stratification and permutation method. In general, when developing machine learning-based prediction models, we experience the overfitting problem. In our paper, we tried to overcome the overfitting problem by using biomedical domain knowledge. As an example of the domain knowledge, we adopt American Society of Anesthesiology (ASA) status; ASA at higher levels indicates the higher possibility of IOH. To obtain the ASA status for developing the IOH prediction model, we used the electronic medical records from a public database VitalDB. Our proposed deep learning model accommodates the loss stratification and the ASA status permutation to consider the domain knowledge. We have found that the model has shown superior IOH prediction performance according to ASA status; this is particularly because it reduces the dependence of ASA status in the learning process.

Keyword-based search exploits the exact match between the index terms of a query and documents. Thus, some documents, although they are relevant to the given query, may not be returned to users unless the documents include the index terms of the query. Some search engines use the authority of documents, which is derived from the links of documents, to help keyword-based search provide more accurate search results. However, unlike the Web documents, if the links between documents do not exist, it is difficult to exploit the authority for ranking documents. In this paper, our goals are to derive the implicit authority of documents that do not have explicit links through semantic collaborative filtering (SCF), and to retrieve documents that are semantically related to the given query. To achieve these goals, we represent users' preferences, queries and documents with their corresponding concepts by extending a Bayesian belief network. It is because the Bayesian belief network provides a clear formalism for mapping the users' preferences, queries and documents to their corresponding concepts. The concepts are extracted from a taxonomic knowledgebase such as the Open Directory Project Web directory. In our experiment, we have shown that the extended Bayesian belief network using taxonomic knowledge outperforms the conventional approaches for personalized search.

This paper discusses a new type of semi-supervised docu-ment clustering that uses partial supervision to partition a large set of documents. Most clustering methods organizes documents into groups based only on similarity measures. Unfortunately, the traditional approaches to document clus-tering are often unable to correctly discern structural details hidden within the document corpus because their algorithms inherently strongly depend on the document themselves and their similarity to each other. In this paper, we attempt to isolate more semantically coherent clusters by employing the domain-specific knowledge provided by a document analyst. By using external human knowledge to guide the clustering mechanism with some flexibility when creating the clusters, clustering efficiency can be considerably enhanced. As a ba-sic clustering strategy, we use a variant of complete-linkage agglomerative hierarchical clustering, and develop the con-cepts (or seeds) of requested clusters by exploiting user-relevance feedback. Although the proposed method is slow when applied to large document collection, it yields higher quality clusters. Through experiments using the Reuters-21578 corpus, we show that the proposed method outper-forms unsupervised clustering method.