Search

Data assimilation schemes blend observational data, with limited coverage, with a short term forecast to produce an analysis, which is meant to be the best estimate of the current state of the atmosphere. Appropriately specifying observation error statistics is necessary to obtain an optimal analysis. Observation error can originate from instrument error as well as the error of representation. While representation error is most commonly associated with unresolved scales and processes, this term is often considered to include contributions from pre-processing or quality control and errors associated with the observation operator. With a focus on practical operational implementation, this chapter aims to define the components of observation error, discusses their sources and characteristics, and provides an overview of current methods for estimating observation error statistics. We highlight the implicit assumptions of these methods, as well as their shortcomings. We will detail current operational practice for diagnosing observation error and accounting for correlated observation error. Finally, we provide a practical methodology for using these diagnostics, as well as the associated innovation-based observation impact, to optimize the assimilation of meteor radar observations in the upper atmosphere.

Being able to estimate future velocity on a road network has applications from vehicle navigation systems to emergency vehicle dispatching systems. The existence of traffic congestion can severely impact travelers’ travel time and in this paper we explore methods to take it into account in velocity forecasting models. Using a data approach, different traffic observations can be classified into classes with and without congestion. Our research shows that using congestion as an attribute can reduce the prediction error when implementing machine learning models, such as random forest or multi-layer perceptron. Furthermore, training separate models for each class performs better than using congestion as an extra attribute. A methodology for the congestion pattern identification is proposed, based only in the velocity and volume values.

Due to the impact of modern information technology, the construction of transportation infrastructure has entered the “Intelligent Era.” The conventional construction technology is also undergoing a paradigm shift. In this paper, the term “intelligent construction” is defined, and the status quo of global transportation infrastructure construction with emerging intelligent construction technologies is summarized. The framework of intelligent construction is presented on all aspects of the integration of modern information technology and the conventional highway technologies. The essential elements of intelligent construction include sensing, analysis, decision-making, and execution. Big data, machine learning, and expert system are applied to provide practical technical solutions for intelligent construction. This paper also elaborates on the process of intelligent decision-making and auto-feedback machine controls. Finally, the future development of intelligent construction is laid out. The globally coordinated efforts by the International Society for Intelligent Construction (ISIC) will help the leading development and implementation of intelligent construction into the future.

Visual programming has become a term that stands for intuitive software development. It dispenses largely with textual notation and instead uses graphical program modules. The use of visual techniques is connected with the hope, among other things, to be able to create and understand programs more easily and better than before, which would open up access to software development for non-programmers. This could lead to a reduction of the enormous application backlog in many places. In this research, we design and build a web-based Visual Query builder tool to query the Waves blockchain database, using a flow-based programming language as well as an empirical evaluation of the tool that, on the one hand, measures the semantic understandability of a visual representation of a query, and compare it to its textual representation in SQL. On the other hand, we evaluate the user acceptance of the tool.

The novel hybrid learning model based on neural networks and grammatical inference is proposed in the paper. The model is used within the multi-derivational parsing of vague language methodology. The foundations of the methodology and the learning algorithms are presented. The model has been used for the implementation of the short term electrical load forecasting system.

In this paper we present a new formulation of the local minimum problem which characterizes the artificial potential field method for obstacle avoidance. Our approach consists of detecting the local-minimum situation by measuring the angle formed by the two force vectors (repulsive force and attractive force vectors). Afterward, we suggest adding a new vector expressed in term of the angle formed by two vectors. The modified APF method we are suggesting ensures a local minimum-free obstacle-free path planning algorithm. The developed algorithm is verified through MATLAB platform simulations. One robot is put in some position moves within a 50 × 50 m2 area to reach a goal point while avoiding three stationary obstacles. Four different initial configurations are simulated. The robot dynamics are represented by first-order integrator model. The obtained trajectories showed a clear effectiveness of the proposed algorithm.

An automatic methodology is proposed for extracting valuable information from customer experience and comments after they purchase products, including the prediction of negative opinions from written phrases and the estimation of potential savings from attending unsatisfied customers. The complete methodology consists in natural language pre-processing, the calculation of the Term Frequency - Inverse Document Frequency (TF-IDF) index, the construction of binary classification models for negative opinions, and the estimation of the potential savings from using the knowledge extracted by the predictive models. The function for estimating the savings makes use of soft measures, based on fuzzy possibility functions and the estimation of client churning (given that they are unsatisfied with the delivery). The prediction algorithms for negative emotions are shown to achieve good results when tested with out-of-sample comments coming from social networks, and the complete methodology obtains significant savings when applied for e-commerce customer support.

Accelerated by current research on the Internet of Things, working with virtual models and things has taken a significant step forward. The term digital twins refers to the development of virtual models of processes, products or services that allow the simulation of system behavior and the early detection of problems before the real system exists. Simulations have been state of the art for a long time, but with the service-oriented possibilities of the Internet a new quality of simulation results. In the meantime, it has become possible, for example, to configure a virtual computer on the Internet and run algorithms on it without this computer in fact existing, see e.g. AWS (AMAZON web services). This paper discusses possibilities that the digital twin concept can offer to educational purposes in remote and virtual laboratories.

