Search

Word embeddings for application in geosciences: development, evaluation and examples of soil-related concepts

Lists of semantically related words are better recalled on immediate memory tests than otherwise equivalent lists of unrelated words. However, measuring the degree of relatedness is not straightforward. We report three experiments that assess the ability of various measures of semantic relatedness—including latent semantic analysis (LSA), GloVe, fastText, and a number of measures based on WordNet—to predict whether two lists of words will be differentially recalled. In Experiment 1, all measures except LSA correctly predicted the observed better recall of the related than the unrelated list. In Experiment 2, all measures except JCN predicted that abstract words would be recalled equally as well as concrete words because of their enhanced semantic relatedness. In Experiment 3, LSA, GLoVe, and fastText predicted an enhanced concreteness effect because the concrete words were more related; three WordNet measures predicted a small concreteness effect because the abstract and concrete words did not differ in semantic relatedness; and three other WordNet measures predicted no concreteness effect because the abstract words were more related than the concrete words. A small concreteness effect was observed. Over the three experiments, only two measures, both based on simple WordNet path length, predicted all three results. We suggest that the results are not unexpected because semantic processing in episodic memory experiments differs from that in reading, similarity judgment, and analogy tasks which are the most common way of assessing such measures.

Being one of the most widely used social media tools, Twitter is seen as an important source of information for acquiring people’s attitudes, emotions, views and feedbacks. Within this context, Twitter sentiment analysis techniques were developed to decide whether textual tweets express a positive or negative opinion. In contrast to lower classification performance of traditional algorithms, deep learning models, including Convolution Neural Network (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM), have achieved a significant result in sentiment analysis. Although CNN can extract high-level local features efficiently by using convolutional layer and max-pooling layer, it cannot effectively learn sequence of correlations. On the other hand, Bi-LSTM uses two LSTM directions to improve the contexts available to deep learning algorithms, but Bi-LSTM cannot extract local features in a parallel way. Therefore, applying a single CNN or single Bi-LSTM for sentiment analysis cannot achieve the optimal classification result. An integrating structure of CNN and Bi-LSTM model is proposed in this study. ConvBiLSTM is implemented; a word embedding model which converts tweets into numerical values, CNN layer receives feature embedding as input and produces smaller dimension of features, and the Bi-LSTM model takes the input from the CNN layer and produces classification result. Word2Vec and GloVe were distinctly applied to observe the impact of the word embedding result on the proposed model. ConvBiLSTM was applied with retrieved Tweets and SST-2 datasets. ConvBiLSTM model with Word2Vec on retrieved Tweets dataset outperformed the other models with 91.13% accuracy.

In recent years, the utilization of natural language processing (NLP) and Machine Learning (ML) techniques in clinical decision support systems have shown their ability in improving and automating the diagnosis process, and reducing potential clinical errors. NLP in the Arabic language is more intricate due to several limitations, such as the lack of datasets and analytical resources compared to other languages like English. However, a clinical decision support system in the Arabic context is of significant importance. A fundamental process in NLP is extracting features from text-based data via text embedding. Word embedding is a representation of words in a numeric format that encodes the statistic, semantic, or context information. Building a neural word embedding model requires hundreds of thousands of data instances to find hidden patterns of relationships within sentences. Essentially, extracting relevant and informative features promotes the performance of the learning algorithms. The objective of this paper is to propose an Arabic neural-based word embedding model in the medical and healthcare context (called “AltibbiVec”). Around 1.5 million medical consultations and questions written in different dialects are obtained from Altibbi telemedicine company and used to train the embedding model. Three different embedding models are developed and compared, which are Word2Vec, fastText, and GloVe. The trained models were evaluated by different criteria, including the word clustering and the similarity of words. Besides, performing a specialty-based question classification. The results show that Word2Vec and fastText capture sufficiently the semantics of text more than GloVe. Hence, they are recommended for healthcare NLP-based applications.

