Hierarchical Classification of Scientific Articles Using Deep Learning (Using the UDC Hierarchy As an Example)

The increasing digital information during the past few decades has radically altered the physical look of knowledge organisation of libraries and the information centres. Traditional classification systems such as the Dewey Decimal Classification (DDC), Universal Decimal Classification (UDC) and Library of Congress Classification (LCC) provided the organised structures upon which printed materials may be ordered in the past. Nevertheless, the rapid growth of electronic resources, including electronic books, online journals, institutional repositories, and multimedia resources has created major issues in the useful and scalable classification systems. The manual systems of classification that highly rely on human knowledge are no longer appropriate in handling volume, complexity and dynamism of the digital information. In these ways, Artificial Intelligence (AI) has become one of the new technological solutions of the new generation that enables automated classification of information resources. Machine learning, natural language processing, deep learning, and text mining are some of the AI technologies that assist with the analysis of written resources and automatically and efficiency assign writing subjects to particular categories and with a high level of reliability. The analysed paper provides a detailed analysis of the AI-based automatic classification in the Knowledge Organization Systems (KOS). It talks about the evolution of classical classification systems and introducing AI technologies, and the principles of automatic classification. The paper also measures the advantages, problems, and future of AI in library science, in terms of the evolving role of the librarians in the automated world. The research is qualitative in nature since it is founded on literature review, conceptual analysis and case-based observations. Keywords: Artificial Intelligence, Automatic Classification, Knowledge Organization Systems, Machine Learning, Natural Language Processing,

Scholars in International Relations Cite Books More Frequently than Journals: More Research is Needed to Better Understand Research Behaviour and Use

Hierarchical classification for acoustic scenes using deep learning

Summary Modern monothetic hierarchical soil classification systems such as the Chinese Soil Taxonomy (CST) are semi‐quantitative and their future development is likely to trend towards a fully quantitative system using the concepts of numerical taxonomy. Previous researchers have calculated the taxonomic distances between individual soils based on soil physiochemical properties, not, however, based on spectra of full soil profiles with different horizons. We hypothesized that numerical taxonomy implemented by cluster analysis of the taxonomic distance matrix based on vis–NIR spectra would accord with some CST Orders assigned by pedologists, and not with others, depending on how closely spectral features represent the diagnostic features used in the classification. Taxonomic distances in spectral space were computed for all pairs of 191 profiles, resulting in a distance matrix on which hierarchical cluster analysis was performed. Different indices were calculated to determine the optimum number of clusters, resulting in four spectral soil classes. These were then compared with CST Orders assigned to the profiles by expert allocation. The numerical classes and CST Orders matched poorly because of the completely different classification philosophies behind numerical taxonomy and the CST, which is based largely on presumed genesis and uses sharp thresholds leading to very similar soils being allocated to different classes. Thus, we consider the numerical classification as information that is complementary to the monothetic hierarchical system. Numerical classification can reveal the taxonomic objective aspects of the relation between the defined classes and can suggest new groupings. Highlights Taxonomic distances can serve as an objective measure of soil similarity. Soil spectra are a good candidate for numerical classification based on taxonomic distances. The conceptual basis of Chinese Soil Taxonomy (CST) does not match that of taxonomic distance. Numerical classification based on spectra can suggest revisions to the CST.

The rapidly evolving landscape of Aspect-Based Sentiment Analysis (ABSA) in the realm of natural language processing necessitates innovative approaches to comprehend and interpret the intricate nature of sentiments expressed in textual data. Traditional ABSA methods have often struggled with the nuanced sentiments inherent in various textual sources, limited in their ability to adapt to domain-specific vernacular and context. This study introduces a novel approach that synergizes rule mining with advanced deep learning techniques, aiming to address these limitations and enhance the precision and contextual understanding in sentiment analysis. Our proposed model integrates rule-based systems with deep learning Transformers, a method recognized for its effectiveness in extracting structured, domain-specific insights. This integration results in a significant enhancement in the model's ability to capture nuanced sentiments, as demonstrated by an 8.5% increase in aspect-based sentiment analysis precision and an 8.3% improvement in accuracy over existing methods. The model employs a combination of techniques including Bi-LSTM, BiGRU-CRF with RNNs, and ontology mapping operations, forming a Hybrid Multi-Level Architecture that effectively captures both sequential data relationships and semantic context. Additionally, the model incorporates Cross-Domain Transfer Learning, utilizing BERT-based pre-trained models with added layers for contextual semantics, which has shown notable generalization capabilities across various domains. Furthermore, this study refines evaluation parameters, tailoring metrics such as domain-specific accuracy, recall, and precision to more accurately assess model performance in particular domains. This is especially pertinent in our analysis of Twitter datasets specific to industries like finance and healthcare. The inclusion of Temporal Dynamics and Aspect Summarization, using techniques like the VARMAx process, provides insights into the evolution of sentiments and aspects over time, an aspect crucial for longitudinal sentiment analysis. The comprehensive testing of our model on diverse spatial and temporal datasets reveals not only improved precision and accuracy in sentiment analysis but also a reduction in delay, demonstrating its efficiency and responsiveness. The study's findings indicate that our approach could significantly impact fields reliant on sentiment analysis, such as market analysis, public opinion research, and social media monitoring, providing a more nuanced and accurate understanding of consumer sentiments and trends. This paper's contribution lies in its innovative amalgamation of rule-based and deep learning techniques, tailored evaluation metrics, and its emphasis on temporal dynamics, setting a new precedent in the field of ABSA and opening avenues for further research and application in real-world scenarios.