Currently, there is a very rapid growth of information published on the Internet, both on social media and news sites. However, a serious problem is a disinformation in the form of fake news. Due to the rapid spread of information on the Internet, it is very important to be able to quickly identify true and fake news. The solution to this problem can be an initial analysis of news by its title and quick selection of true or fake news. Additionally, the possibility of balancing precision and recall as the quality of classification measures could allow for better news selection. In this paper, we propose the use of the adaptive goal function of ant colony optimization algorithms in fake news detection. The goal of this solution is to increase recall or precision of the selected class – in this case fake or true news. We use natural language processing (NLP) to describe the title of the news. In addition, a constrained term matrix is used. The choice of titles alone and the restriction of the words analyzed are related to speeding up the initial classification. Eventually, we present an analysis of a real dataset and classification results (detailing recall and precision) of news using the adaptive goal function of the ACDT algorithm.

The paper deals with development of iterative schemes for an implicit finite difference approximation of one-dimensional model for Darcy flow of magma in rock. The difficulties are that the solution typically lacks regularity due to the degenerate non-linear diffusion term, and the non-local non-linear source. We construct and compare four iterative algorithms for realization of the fully implicit discretizations. We discuss positivity and boundness of the numerical solutions as well as the convergence of the iteration processes. Results from numerical experiments, using the presented algorithms are reported.

The problem of limited labelled data availability causes under-fitting, which negatively affects the development of accurate time series based prediction models. Two-hybrid deep neural network architectures, namely the CNN-BiLSTM and the Conv-BiLSTM, are proposed for time series based transductive transfer learning and compared to the baseline CNN model. The automatic feature extraction abilities of the encoder CNN module combined with the superior recall of both short and long term sequences by the decoder LSTM module have shown to be advantageous in transfer learning tasks. The extra ability to process in both forward and backward directions by the proposed models shows promising results to aiding transfer learning. The most consistent transfer learning strategy involved freezing both the CNN and BiLSTM modules while retraining only the fully connected layers. These proposed hybrid transfer learning models were compared to the baseline CNN transfer learning model and newly created hybrid models using the \(R^2\), MAE and RMSE metrics. Three electrical vehicle data-sets were used to test the proposed transfer frameworks. The results favour the hybrid architectures for better transfer learning abilities relative to utilising the baseline CNN transfer learning model. This study offers guidance to enhance time series-based transfer learning by using available data sources.

Accurate prediction of power load plays an important role in the optimal scheduling of resources. However, the lack of power data in the traditional automatic acquisition system inevitably affects the subsequent data analysis. With the help of on-site real-time monitoring, the integrity of data collection can be ensured. In this paper, a load forecasting model based on the fusion of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) is proposed. Through training the historical data collected by the on-duty robot, a complete network model is constructed. The network extracts the effective sequence features of the input data through CNN network, and gets the load prediction results through LSTM network. The experimental results show that the fusion network of CNN and LSTM obtains higher prediction accuracy than present algorithms.

Consumer reviews show inconsistent ratings when compared to their contents as a result of sarcastic feedback. Consequently, they cannot provide valuable feedback to improve products and services of the firms. One possible solution is to utilize consumer review contents to identify the true ratings. In this work, different multi-class classification methods were applied to assign automatic ratings for consumer reviews based on a 5-star rating scale, where the original review ratings were inconsistent with the content. Two term weighting schemes (i.e. tf-idf and tf-igm) and five supervised machine learning algorithms (i.e. k-NN, MNB, RF, XGBoost and SVM) were compared. The dataset was downloaded from the Amazon website, and language experts helped to correct the real rating for each consumer review. After verifying the effectiveness of the proposed methods, the multi-class classifier model developed by SVM along with tf-igm returned the best results for automatic ratings of consumer reviews, with average improved scores of accuracies and F1 over the other methods at 11.7% and 10.5%, respectively.

The term ‘ballroom dancing’ refers to a social and competitive type of partnered dance. Competitive ballroom dancing consists of 10 different types of dances performed to specific styles of music unique to each type of dance. There are few algorithms attempting to differentiate between pieces of music and classify them into the categories, making it hard for beginners to identify which dance corresponds to a certain piece of music they may be listening to. In our research, we attempted to use an end-to-end machine learning approach to easily and accurately classify music into the 10 different types of dance. We experimented with four types of machine learning models and received the highest accuracy of 83% using a Deep Neural Network with three hidden layers. With this algorithm, we can facilitate the learning experience of beginner ballroom dancers by aiding them to distinguish between different types of ballroom dancing music.

Part 5: Autonomous Agents; International audience; Real-word intelligent agents must be able to detect sudden and unexpected changes to their task environment and effectively respond to those changes in order to function properly in the long term. We thus isolate a set of perturbations that agents ought to address and demonstrate how task-agnostic perturbation detection and mitigation mechanisms can be integrated into a cognitive robotic architecture. We present results from experimental evaluations of perturbation mitigation strategies in a multi-robot system that show how intelligent systems can achieve higher levels of autonomy by explicitly handling perturbations.