Klasik duygu analizi yöntemlerinden farklı olarak hedef tabanlı duygu analizi (HTDA), birden fazla kategorinin olduğu karmaşık yapıdaki çevrimiçi tüketici geribildirimlerini değerlendirmede daha başarılı bir performans ortaya koyabilmektedir. Nitekim bir platformda yer alan tüketici geri bildirimleri bir ürüne ilişkin birden farklı hedefe atfedilebilmektedir ve standart duygu analizleri bu geribildirimleri analiz etmede yetersiz kalmaktadır. Literatürdeki gelişmeler gözden geçirildiğinde, HDTA çalışmalarının, duygu analizine odaklanan diğer çalışmalar içinde oldukça popüler olduğu anlaşılmaktadır. SemEval ABSA-2016 yarışmasında, HTDA için 8 farklı dilde veri setleri yayınlanmış ve ekipler duygu analizi için yarışmışlardır. Yarışmada hedef terim, kategori ve duygu sınıfı tespit etmek gibi farklı alt görevler bulunmaktadır. Bu alt görevlerin içindekilerden biri, hedef terimin tespit edilmesidir. Türkçe dili için HTDA çalışmaları oldukça sınırlıdır. Farklı diller ve farklı kelime temsil yöntemleri kullanan çalışmalar vardır. SemEval Absa 2016 yarışması Türkçe veri seti için kelime temsil yöntemlerinin etkisini inceleyen çalışma bulunmamaktadır. Bu çalışma, müşteri yorumlarındaki hedef terimlerin tespitinde farklı kelime temsil yöntemlerinin başarısının incelenmesi amacıyla gerçekleştirilmiştir. Word2Vec, Glove ve Fasttext kelime temsil yöntemleri analiz kapsamında incelenmiş ve hedef terimi en başarılı tespit edebilen yöntemin Fasttext kelime temsil yöntemi olduğu görülmüştür. Çalışmada ayrıca F-1 sınıflandırma ölçütü açısından %77 başarı oranı ile Türkçe veri seti için literatürdeki en yüksek sınıflandırma başarısı elde edilmiştir., Unlike classical sentiment analysis methods, Aspect-Based Sentiment Analysis (ABSA) can demonstrate a more successful performance in evaluating complex online consumer feedbacks including more than one category. As a matter of fact, consumer feedbacks on a platform can be referred to more than one aspect regarding a product, and standard sentiment analysis method is insufficient to analyse these comments. When the developments in the literature are reviewed, it is understood that HDTA studies are very popular among other studies focusing on sentiment analysis. In the SemEval ABSA-2016 competition, datasets were published in 8 different languages for HTDA and the teams competed for sentiment analysis. There are different subtasks in the competition, determining sub-categories such as aspect term, category and sentiment class. One of these subtasks is to determine the aspect term. HTDA studies for Turkish language are quite limited. There are studies using different languages and different word representation methods. There is no study examining the effect of word representation methods for the Turkish data set of SemEval Absa 2016 competition. This study was carried out to examine the success of different word representation methods in identifying aspect terms in customer comments. This study was carried out with the aim of examining the success of different word representation methods in identifying target terms in customer comments. Word2Vec, Glove and Fasttext word representation methods were examined within the scope of the analysis and it was seen that the method that could detect the aspect term most successfully was the Fasttext word representation method. The highest classification success for Turkish dataset in the literature with a success rate of 77% in terms of the F-1 score was also achieved in the study.