In 2015 the United Nations proposed the Sustainable Development Goals (SDGs), a set of universal goals for meeting the urgent environmental, political and economic challenges in the world. Universities play an important role to support and contribute to the SDGs mainly through education and research. To evaluate the contributions through research, universities aim at relating their scientific publications to SDGs, and automatically quantify the connectedness of these publications to the detailed targets and the unique indicators under SDGs. In this paper, we apply deep learning techniques to estimate the unknown indicators (third level) and targets (second level) for each publication, and output all its possible goals (first level). Specifically, we first exploit the dependency of categories at different levels (goals, targets, and indicators) to extract the dependent label features. Then we calculate the degree of matching between categories and publications in a bottom-up way and design a hierarchical structure to transfer such matching information level by level until obtaining the predicted SDGs of the publications. This is the first application of a deep learning method on this SDG prediction task and our experiments clearly demonstrate the good performance of our model on this real-world SDGs matching task, the extraction of key information as well as the prediction of potential sub-categories. As auxiliary analysis, we visualize the extraction of key semantic information and the probability of the hierarchical SDG categories.

Anti-virus software based on unsupervised hierarchical clustering (HC) of malware samples has been shown to be vulnerable to poisoning attacks. In this kind of attack, a malicious player degrades anti-virus performance by submitting to the database samples specifically designed to collapse the classification hierarchy utilized by the anti-virus (and constructed through HC) or otherwise deform it in a way that would render it useless. Though each poisoning attack needs to be tailored to the particular HC scheme deployed, existing research seems to indicate that no particular HC method by itself is immune. We present results on applying a new notion of entropy for combinatorial dendrograms to the problem of controlling the influx of samples into the data base and deflecting poisoning attacks. In a nutshell, effective and tractable measures of change in hierarchy complexity are derived from the above, enabling on-the-fly flagging and rejection of potentially damaging samples. The information-theoretic underpinnings of these measures ensure their indifference to which particular poisoning algorithm is being used by the attacker, rendering them particularly attractive in this setting.

Some aspects of past development are outlined in regard to the identification and classification of nature’s benefits nowadays conceptualized as ecosystem services. The knowledge about the multitude and diversity of the useful benefits has grown tremendously and yet a large part of biodiversity is still unknown. The conclusion is that the variety and complexity of ecosystem goods and services can only be properly categorized and managed by using hierarchical classifications. The Common International Classification of Ecosystem Services (CICES) represents the most concentrated effort to develop hierarchic, systematic and multipurpose classification for ecosystem services. CICES has been applied for the classification of the services of major inland ecosystems of Finland: forests, agricultural fields, peatland and freshwaters. The major observations has been that the flexibility, which the hierarchic system provides for moving towards more detailed classification levels is really a needed advantage and was used in the classifications done. Sample examples from the expanded classification are given besides other observations. The results of the classification efforts, which are reported elsewhere, are first applications of CICES in the boreal ecosystems.