For several years now, thanks to the emergence of new information and communication technologies, information has become increasingly available and accessible. The IT field today has a very large amount of data allowing the search for any information. However, exploiting this large amount of data makes finding and classifying precise data complex and time consuming. This difficulty has motivated the development of new adapted data classification tools. Classification is first used to refer to the division of a set of individuals into classes so that every individual belongs to one class and only one. But the term classification is also used to designate nested systems of classes so we can say that it is a statistical operation which consists in grouping objects (individuals or variables or observations) into a limited number of groups (classes, segments), and to classify individuals according to some of their characteristics. There are different types of classification, but one of the most intuitive and widely used is supervised classification. The overall objective of the classification is to identify the classes to which objects belong based on descriptive features (attributes, characteristics). Data classification is a delicate problem that arises in many sciences such as data mining analysis. The broker's primary goal is to reduce the overhead costs associated with managing web service level agreements, while maintaining compatibility with existing infrastructure and vendors.

Disentangled representation learning has undoubtedly benefited from objective function surgery. However, a delicate balancing act of tuning is still required in order to trade off reconstruction fidelity versus disentanglement. Building on previous successes of penalizing the total correlation in the latent variables, we propose TCWAE (Total Correlation Wasserstein Autoencoder). Working in the WAE paradigm naturally enables the separation of the total-correlation term, thus providing disentanglement control over the learned representation, while offering more flexibility in the choice of reconstruction cost. We propose two variants using different KL estimators and perform extensive quantitative comparisons on data sets with known generative factors, showing competitive results relative to state-of-the-art techniques. We further study the trade off between disentanglement and reconstruction on more-difficult data sets with unknown generative factors, where we expect improved reconstructions due to the flexibility of the WAE paradigm.

Dark jargons are benign-looking words that have hidden, sinister meanings and are used by participants of underground forums for illicit behavior. For example, the dark term “rat” is often used in lieu of “Remote Access Trojan”. In this work we present a novel method towards automatically identifying and interpreting dark jargons. We formalize the problem as a mapping from dark words to “clean” words with no hidden meaning. Our method makes use of interpretable representations of dark and clean words in the form of probability distributions over a shared vocabulary. In our experiments we show our method to be effective in terms of dark jargon identification, as it outperforms another baseline on simulated data. Using manual evaluation, we show that our method is able to detect dark jargons in a real-world underground forum dataset.

In this paper, a Deep learning-based approach is analyzed to improve the accuracy of the Language Modelling of a system that is a part of ASR. The main intent is to build an End to End Neural Network Language Model. Generally, ASR works in three different modes. The modes are Acoustic Modeling, Lexicons, and Language Modeling. In ASR, G2P models are vital in the case of analyzing the Out Of Vocabulary words. Mostly these G2P models were fabricated using the traditional approaches such as Bi-gram, n-gram language models. The traditional approaches are deceptive in aiding the meanings, long term dependencies of the words and are also impotent in rendering the phonemes of new words. Thus Seq2Seq G2P models exceed the state of the art of the traditional approach. As we all know, the ASR (Speech to Text) system is widely used in AI applications such as voice dialing, system control, navigation, etc. The method which is applied here is the Statistical based one. A method other than the Rule-based approach is trialed out. A suitable neural network of LSTM and BI-LSTM is proposed which enhances the accuracy of the system.

Term recognition is the foundation of natural language processing, especially in chemical domain. The most important feature of domain text is terminology. Because of the variety and the complexity of chemical terms, chemical term NER is too difficult to be used in a production environment such as patents search. Based on the theory of morphemes and data of Chinese patents, we construct a chemical morpheme classification according to the composition characteristics and formation rules of chemical terms. A simple and stable HMM (The Hidden Markov Model) model is used to model the word formation relationship among different morphemes. The improved forward algorithm is applied to estimate the word formation probability in terminology recognition. It solves the problem of poor recognition of the long terms in low frequency in traditional statistical methods. It also solves the problem of subjectivity and completeness in establishing rules to some extent. Finally, the F1 reaches 91.58% in chemical patents.

Traffic prediction is essential to public transportation management in cities. However, long-term traffic prediction involves complex spatio-temporal correlations changing dynamically, which is highly challenging to capture in road networks. We focus on these dynamic correlations and propose a spatio-temporal graph modeling method to solve the long-term traffic prediction problem. Our proposed method builds a Spatio-Temporal Graph Attention network for Traffic Prediction (STGATP), exploring and capturing the complex spatial-temporal nature in traffic networks. We apply dilated causal convolution with a gated fusion in the temporal modeling block, and diffusion convolution with the attention mechanism in the spatial modeling block. This results in that STGATP can simultaneously capture spatial dependencies and temporal dependencies in road networks. Finally, we conduct the experiments on public traffic datasets METR-LA and PEMS-BAY, and our method reaches superior performance. In particular, STGATP surpasses state-of-the-art methods by up to 11% improvement of RMSE measure on the PEMS-BAY datasets.