Günümüzde, sosyal medya platformları duyguları ifade etmenin en iyi yoludur. Yaklaşık iki yıldır, Covid-19 yeni koronavirüs salgının ortaya çıkması tüm dünyada olduğu gibi ülkemizde de insanların üzerinde benzeri görülmemiş karmaşık duygular yarattı. Covid-19’a karşı aşı çalışmalarının başlamasından sonra insanların duyguları daha karmaşık hale geldi. Daha yakın zamanda, Covid-19’un Delta, Omicron vb. varyantlarının çıkması da toplumda yeniden büyük bir korku yarattı. İnsanlar, bu süreçte duygu ve düşüncelerini paylaşmak üzere Twitter gibi sosyal medya araçlarına yöneldi. Twitter’da duygu analizi yapmak çok önemli ve zorlu bir görevdir. Bu çalışmada amacımız, derin öğrenme mimarilerinin gücünden faydalanarak Türk halkının aşılama süreciyle ilgili farklı duygularını araştırmak ve halkın mevcut aşılama girişimlerine yönelik tepkilerine genel bir bakış sunmaktır. Çalışmada, Twitter’da 16 Haziran 2021 ve 18 Eylül 2021 arasında paylaşılan Türkçe tweetler toplanmıştır. İnsanların her türden aşılarla ilgili duyguları, doğal dil işleme (NLP) aracı olan TextBlob kullanılarak değerlendirildi. Daha sonra, duygu sınıflandırması için yeni bir model önerildi. Önerilen model, Glove kelime gömme vektörüyle tek katmanlı Çift-yönlü Geçitli Tekrarlayan Birim (Bi-GRU) ve Evrişimli Sinir Ağı (CNN) modelini kullanan BiGRU-CNN modelidir. Önerilen yöntemin deneysel sonuçları en son modellerle kıyaslandığında umut vericidir. Bu çalışma, halkın COVID-19 aşıları hakkındaki görüşlerinin anlaşılmasını geliştirmekte ve koronavirüsü dünyadan yok etme hedefini desteklemektedir., Nowadays, social media platforms are the best way to express emotions. For nearly two years, the emergence of the Covid-19 new coronavirus epidemic has created unprecedented complex emotions on people in our country as well as all over the world. People's emotions became more complex after the start of vaccine studies against Covid-19. More recently, Covid-19's Delta, Omicron etc. The emergence of variants also created a great fear in the society again. In this process, people turned to social media tools to share their feelings and thoughts. Achieving sentiment analysis on Twitter is a very important and challenging task. The aim of this study is to investigate the different feelings of the Turkish people about the vaccination process by making use of the power of deep learning architectures and to provide an overview of the public's reactions to the current vaccination initiatives. In the study, Turkish tweets shared on Twitter between 16 June 2021 and 18 September 2021 were collected. People's feelings about vaccines of all kinds were assessed using TextBlob, a natural language processing (NLP) tool. Next, a new model for emotion classification was proposed. The proposed model is the BiGRU-CNN model using a single-layer Bi-directional Gateway Recurrent Unit (Bi-GRU) and Convolutional Neural Network (CNN) model with the Glove word embedding vector. The experimental results of the proposed method are promising when compared with the latest models. This work improves understanding of the public's views on COVID-19 vaccines and supports the goal of eradicating the coronavirus disease from the world.

Terminology is the most basic information that researchers and literature analysis systems need to understand. Mining terms and revealing the semantic relationships between terms can help biomedical researchers find solutions to some major health problems and motivate researchers to explore innovative biomedical research issues. However, how to mine terms from biomedical literature remains a challenge. At present, the research on text segmentation in natural language processing (NLP) technology has not been well applied in the biomedical field. Named entity recognition models usually require a large amount of training corpus, and the types of entities that the model can recognize are limited. Besides, dictionary-based methods mainly use pre-established vocabularies to match the text. However, this method can only match terms in a specific field, and the process of collecting terms is time-consuming and labour-intensive. Many scenarios faced in the field of biomedical research are unsupervised, i.e. unlabelled corpora, and the system may not have much prior knowledge. This paper proposes the TermInformer project, which aims to mine the meaning of terms in an open fashion by calculating terms and find solutions to some of the significant problems in our society. We propose an unsupervised method that can automatically mine terms in the text without relying on external resources. Our method can generally be applied to any document data. Combined with the word vector training algorithm, we can obtain reusable term embeddings, which can be used in any NLP downstream application. This paper compares term embeddings with existing word embeddings. The results show that our method can better reflect the semantic relationship between terms. Finally, we use the proposed method to find potential factors and treatments for lung cancer, breast cancer, and coronavirus.