Data sets on wetlands required for the representation of aquatic ecosystem biodiversity and systematic wetland conservation planning are typically not available or are inadequate, particularly at country‐wide scale, which hinders conservation planning. The improvement in hierarchical classification systems and increased availability of broad‐scale data sets offers new opportunities to overcome these limitations. This study demonstrates replicable methods for classifying wetland ecosystem types and condition country‐wide using broad‐scale data sets in data‐scarce countries. A country‐wide data set, compiled primarily using remote sensing techniques, was combined with regional and landscape‐setting data sets to reflect the ecological and geomorphic biodiversity of wetlands. Geographical Information Systems (GIS) were employed to model wetland types, disturbance indices and identify priority wetlands through threatened faunal species associations using existing data. Accuracy of the national data was assessed through a congruency with two local data sets. Most of the 1 680 306 ha of inland wetlands were classified as Natural (80%), of which the majority were located on Valley Floors (68%). However, the national data were found only to represent 54% of wetlands mapped at a local scale, and comparison with local data showed inaccuracies in the types and condition classifications. Problems regarding spatial data quality and scale are discussed and suggestions for improvement are provided. The desktop classification steps can be reproduced easily for other data‐scarce countries. Data sets on freshwater ecosystems can assist in raising awareness and influence policy at a national scale. Copyright © 2015 John Wiley & Sons, Ltd.

The number of notational symbols and the length of codes in hierarchical classification systems

Background Machine learning (ML) algorithm support vector machine (SVM) performed better than Thrombolysis in Myocardial Infarction (TIMI) score for ASIAN STEMI patients. However, Deep Learning (DL) effectiveness in the multiethnic ASIAN population has yet to be determined. DL has automatic learning of the feature from a given dataset without the need to conduct feature selection. However, the selected features by the algorithm is black box. Identifying features associated with mortality is essential to recognize characteristics of patients with high risk for better patient management. Purpose To develop a DL algorithm for in-hospital mortality in multiethnic STEMI patients using predictors identified from the SVM algorithm. To investigate DL performance constructed using predictors from SVM feature extraction and expert-recommended predictors. Methods We constructed four algorithms; a) DL and SVM algorithms with predictors identified from the SVM variable importance b) DL and SVM using predictors based on expert recommendation. We used registry data from the National Cardiovascular Disease Database of 11397 patient's. Fifty parameters including demographics, cardiovascular risk, medications and clinical variables were considered. The Area under the curve (AUC) is the performance evaluation metric. Algorithms were validated against the TIMI and tested using the same validation data. SVM variable importance with backward elimination was used to select and rank important variables. Results DL algorithms outperform SVM and TIMI on the validation dataset; i) DL with SVM selected predictors (15 predictors, AUC = 0.97), ii) DL with expert-recommended predictors (16 predictors, AUC = 0.96), iii) SVM with selected predictors (15 predictors, AUC = 0.92), iv) SVM with expert-recommended predictors (AUC = 0.89) and TIMI (AUC = 0.82). Common predictors across SVM feature selection, expert-recommendation and TIMI are: age, heart rate, Killip class, fasting blood glucose, systolic blood pressure, comorbid diseases and ST-elevation. SVM feature selection also identified diuretics, PCI and pharmacotherapy drugs as predictors that improve mortality prediction in STEMI patients. Our findings suggest that the TIMI score underestimates patients risk of mortality. DL algorithm using selected predictors classified 35% of nonsurvival patients as high risk (risk probabilities >50%) compared to only 12.7% nonsurvival patients by TIMI (score >5) (Figure below). Conclusions In the ASIAN population, patients with STEMI can be better classified using the DL algorithm compared to the ML and TIMI score. Combining ML feature selection with DL allows the identification of distinct factors in a unique ASIAN population for better mortality prediction than relying solely on an expert recommendation as it is a very subjective approach. Continuous validation on population-specific algorithms using DL and ML is needed before implementing in a real clinical setting. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1 TIMI performance on validation setDL performance on validation set

Recent health reforms have created incentives for cardiologists and accountable care organizations to participate in value-based care models for heart failure (HF). Accurate risk stratification of HF patients is critical to efficiently deploy interventions aimed at reducing preventable utilization. The goal of this paper was to compare deep learning approaches with traditional logistic regression (LR) to predict preventable utilization among HF patients. We conducted a prognostic study using data on 93,260 HF patients continuously enrolled for 2-years in a large U.S. commercial insurer to develop and validate prediction models for three outcomes of interest: preventable hospitalizations, preventable emergency department (ED) visits, and preventable costs. Patients were split into training, validation, and testing samples. Outcomes were modeled using traditional and enhanced LR and compared to gradient boosting model and deep learning models using sequential and non-sequential inputs. Evaluation metrics included precision (positive predictive value) at k, cost capture, and Area Under the Receiver operating characteristic (AUROC). Deep learning models consistently outperformed LR for all three outcomes with respect to the chosen evaluation metrics. Precision at 1% for preventable hospitalizations was 43% for deep learning compared to 30% for enhanced LR. Precision at 1% for preventable ED visits was 39% for deep learning compared to 33% for enhanced LR. For preventable cost, cost capture at 1% was 30% for sequential deep learning, compared to 18% for enhanced LR. The highest AUROCs for deep learning were 0.778, 0.681 and 0.727, respectively. These results offer a promising approach to identify patients for targeted interventions. Funding Acknowledgement Type of funding sources: Private company. Main funding source(s): internally funded by Diagnostic Robotics Inc.