In this chapter, we describe the user interface of the FasTensor library, a C++ implementation of the FasTensor programming model presented in previous chapters. Hence, the term FasTensor will mostly refer to the FasTensor library that implements the FasTensor programming model. Currently, the FasTensor programming model is available in C++. Support for other languages, such as Java, Python, and Julia, are planned.

Natural language processing (NLP) is a problem which attracts lots of attention from researchers. This study analyzes and compares a different method to classify text sentences or paragraphs in Vietnamese into different categories. The work utilizes a sequence of techniques for data-preprocessing, customize learning model and methods before using Term Frequency-Inverse Document Frequency (TF-IDF) for model training. This classification model could contribute positively to many Vietnamese text-analyzing based businesses, such as social network, e-commerce, or data mining in general. This problem’s challenge relies on two main aspects: the Vietnamese language itself and current NLP researches for the Vietnamese language. The paper utilizes the pros of many different classification methods to provide better accuracy in text classification.

Advances in metering technologies and machine learning methods provide both opportunities and challenges for predicting building energy usage in the both the short and long term. However, there are minimal studies on comparing machine learning techniques in predicting building energy usage on their rolling horizon, compared with comparisons based upon a singular forecast range. With the majority of forecasts ranges being within the range of one week, due to the significant increases in error beyond short term building energy prediction. The aim of this paper is to investigate how the accuracy of building energy predictions can be improved for long term predictions, in part of a larger study into which machine learning techniques predict more accuracy within different forecast ranges. In this case study the ‘Clarendon building’ of Teesside University was selected for use in using it’s BMS data (Building Management System) to predict the building’s overall energy usage with Support Vector Regression. Examining how altering what data is used to train the models, impacts their overall accuracy. Such as by segmenting the model by building modes (Active and dormant), or by days of the week (Weekdays and weekends). Of which it was observed that modelling building weekday and weekend energy usage, lead to a reduction of 11% MAPE on average compared with unsegmented predictions.

Text classification plays a vital role in many real-life applications. There are different methods for text classification primarily Naive Bayes classifier, support vector machine, etc. A good text classifier must efficiently classify large set of unstructured documents with optimal accuracy. Many techniques have been proposed for text classification. In this paper, we propose an integrated approach for text classification which works in two phases. In initial preprocess phase, we select the frequent terms and adjust the term weight by use of information gain and support vector machines. Second phase consists of applying Naive Bayes classifier to the document vector. The experiment has been performed on the open research dataset of Forum of Information Retrieval (FIRE). In association rule, the correlation between data items is obtained with no requirement of external knowledge, whereas in classification, attention is given to small set of rules with the help of external knowledge. The proposed work uses FP-growth algorithm with absolute pruning for obtaining frequent text sets, and then, Naive Bayes classifier model is used for training and constructing a model for classification. Our experimental result shows increase in efficiency while comparing with other traditional text classification methods.

Credit card fraud detection over the years remains a major concern for all and this has made the field to receive huge attention from researchers. One way of addressing this concern is for financial sectors and government agencies to reliably detect fraud in any transaction so as to prevent financial losses incurred by these sectors and card owners. Exiting fraud detection system is plagued with misclassification and high false-positive rate. To prevent misclassification, this research therefore aims to deploy Long Short Term Memory-Recurrent Neural Network (LSTM-RNN) for classifying financial transaction as fraudulent or not. The system extracted desired features from two different dataset gotten from kaggle repository using Principal Component Analysis and preprocessed using Arbitrary Assignment Method and Min–Max scalar algorithm for Normalization. The work was implemented using python programming language. The relevant features selected were then fed into the LSTM-RNN for classification. The results obtained were compared with past work and our fraud model recorded high classification accuracy as well as reduced false alarm rate. It has 99.58% Prediction Accuracy, 99.6% Precision, and Recall of 80%. The system will enable financial institutions and government agencies involved in financial transaction to detect the occurrence of fraud and be able to proffer the necessary solution as appropriate.

Aspect category sentiment classification aims at predicting the sentiment polarity of the given aspect category. Since the aspect category may not occur in the sentence, it is hard for the model to directly find the appropriate sentiment words for the aspect category and disregard unrelated ones. To address it, previous works have explored leveraging implicitly the information of the aspect term in the sentence and demonstrated the effectiveness of such information. Inspired by this conclusion, we propose a two-stage strategy named Locate-Combine(LC) to utilize the aspect term in a more straightforward way, which first locates the aspect term and then takes it as the bridge to find the related sentiment words. Specifically, in the “Locate” stage, we locate the aspect term corresponding to the given aspect category in the sentence, which can crystallize the target and further enable our model to focus on the target-related words. In the “Combine” stage, we first apply the graph convolutional network (GCN) over the dependency tree of the sentence to combine the information of the aspect term and related sentiment words and then take the output representation corresponding to the located aspect term to predict the sentiment polarity. The experimental results on the public datasets show that the proposed two-stage strategy is effective, which achieves state-of-the-art performance. Furthermore, our model can output explainable intermediate results for model analysis. (Code can be found at https://github.com/SCIR-MSA-Team/LC-ACSA)