The rapid change of the music market from analog to digital has caused a rapid increase in the amount of music that is spread throughout the world as well because music is easier to make and sell. The amount of music available has changed the way people find music, one of which is based on the emotion of the song. The existence of music emotion recognition and recommendation helps music listeners find songs in accordance with their emotions. Therefore, the classification of emotions is needed to determine the emotions of a song. The emotional classification of a song is largely based on feature extraction and learning from the available data sets. Various learning algorithms have been used to classify song emotions and produce different accuracy. In this study, the Bidirectional Long-short Term Memory (Bi-LSTM) deep learning method with weighting words using GloVe is used to classify the song's emotions using the lyrics of the song. The result shows that the Bi-LSTM model with dropout layer and activity regularization can produce an accuracy of 91.08%. Dropout, activity regularization and learning rate decay parameters can reduce the difference between training loss and validation loss by 0.15., Perubahan pasar musik yang cukup cepat dari analog ke digital menyebabkan bertambahnya jumlah musik yang tersebar di dunia secara cepat juga karena musik lebih mudah untuk dibuat dan dijual. Banyaknya musik yang tersedia menyebabkan berubahnya cara orang menemukan musik, salah satunya yaitu berdasarkan emosi lagu. Adanya music emotion recognition and recommendation membantu pendengar musik menemukan lagu sesuai dengan emosi mereka. Oleh karena itu, klasifikasi emosi dibutuhkan untuk menentukan emosi sebuah lagu. Klasifikasi emosi pada sebuah lagu sebagian besar didasarkan pada ekstraksi fitur dan learning dari set data yang tersedia. Berbagai algoritma learning telah digunakan untuk mengklasifikasikan emosi lagu dan menghasilkan akurasi yang berbeda. Dalam penelitian ini, metode deep learning Bidirectional Long-Short Term Memory (Bi-LSTM) dengan pembobotan kata menggunakan GloVe digunakan untuk mengklasifikasikan emosi lagu menggunakan lirik dari lagu tersebut. Pada penelitian ini, hasilnya menunjukkan bahwa model Bi-LSTM dengan dropout layer dan activity regularization dapat menghasilkan akurasi sebesar 91.08%. Parameter dropout, activity regularization dan learning rate decay dapat mengurangi selisih training loss dan validation loss sebesar 0.15.

Introduction Various orthopedic surgical procedures cause mechanical stress for gloves. In some cases, sharp-edged objects impact on the glove surfaces. The systematic description of lesions is still missing. Methods 2289 gloves from 409 surgeries [primary hip and knee arthroplasties (PA), revisions arthroplasties (RA) and arthroscopic shoulder, hip and knee surgery (AY)] from 3 clinics were examined for lesions using water tightening test according to the European norm EN 455-1. Results Arthroscopies showed the lowest rate of operations with damaged gloves (6.9%). Depending on clinic, 32.7% and 59.2% of PA surgeries generated damaged gloves, while in RA, these numbers rose to 76.0% and 72.8%, respectively. In PA and RA, the most affected finger was the index finger, whereas in arthroscopies, more damage occurred on the middle finger and the thumb. The size of the lesions was rather small with the vast majority being 1 mm or 2 mm in size. Conclusion All investigated interventions led to glove lesions. With increasing mechanical stress, the number of glove defects increased. EN 455 does not account for the intraoperative tear risk. Stricter requirements for gloves should be introduced. Glove change intervals should be defined and implemented, and new materials should be developed.