Diagnosis of pulmonary lesions from computed tomography (CT) is important but challenging for clinical decision making in lung cancer related diseases. Deep learning has achieved great success in computer aided diagnosis (CADx) area for lung cancer, whereas it suffers from label ambiguity due to the difficulty in the radiological diagnosis. Considering that invasive pathological analysis serves as the clinical golden standard of lung cancer diagnosis, in this study, we solve the label ambiguity issue via a large-scale radio-pathomics dataset containing 5,134 radiological CT images with pathologically confirmed labels, including cancers (e.g., invasive/non-invasive adenocarcinoma, squamous carcinoma) and non-cancer diseases (e.g., tuberculosis, hamartoma). This retrospective dataset, named Pulmonary-RadPath, enables development and validation of accurate deep learning systems to predict invasive pathological labels with a non-invasive procedure, i.e., radiological CT scans. A three-level hierarchical classification system for pulmonary lesions is developed, which covers most diseases in cancer-related diagnosis. We explore several techniques for hierarchical classification on this dataset, and propose a Leaky Dense Hierarchy approach with proven effectiveness in experiments. Our study significantly outperforms prior arts in terms of data scales (6x larger), disease comprehensiveness and hierarchies. The promising results suggest the potentials to facilitate precision medicine.

Air quality, both indoors and outdoors, is crucial for public health as it affects respiratory conditions and overall well-being. Machine learning (ML) techniques offer innovative solutions for monitoring and predicting air quality. However, choosing the right ML algorithms and evaluation metrics is essential for creating accurate air quality prediction models, as these choices directly impact the accuracy and reliability of the results. This study aims to explore and summarize the literature on the use of ML techniques in predicting and optimizing air quality, addressing the urgent issue of air pollution. The review analyzed papers from two electronic databases, namely Scopus and Science Direct. Information on ML techniques for predicting air quality and identifying the main sources of pollution was extracted from 26 studies. The study focuses on common ML techniques employed in air quality prediction, including classification, deep learning, regression, ensemble learning, and combinations of regression and deep learning. It also identifies the evaluation metrics used to assess the performance of these models, such as recall, root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). By synthesizing existing knowledge, this study provides a comprehensive understanding of the metrics researchers use to measure the overall effectiveness of ML algorithms. It serves as a benchmark for future research and guides the selection of appropriate evaluation metrics in the field of air quality prediction.

Recently biometric systems have shown improved capabilities because of the remarkable success of deep learning in solving various computer vision tasks. In ear recognition, the use of deep learning techniques is seldom due to training data scarcity. The existing work has shown poor performance as the majority of techniques are based on either handcraft features or pre-trained models. Besides this, transfer-learning has also shown poor performance because of the diversity among the tasks. To circumvent the existing issues, in this work, we have presented an end-to-end framework for ear recognition. It consist of the Ear Mask Extraction (EME) network to segment the ear, a normalization algorithm to align the ear, and a novel siamese-based CNN (CG-ERNet) for deep ear feature learning. CG-ERNet exploits domain-specific knowledge by using Curvature Gabor filters and uses triplet loss, triplet selection, and adaptive margin for better convergence of the loss. For comparative analysis, we trained state-of-the-art deep learning models like Face-Net, VGG19, ResNet50, Inception, Exception, and Mobile-Net for ear-recognition. The performance is assessed using five well-known evaluation metrics. In the extensive experimentation, our proposed model (CG-ERNet) outperformed the deep learning models and handcrafted feature based methods on four different, publicly available, benchmark datasets. To make the results more interpretable, we employ the t-SNE visualization of learned features. Additionally, our proposed method has shown robustness to various environmental challenges like Gaussian noise, Gaussian blur, up to ± 30 degrees of rotation, and 20% of occlusion.