Frequent communications among massive terminal devices are ubiquitous in forthcoming 5G Internet of Thing (IoT) networks. It strengthens links of massive machine-type-communication (MMTC), pushes forward the process of Internet of everything. However, due to continual interactions among different devices and the broadcast characteristic of wireless channels, it also brings new security challenges. Recently, physical layer security launches a new solution to guarantee information theoretic security. To enhance the physical layer security performance of massive intelligent devices, especially in D2D communications, the game theory and machine learning methods are introduced. In this paper, we first review physical layer security problems on D2D communications under different attack scenarios. Game theory is proposed to describe hierarchical and heterogeneous interactions among legitimate users and active attackers in 5G IoT networks, then some distributed machine learning methods are proposed to obtain equilibrium states among different agents. Moreover, numerical results are provided to verify availability and efficiency of proposed game-theoretic learning approaches. Finally, we discuss open issues and future research directions in term of anti-eavesdropping and anti-jamming problems in D2D communications when facing active attackers.

The long term privacy of voting systems is of increasing concern as quantum computers come closer to reality. Everlasting privacy schemes offer the best way to manage these risks at present. While homomorphic tallying schemes with everlasting privacy are well developed, most national elections, using electronic voting, use mixnets. Currently the best candidate encryption scheme for making these kinds of elections everlastingly private is PPATC, but it has not been shown to work with any mixnet of comparable efficiency to the current ElGamal mixnets. In this work we give a paper proof, and a machine checked proof, that the variant of Wikstrom’s mixnet commonly in use is safe for use with the PPATC encryption scheme.

Artificial’ ability the term artificial intelligence (AI) was first coined by John McCarthy in 1956 when he had the initial theoretical meeting on this topic. But this journey to see if machines will really believe started much before this. Five years later Alan Turing published the article on this idea of machines being able to imitate human race and the ability to do intelligent things, e.g., play chess. Yet, most people be uncertain about the purpose of AI. What if he produces the AI that has the only aim of estimating the amount of Pi, but has decided that by getting earthly death and giving human lives, it could gain another digit to its collection of digits efficiently? Therefore, the entire idea of knowing in machines is unlikely. The only plausible explanation of the power to philosophize, emotionally interact with humans and reason motivation is artificial awareness.

Hospital information system stores all clinical information, whose major part is electronic patient records written by doctors, nurses and other medical staff. Since records are described by medical experts, they are rich in knowledge about medical decision making. This paper proposes an approach to extract clinical knowledge from the texts of clinical records. The method consists of the following three steps. First, discharge summaries, which include all clinical processes during the hospitalization, are extracted from hospital information system. Second, morphological and correspondence analysis generates a term matrix from text data. Then, finally, machine learning methods are applied to a term matrix in order to acquire classification knowledge. We compared several machine learning methods by using discharge summaries stored in hospital information system. The experimental results show that random forest is the best classifier, compared with deep learning, SVM and decision tree. Furthermore, random forest gains more than 90% classification accuracy.

Bilingual term resources are helpful on collaborative translation tasks. Firstly, we build an English-Chinese parallel term resource of social sciences (SocialSciTerm) based on the bilingual big data and advanced theories of information extraction and alignment. Secondly, we verify the SocialSciTerm’s accuracy and practicality through a translation task of social sciences. The statistical results indicate that the coverage rate is high in common entity names and terms of social sciences disciplines while that of text-specific word still needs to be improved. Finally, some advice is proposed for the future study of information extraction and alignment.

Most existing aspect-term level sentiment analysis (ATSA) approaches combined neural networks with attention mechanisms built upon given aspect to generate refined sentence representation for better predictions. In these methods, aspect terms are always provided in both training and testing process which may degrade aspect-level analysis into sentence-level prediction. However, the annotated aspect term might be unavailable in real-world scenarios which may challenge the applicability of the existing methods. In this paper, we aim to improve ATSA by discovering the potential aspect terms of the predicted sentiment polarity when the aspect terms of a test sentence are unknown. We access this goal by proposing a capsule network based model named CAPSAR. In CAPSAR, sentiment categories are denoted by capsules and aspect term information is injected into sentiment capsules through a sentiment-aspect reconstruction procedure during the training. As a result, coherent patterns between aspects and sentimental expressions are encapsulated by these sentiment capsules. Experiments on three widely used benchmarks demonstrate these patterns have potential in exploring aspect terms from test sentence when only feeding the sentence to the model. Meanwhile, the proposed CAPSAR can clearly outperform SOTA methods in standard ATSA tasks.

The subject of the study was the recognition of sounds of critical situations in the audio signal. The term “critical situation” is understood as an event, the characteristic sound signs of which can speak about acoustic artifacts as a shot, a scream, a glass crash, an explosion, a siren, etc.. The paper considers the scope of audio analytics, its advantages, the history of spectral analysis, as well as analyzes and selects tools for further development of system components. In the paper, we propose our dataset that consists of 14 classes that contains 1000 sounds of each, and a model to detect emergency situations using audio processing and analytics.