The growth of popularity of online platforms which allow users to communicate with each other, share opinions about various events, and leave comments boosted the development of natural language processing algorithms. Tens of millions of messages per day are published by users of a particular social network need to be analyzed in real time for moderation in order to prevent the spread of various illegal or offensive information, threats and other types of toxic comments. Of course, such a large amount of information can be processed quite quickly only automatically. that is why there is a need to and a way to teach computers to “understand” a text written by humans. It is a non-trivial task even if the word “understand” here means only “to classify”. the rapid evolution of machine learning technologies has led to ubiquitous implementation of new algorithms. A lot of tasks, which for many years were considered almost impossible to solve, are now quite successfully solved using deep learning technologies. this article considers algorithms built using deep learning technologies and neural networks which can successfully solve the problem of detection and classification of toxic comments. In addition, the article presents the results of the developed algorithms, as well as the results of the ensemble of all considered algorithms on a large training set collected and tagged by Google and Jigsaw.

Bug triage processes are intended to assign bug reports to appropriate developers effectively, but they typically become bottlenecks in the development process-especially for large-scale software projects. Recently, several machine learning approaches, including deep learning-based approaches, have been proposed to recommend an appropriate developer automatically by learning past assignment patterns. In this paper, we propose a deep learning-based bug triage technique using a convolutional neural network (CNN) with three different word representation techniques: Word to Vector (Word2Vec), Global Vector (GloVe), and Embeddings from Language Models (ELMo). Experiments were performed on datasets from well-known large-scale open-source projects, such as Eclipse and Mozilla, and top-k accuracy was measured as an evaluation metric. The experimental results suggest that the ELMo-based CNN approach performs best for the bug triage problem. GloVe-based CNN slightly outperforms Word2Vec-based CNN in many cases. Word2Vec-based CNN outperforms GloVe-based CNN when the number of samples per class in the dataset is high enough.

In order to accelerate the performance of various Natural Language Processing tasks for Roman Urdu, this article for the very first time provides 3 neural word embeddings prepared using most widely used approaches namely Word2vec, FastText, and Glove. The integrity of generated neural word embeddings is evaluated using intrinsic and extrinsic evaluation approaches. Considering the lack of publicly available benchmark datasets, it provides a first-ever Roman Urdu public dataset which consists of 3241 sentiments annotated against positive, negative, and neutral classes. To provide benchmark baseline performance over the presented dataset for Roman Urdu sentiment analysis, we adapt diverse machine learning (Support Vector Machine, Logistic Regression, Naive Bayes), deep learning (convolutional neural network, recurrent neural network), and hybrid deep learning approaches. Performance impact of generated neural word embeddings based representation is compared with other most widely used bag of words based feature representation approaches using diverse machine and deep learning classifiers. In order to improve the performance of Roman Urdu sentiment analysis, it proposes a novel precisely extreme multi-channel hybrid methodology which makes use of convolutional and recurrent neural networks along with pre-trained neural word embeddings. The proposed hybrid approach outperforms adapted machine learning approaches by the significant figure of 9% and deep learning approaches by the figure of 4% in terms of F1-score.

This study responds to a descriptive-technological study, developed with the objective of developing and implementing an electronic glove to promote productivity and safety in the tasks of the industrial workshop of the INEM Educational Institution in the city of Monteria, Cordoba. Among the stages that were considered in the development of this study, the application of surveys before and after the implementation of the electronic device is highlighted, with the purpose of determining the impact of this in relation to the productivity of the tasks and the welfare of the operators of the industrial workshop. As well as the development of a web page that stores the data collected by the electronic glove to allow a better management of the information. The research yielded optimal results that demonstrate the relevant intervention that this type of technology has within the optimization of industrial tasks, in terms of time management, inventory management and ease in the handling of the different tools of the workshop; along with the efficient management of information, this because it caused a better manageability of processes and items, as well as offers a new innovative and creative tool for the industry. Resumen VII Abstract VIII Introducción 15 Capítulo 1. Consideraciones Generales 17 1.1 Planteamiento del Problema 17 Árbol del Problema 19 1.2 Objetivos 21 1.2.1 Objetivo General 21 1.2.2 Objetivos Específicos 21 1.3 Hipótesis 22 1.4 Alcances de la Investigación 22 1.5 Justificación 23 Capítulo 2. Marco Referencial 24 2.1 Estado del Arte 24 2.2 Marco Conceptual 42 2.2.1 Anatomía de la Mano Humana 42 2.2.2 Electrónica 42 Dispositivo Electrónico 42 2.2.5 Industrialización 43 Industria 43 2.2.6 Sistema 43 2.2.8 Software 44 2.2.9 Hardware 44 2.2.10 Componentes que Constituyen al Sistema de Información Computado 44 Capítulo 3. Marco Metodológico 45 3.1 Tipo de Trabajo 45 3.1.1 Población Y Muestra 45 3.2 Estrategias de Recolección de la Información 45 3.3. Proceso de la Investigación 46 Variables de la Investigación 46 3.3.1. Fase I: Estudio, Análisis e Interpretación del Sistema 47 3.3.2. Fase II: Caracterización del Sistema 47 3.3.3. Fase III: Diseño, Desarrollo e Implementación del Sistema 47 3.3.4 Fase IV: Descripción de Pruebas 48 3.4 Especificación de Requisitos 48 3.5 Requisitos Funcionales 50 3.6 Requisitos No Funcionales 51 3.7 Diagramas 51 3.7.1 Diagramas de Casos de Uso 51 3.8 Documentación Casos de Uso 57 3.9 Diagrama de Clase 61 3.10 Diagrama de Componentes 61 3.11 Modelo Entidad Relación E R 62 3.12 Modelo Relacional 63 3.13 Diagramas de Secuencia 63 3.14 Diagramas de Actividad 68 3.15 Diseño del Software 72 3.15.1 Diseño Arquitectónico del Software 74 3.15.2 Diseño del Dispositivo 75 3.16 Pruebas del Hardware 78 3.16.1.1 Casos de Prueba 78 3.16.1.2 Preparación de Datos de Entrada 80 3.16.1.3 Prueba de Caja Negra 81 3.16.1.4 Prueba de Caja Blanca 86 3.16.1.5 Informe de Pruebas 90 Capítulo 4. Análisis e Interpretación de Resultados 91 ProGlove: El Guante Inteligente que Quiere Revolucionar el Sector Industrial 92 Capítulo 5 103 5.1 Conclusiones 103 Recomendaciones 104 Trabajos Futuros 104 Anexos 105 Manual de Usuario 105 Evidencias Fotográficas 113 Referencias 114 El presente estudio responde a un estudio de tipo descriptivo-tecnológico, desarrollado con el objetivo de desarrollar e implementar un guante electrónico para favorecer la productividad y seguridad en las tareas del taller industrial de la Institución Educativa INEM de la Ciudad de Montería, Córdoba. Entre las etapas que se tomaron en cuenta en el desarrollo de este estudio se resalta la aplicación de encuestas antes y después de la implementación del dispositivo electrónico, con el propósito de determinar el impacto de este en relación con la productividad de las tareas y el bienestar de los operarios del taller industrial. Así como el desarrollo de una aplicación web que almacena los datos recogidos por el guante electrónico para permitir una mejor administración de la información. La investigación arrojó resultados óptimos que demuestran la intervención relevante que tiene este tipo de tecnología dentro de la optimización de las labores industriales, en cuanto al manejo del tiempo, la gestión de inventarios y facilidad en la manipulación de las distintas herramientas del taller; junto con la gestión eficiente de la información, esto debido a que provocó una mejor manejabilidad de los procesos y de los artículos, así como ofrece una nueva herramienta innovadora y creativa para la industria. Pregrado Ingeniero(a) de Sistemas Trabajos de Investigación y/o Extensión