Aspect-Based Opinion Mining (ABOM) mainly focuses on mining the aspect terms (product’s features) and related opinion polarities (e.g., Positive, Negative, and Neutral) from user’s reviews. The most prominent neural network-based methods to perform ABOM tasks include BERT-based approaches, such as BERT-PT and BAT. These approaches build separate models to complete each ABOM subtasks, such as aspect term extraction (e.g., pizza, staff member) and aspect sentiment classification. Both approaches use different training algorithms, such as Post-Training and Adversarial Training. Also, the BERT-LSTM/Attention approach uses different pooling strategies on the intermediate layers of the BERT model to achieve better results. Moreover, they do not consider the subtasks of aspect categories (e.g., a category of aspect pizza in a review is food) and related opinion polarity. This paper proposes a new system for ABOM, called BERT-MTL, which uses Multi-Task Learning (MTL) approach and differentiates from these previous approaches by solving two tasks such as aspect terms and categories extraction simultaneously by taking advantage of similarities between tasks and enhancing the model’s accuracy as well as reduce the training time. Our proposed system also builds models to identify user’s opinions for aspect terms and aspect categories by applying different pooling strategies on the last layer of the BERT model. To evaluate our model’s performance, we have used the SemEval-14 task 4 restaurant dataset. Our model outperforms previous models in several ABOM tasks, and the experimental results support its validity.

Up to date, various neural-based methods have been proposed for joint mention span detection and coreference resolution. However, existing studies on coreference resolution mainly depend on mention representations, while the rest spans in the text are largely ignored and directly eliminated. In this paper, we aim at investigating whether those eliminated spans are totally worthless, or to what extent they can help improve the performance of coreference resolution. To achieve this goal, we propose to refine the representation of mentions with global spans including these eliminated ones leveraged. On this basis, we further introduce an additional loss term in this work to encourage the diversity between different entity clusters. Experimental results on the document-level CoNLL-2012 Shared Task English dataset show that the eliminated spans are indeed useful and our proposed approaches show promising results in coreference resolution.

PICO recognition is an information extraction task for detecting parts of text describing Participant (P), Intervention (I), Comparator (C), and Outcome (O) (PICO elements) in clinical trial literature. Each PICO description is further decomposed into finer semantic units. For example, in the sentence ‘The study involved 242 adult men with back pain.’, the phrase ‘242 adult men with back pain’ describes the participant, but this coarse-grained description is further divided into finer semantic units. The term ‘242’ shows “sample size” of the participants, ‘adult’ shows “age”, ‘men’ shows “sex”, and ‘back pain’ show the participant “condition”. Recognizing these fine-grained PICO entities in health literature is a challenging named-entity recognition (NER) task but it can help to fully automate systematic reviews (SR). Previous approaches concentrated on coarse-grained PICO recognition but focus on the fine-grained recognition still lacks. We revisit the previously unfruitful neural approaches to improve recognition performance for the fine-grained entities. In this paper, we test the feasibility and quality of multitask learning (MTL) to improve fine-grained PICO recognition using a related auxiliary task and compare it with single-task learning (STL). As a consequence, our end-to-end neural approach improves the state-of-the-art (SOTA) F1 score from 0.45 to 0.54 for the “participant” entity and from 0.48 to 0.57 for the “outcome” entity without any handcrafted features. We inspect the models to identify where they fail and how some of these failures are linked to the current benchmark data.

Aspect Sentiment Triplet Extraction (ASTE) aims to extract aspect term (aspect), sentiment and opinion term (opinion) triplets from sentences and can tell a complete story, i.e., the discussed aspect, the sentiment toward the aspect, and the cause of the sentiment. ASTE is a charming task, however, one triplet extracted by ASTE only includes one opinion of the aspect, but an aspect in a sentence may have multiple corresponding opinions and one opinion only provides part of the reason why the aspect has this sentiment, as a consequence, some triplets extracted by ASTE are hard to understand, and provide erroneous information for downstream tasks. In this paper, we introduce a new task, named Aspect Sentiment Multiple Opinions Triplet Extraction (ASMOTE). ASMOTE aims to extract aspect, sentiment and multiple opinions triplets. Specifically, one triplet extracted by ASMOTE contains all opinions about the aspect and can tell the exact reason that the aspect has the sentiment. We propose an Aspect-Guided Framework (AGF) to address this task. AGF first extracts aspects, then predicts their opinions and sentiments. Moreover, with the help of the proposed Sequence Labeling Attention (SLA), AGF improves the performance of the sentiment classification using the extracted opinions. Experimental results on multiple datasets demonstrate the effectiveness of our approach (Data and code can be found at https://github.com/l294265421/ASMOTE).