Technology-aided hand functional assessment has received considerable attention in recent years. Its applications are required to obtain objective, reliable, and sensitive methods for clinical decision making. This systematic review aims to investigate and discuss characteristics of technology-aided hand functional assessment and their applications, in terms of the adopted sensing technology, evaluation methods and purposes. Based on the shortcomings of current applications, and opportunities offered by emerging systems, this review aims to support the design and the translation to clinical practice of technology-aided hand functional assessment. To this end, a systematic literature search was led, according to recommended PRISMA guidelines, in PubMed and IEEE Xplore databases. The search yielded 208 records, resulting into 23 articles included in the study. Glove-based systems, instrumented objects and body-networked sensor systems appeared from the search, together with vision-based motion capture systems, end-effector, and exoskeleton systems. Inertial measurement unit (IMU) and force sensing resistor (FSR) resulted the sensing technologies most used for kinematic and kinetic analysis. A lack of standardization in system metrics and assessment methods emerged. Future studies that pertinently discuss the pathophysiological content and clinimetrics properties of new systems are required for leading technologies to clinical acceptance.

This article presents results from a study that developed and tested a word embedding trained on a dataset of South African news articles. A word embedding is an algorithm-generated word representation that can be used to analyse the corpus of words that the embedding is trained on. The embedding on which this article is based was generated using the Word2Vec algorithm, which was trained on a dataset of 1.3 million African news articles published between January 2018 and March 2021, containing a vocabulary of approximately 124,000 unique words. The efficacy of this Word2Vec South African news embedding was then tested, and compared to the efficacy provided by the globally used GloVe algorithm. The testing of the local Word2Vec embedding showed that it performed well, with similar efficacy to that provided by GloVe. The South African news word embedding generated by this study is freely available for public use.

Text classification is an essential task in natural language processing. While graph neural networks (GNNs) have successfully been applied to this problem both through graph classification and node classification approaches, their typical applications suffer from several issues. In the graph classification case, common graph construction techniques tend to leave out syntactic information. In the node classification case, most widespread datasets and applications tend to suffer from encoding relatively little information in the chosen node features. Finally, there are great benefits to be gained from combining the two GNN approaches. To tackle these concerns, we propose DepNet, a two-stage framework for text classification using GNN models. In the first stage we replace current graph construction methods by utilizing syntactic dependency parsing in order to include as much syntactic information in the GNN input as possible. In the second stage we combine both graph classification and node classification methods by utilizing the former to produce node embeddings for the latter, maximizing the potential of GNNs for text classification. We find that this technique significantly improves the performance of both graph classification and node classification approaches to text classification.

Twitter is one of the popular social media to channel opinions in the form of criticism and suggestions. Criticism could be a form of hate speech if the criticism implies attacking something (an individual, race, or group). With the limit of 280 characters in a tweet, there is often a vocabulary mismatch due to abbreviations which can be solved with word embedding. This study utilizes feature expansion to reduce vocabulary mismatches in hate speech on Twitter containing Indonesian language by using Global Vectors (GloVe). Feature selection related to the best model is carried out using the Logistic Regression (LR), Random Forest (RF), and Artificial Neural Network (ANN) algorithms. The results show that the Random Forest model with 5.000 features and a combination of TF-IDF and Tweet corpus built with GloVe produce the best accuracy rate between the other models with an average of 88,59% accuracy score, which is 1,25% higher than the predetermined Baseline. The number of features used is proven to improve the performance of the system.

One of the trends in Natural Language Processing (NLP) is the use of word embedding. Its aim is to build a low dimensional vector representation of words from text corpora. Global Vectors for Word Representation (GloVe) and Sikp-Gram with Negative Sampling (SGNS) are two representative word embedding methods. Existing papers have different conclusions on the performance of these two methods. This thesis focuses on GloVe and studies its commonalities and differences with SGNS. Word co-occurrence is the cornerstone of all word embedding algorithms. One difference between GloVe and SGNS is the definition of co-occurrence. The weight of co-occurring words tapers o↵ with the distance between them. GloVe and SGNS adopts different weighting schemes. In SGNS, weight decreases linearly with the distance. In GloVe, the weight decreases harmonically, giving less weight to the words in the center of the window. We propose GloVe-L (GloVe Linear), by changing the weighting scheme to the linear weighting. We find that GloVe-L outperforms GloVe consistently in word similarity tasks. The conclusion is supported by extensive experiments on 8 Word evaluation benchmarks on Wikipedia training corpus. The thesis also explores the impact of hyper-parameters on the result, including window size and xmax in GloVe. Another interesting observation is that Glove-L does not work well for word analogy tasks.

Internet technologies are emerging very fast nowadays, due to which web pages are generated exponentially. Web page categorization is required for searching and exploring relevant web pages based on users’ queries and is a tedious task. The majority of web page categorization techniques ignore semantic features and the contextual knowledge of the web page. This paper proposes a web page categorization method that categorizes web pages based on semantic features and contextual knowledge. Initially, the GloVe model is applied to capture the semantic features of the web pages. Thereafter, a Stacked Bidirectional long short-term memory (BiLSTM) with symmetric structure is applied to extract the contextual and latent symmetry information from the semantic features for web page categorization. The performance of the proposed model has been evaluated on the publicly available WebKB dataset. The proposed model shows superiority over the existing state-of-the-art machine learning and deep learning methods.

Velik broj ljudi koristi društvene mreže. Gotovo svatko može objaviti informacije koje ne moraju biti točne. Kako bismo bolje razumjeli što čini takve informacije popularnim, pojavila se potreba za modelima koji mogu obraditi golem skup informacija. U ovom radu smo konstruirali skup podataka sačinjen od objava prikupljenih s Twittera. Svaka je objava iz domene obrade prirodnog jezika te uključuje sažetak i naslov znanstvenog rada dohvaćenog preko poveznice iz objave. U ovom radu smo proširili trenutna rješenja tako što smo dodali novi skup informacija dobivenih iz teksta znanstvenog rada, informacijama o pratiteljima autora objave te informacijama o trendu korištenih riječi u samoj objavi. U rezultatima smo pokazali u kojoj mjeri takve informacije utječu na predikciju. Pokazano je kako popularnost samog znanstvenog rada ne ovisi o tekstu rada. Modeli koje smo koristili su bazirani na GloVe-u, kombinaciji GloVe-a i GRU-a, te BERT-u. Uspoređujući navedene modele nismo dobili veliku razliku u točnosti predikcije unutar istog zadatka. Uobičajeno autori rješavaju problem kao klasifikaciju dok smo mi pokazali kako regresijski pristup poboljšava performanse modela. Na kraju smo pokazali kako normalizacija prediktivne varijable brojem pratitelja olakšava problem i konsekventno povlači povećanje performanse modela. With the advent of social networks, information sharing has become available to the general public, empowering individuals to be sources of news as well as misinformation. In order to better gauge the reach a post will have on a platform, an opportunity has arisen for models which can predict the popularity of a user's post on a platform. In this thesis, we constructed the dataset that consists of posts from the NLP community and contains arXiv pre-prints, extended the current approaches with graph-based features computed from user's social network, trend-based features computed from the historically popular tweets, as well as features extracted from arXiv pre-prints. By adding those features, we increased the model's performance, showed what features influence prediction the most, and confirmed that the popularity of pre-print isn't affected much by its content. We used GloVe averaging, GRU with GloVe, and BERT variants for which we compared the results and showed that there isn't much difference in their performance. Traditionally, scholars framed the task as classification and we showed how framing the task as regression helps improve the model's performance. Finally, we introduced a normalization algorithm that made popularity user-dependent, which simplified the problem and increased the model's performance.

This article focuses on a sensorial glove prototype capable of acquiring hand motion and estimating its pose. The presented solution features twelve inertial measurement units (IMUs) to track hand orientation. The sensors are attached to a glove to decrease the project cost. The system also focuses on sensor fusion algorithms for the IMUs and further implementations, presenting the algebraic quaternion algorithm (AQUA), used because of its modularity and intuitive implementation. An adaptation of a human hand model is proposed, explaining its advantages and its limitations. Considering that the calibration is a very important process in gyroscope performance, the online and offline calibration data was analyzed, pointing out its challenges and improvements. To better visualize the model and sensors a simulation was conducted in Unity.