In the last decades, swarm intelligence algorithms have become a powerful tool for solving hard optimization problems. Nowadays numerous algorithms are proved to be good for different problems. With the overwhelming number of algorithms, it became hard for a common user to choose an appropriate method for solving a certain problem. To provide guidelines, it is necessary to classify optimization metaheuristics according to their capabilities. Deep statistical comparison represents a novel method for comparing and analyzing optimization algorithms. In this paper, the deep statistical comparison method was used for comparing different versions of the widely used fireworks algorithm. The fireworks algorithm was developed and improved in the last ten year, and this paper provides a theoretical analysis of five different versions, a cooperative framework for FWA, bare bones FWA, guided FWA, loser-out tournament based FWA, and dynamic search FWA. Based on the obtained results, the loser-out tournament based FWA has the best performance in the term of the solution quality, while the dynamic search FWA is the best in term of the solutions distribution in the search space.

Following the familiar distinction between software and hardware, this chapter argues that web archives deserve to be treated as a third category—memoryware: specific forms of preservation techniques which involve both software and hardware, but also crawlers, bots, curators, and users. While historically the term memoryware refers to the art of cementing together bits and pieces of sentimental objects to commemorate loved ones, understanding web archives as a complex socio-technical memoryware moves beyond their perception as bits and pieces of the live Web. Instead, understanding web archives as memoryware hints at the premise of the web’s exceptionalism in media and communication history and calls for revisiting some of the concepts and best practices in web archiving and web archive research that have consolidated over the years. The chapter, therefore, presents new challenges for web archive research by turning a critical eye on web archiving itself and on the specific types of histories that are constructed with web archives.

This Chapter shows how long-term raw vibration data could be analyzed. For this purpose, synthetic data generated in previous Chapter is used. Analysis methods include statistical techniques, narrowband spectral analysis, spectral comparison, and 3-dimensional data visualization. Each analysis is used to detect shaft imbalance, gearbox degradation, and REB inner race failure.

Deep learning models have achieved great success in recent years but progress in some domains like cybersecurity is stymied due to a paucity of realistic datasets. Organizations are reluctant to share such data, even internally, due to privacy reasons. An alternative is to use synthetically generated data but existing methods are limited in their ability to capture complex dependency structures, between attributes and across time. This paper presents STAN (Synthetic network Traffic generation with Autoregressive Neural models), a tool to generate realistic synthetic network traffic datasets for subsequent downstream applications. Our novel neural architecture captures both temporal dependencies and dependence between attributes at any given time. It integrates convolutional neural layers with mixture density neural layers and softmax layers, and models both continuous and discrete variables. We evaluate the performance of STAN in terms of quality of data generated, by training it on both a simulated dataset and a real network traffic data set. Finally, to answer the question—can real network traffic data be substituted with synthetic data to train models of comparable accuracy?—we train two anomaly detection models based on self-supervision. The results show only a small decline in accuracy of models trained solely on synthetic data. While current results are encouraging in terms of quality of data generated and absence of any obvious data leakage from training data, in the future we plan to further validate this fact by conducting privacy attacks on the generated data. Other future work includes validating capture of long term dependencies and making model training more efficient.

Student response systems (SRS) are a popular and effective tool to promote active learning on site, improving student engagement and attention, motivation and learning performance. Traditionally, SRS are designed for on-site settings. However, the safety measures in relation to the recent COVID-19 pandemic result in remote teaching at an unprecedented scale, with online courses becoming the rule. In this paper, we discuss the utilization of interactive SRS in such remote settings for which they initially were not designed. Over the last term, we conducted several empirical surveys across different groups of Computer Science students on undergraduate and postgraduate levels, and covering a broad age spectrum. The results indicate that, while common interactive features of videoconferencing tools, such as chat or polls, are well appreciated, there is still a need for dedicated SRS with game-based elements and feature sets beyond standard multiple-choice questions.

Several brain disorders can be detected by observing alterations in the brain’s structural and functional connectivities. Neurological findings suggest that early diagnosis of brain disorders, such as mild cognitive impairment (MCI), can prevent and even reverse its development into Alzheimer’s disease (AD). In this context, recent studies aimed to predict the evolution of brain connectivities over time by proposing machine learning models that work on brain images. However, such an approach is costly and time-consuming. Here, we propose to use brain connectivities as a more efficient alternative for time-dependent brain disorder diagnosis by regarding the brain as instead a large interconnected graph characterizing the interconnectivity scheme between several brain regions. We term our proposed method Recurrent Brain Graph Mapper (RBGM), a novel efficient edge-based recurrent graph neural network that predicts the time-dependent evaluation trajectory of a brain graph from a single baseline. Our RBGM contains a set of recurrent neural network-inspired mappers for each time point, where each mapper aims to project the ground-truth brain graph onto its next time point. We leverage the teacher forcing method to boost training and improve the evolved brain graph quality. To maintain the topological consistency between the predicted brain graphs and their corresponding ground-truth brain graphs at each time point, we further integrate a topological loss. We also use l1 loss to capture time-dependency and minimize the distance between the brain graph at consecutive time points for regularization. Benchmarks against several variants of RBGM and state-of-the-art methods prove that we can achieve the same accuracy in predicting brain graph evolution more efficiently, paving the way for novel graph neural network architecture and a highly efficient training scheme. Our RBGM code is available at https://github.com/basiralab/RBGM.

Existing Supervised Query Expansion (SQE) spends much time in term feature extraction but generates sub-optimal expanded terms. In this paper, we introduce Generative Adversarial Nets (GANs) and propose a GAN-based SQE method (SQE-GAN) to get helpful query expansion terms. We unify two types of models in query expansion: the generative model and the discriminative one. The generative (resp., discriminative) model focuses on predicting relevant terms (resp., relevancy) given a query (resp., a query-term pair). We iteratively optimize both models with a game between them. Besides, a BiLSTM layer is adopted to encode the utility of a term with respect to the query. As a result, the costly feature calculation in SQE schemes is avoided, such that the efficiency can be significantly improved. Moreover, by introducing GAN into expansion, the expanded terms are possible to be more effective with respect to the eventual needs of the user. Our experimental results demonstrate that SQE-GAN can be 37.3% faster than state-of-the-art SQE solutions while outperforming some recently proposed neural models in the retrieval quality.

Digital media have opened up new opportunities to review e-learning. The three ages of media describe a shift from technology for distance education, to technology-supported groups and, currently, technology for Community development. Reflecting on the profile of the e-tutor in the third phase of e-learning, our research was guided by three key-questions: what functions does the e-tutor fulfill? What tools does the e-tutor need? What actions should the e-tutor take care of, in term of communication? The research plan, carried out on 9 online courses, is based on quanti-qualitative method (a pre-course survey, a post-course survey, learning analytics, content analysis of tutor’s interactions and self-report). The results highlight the strategic role of the e-tutor in new contexts of learning.

Searching for suitable entities in a datasets is still a challenging task, because the entities’ attributes are often expressed by various data types including numerical, categorical, and fuzzy data. In addition, an attribute in a dataset may convey different data types for diverse records. In the query, the user may explain requirements by a different data type comparing with the one stored in a dataset, i.e. by linguistic term(s), whereas the respective attribute in a dataset is recorded as a real number and vice versa. Further, the user may provide complex preferences among atomic conditions. In this paper, we propose a robust framework capable to manage user requirements and match them with records in a dataset. The former is solved by the conformance measure, whereas for the latter different aggregation functions belonging to the conjunctive, averaging and hybrid classes have been suggested to cover particular aggregation needs like coalitions among atomic predicates and quantified conditions. The proposed method can be applied for selecting suitable records from any dataset containing numerical, categorical, fuzzy and binary data. The important characteristic of the presented method is the efficient applicability in the mentioned data collections, because conformance measure can manage different data types in the same manner. Finally, we discuss benefits, drawbacks and outline further activities.

User reviews contain rich information about user interests in items. Recently, many deep learning methods have attempted to integrate review contents into user preference rating prediction, helping to solve data sparseness problems. However, existing methods suffered from an inherent limitation that many reviews are noisy and contain non-consecutive viewpoints, besides, they are insufficient to capture inter-viewpoint relationships. Incorporating useful information is helpful for more accurate recommendations. In this paper, we propose a neural recommendation approach with a Diversity Penalty mechanism and Capsule Networks, named DPCN. Specifically, the diversity penalty component employs weight distributed matrices with the penalization term to capture different viewpoints in textual reviews. The capsule networks are designed to aggregate individual viewpoint vectors to form high-level feature representations for feature interaction. Then we combine the review feature representations with the user and item ID embedding for final rating prediction. Extensive experiments on five real-world datasets validate the effectiveness of our approach.

The use of traditional chalkboards, the illnesses that result from the use of chalk such as allergies, students who find it difficult to copy their lessons written by the teacher, make our daily classes lack intelligence, however, these difficulties manifest themselves in the difficult recognition of handwritten letters and words. Hence, we are discussing these current problems by providing future solutions that are smart, to use a new term it is "the smart classroom". This has prompted extensive research into the use of deep learning in the fields of computer vision and image processing based on artificial neural networks. In this article, a new approach to the recognition of Arabic handwritten words is presented. Certainly, the recognition of Arabic handwriting is a very recent research topic in computer vision, which has several attractive applications, in particular intelligent systems, video conferencing, real-time applications etc.… In our proposed approach, Deep Convolutional Neural Networks (DCNN) are adapted to perform the classification phase. Notably the architecture "ResNet", and "VGG16", using an enriched dataset containing images extracted from the IFN/ENIT database (The names of Tunisian villages). This technique has shown very good performance for the frequent recognition problems. According to the results obtained, the developed system gives very interesting recognition rates.

Aspect-based sentiment analysis is currently a main focus within the domain of sentiment analysis, whose target is to identify the sentiment polarities of specific aspect terms. The ongoing research is absent of exploiting the inter-aspect relationship while mainly focus on modeling the aspect terms and its context independently. To address this problem, we propose a model integrating the conjunction information and the sentiment of the preceding aspect term. As such, the inter-aspect relation between adjacent aspect terms can be precisely modeled and applied to sentiment classification. Experimental results on SemEval 2014 and MAMS show that our model outperform the baseline methods, especially dealing with the multi-aspect terms, which establishes a strong evidence of the effectiveness of the proposed method.