Adaptable Closed-Domain Question Answering Using Contextualized CNN-Attention Models and Question Expansion

While diverse question answering (QA) datasets have been proposed and contributed significantly to the development of deep learning models for QA tasks, the existing datasets fall short in two aspects. First, we lack QA datasets covering complex questions that involve answers as well as the reasoning processes to get them. As a result, the state-of-the-art QA research on numerical reasoning still focuses on simple calculations and does not provide the mathematical expressions or evidence justifying the answers. Second, the QA community has contributed a lot of effort to improve the interpretability of QA models. However, they fail to explicitly show the reasoning process, such as the evidence order for reasoning and the interactions between different pieces of evidence. To address the above shortcoming, we introduce NOAHQA, a conversational and bilingual QA dataset with questions requiring numerical reasoning with compound mathematical expressions. With NOAHQA, we develop an interpretable reasoning graph as well as the appropriate evaluation metric to measure the answer quality. We evaluate the state-of-the-art QA models trained using existing QA datasets on NOAHQA and show that the best among them can only achieve 55.5 exact match scores, while the human performance is 89.7. We also present a new QA model for generating a reasoning graph where the reasoning graph metric still has a large gap compared with that of humans, eg, 28 scores.

The development of Question Answering (QA) systems has become important because it reveals research issues that require insight from a variety of disciplines, including Artificial Intelligence, Information Extraction, Natural Language Processing, and Psychology. Our goal here is to develop a QA approach to answer yes/no questions relevant to civil laws in legal bar exams. A bar examination is intended to determine whether a candidate is qualified to practice law in a given jurisdiction. We have found that the development of a QA system for this task provides insight into the challenges of formalizing reasoning about legal text, and about how to exploit advances in computational linguistics. We separate our QA approach into two steps. The first step is to identify legal documents relevant to the exam questions; the second step is to answer the questions by analyzing the relevant documents. In our initial approach described here, the first step has been already solved for us: the appropriate articles for each question have been identified by legal experts. So here, we focus on the second task, which can be considered as a form of Recognizing Textual Entailment (RTE), where input to the system is a question sentence and its corresponding civil law article(s), and the output is a binary answer: whether the question sentence is entailed from the article(s). We propose a hybrid method, which combines simple rules and an unsupervised learning model using deep linguistic features. We first construct a knowledge base for negation and antonym words for the legal domain. We then identify potential premise and conclusion components of input questions and documents, based on text patterns and separating commas. We further classify the questions into easy and difficult ones, and develop a two-phase method for answering yes/no questions. We answer easy questions by negation/antonym detection. For more difficult questions, we adapt an unsupervised machine learning method based on morphological, syntactic, and lexical semantic analysis on identified premises and conclusions. This provides the basis to compare the semantic correlation between a question and a legal article. Our experimental results show reasonable performance, which improves the baseline system, and outperforms an SVM-based supervised machine learning model.

Millions of clinicians rely daily on automated preliminary electrocardiogram (ECG) interpretation. Critical comparisons of machine learning-based automated analysis against clinically accepted standards of care are lacking. To use readily available 12-lead ECG data to train and apply an explainability technique to a convolutional neural network (CNN) that achieves high performance against clinical standards of care. This cross-sectional study was conducted using data from January 1, 2003, to December 31, 2018. Data were obtained in a commonly available 12-lead ECG format from a single-center tertiary care institution. All patients aged 18 years or older who received ECGs at the University of California, San Francisco, were included, yielding a total of 365 009 patients. Data were analyzed from January 1, 2019, to March 2, 2021. A CNN was trained to predict the presence of 38 diagnostic classes in 5 categories from 12-lead ECG data. A CNN explainability technique called LIME (Linear Interpretable Model-Agnostic Explanations) was used to visualize ECG segments contributing to CNN diagnoses. Area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated for the CNN in the holdout test data set against cardiologist clinical diagnoses. For a second validation, 3 electrophysiologists provided consensus committee diagnoses against which the CNN, cardiologist clinical diagnosis, and MUSE (GE Healthcare) automated analysis performance was compared using the F1 score; AUC, sensitivity, and specificity were also calculated for the CNN against the consensus committee. A total of 992 748 ECGs from 365 009 adult patients (mean [SD] age, 56.2 [17.6] years; 183 600 women [50.3%]; and 175 277 White patients [48.0%]) were included in the analysis. In 91 440 test data set ECGs, the CNN demonstrated an AUC of at least 0.960 for 32 of 38 classes (84.2%). Against the consensus committee diagnoses, the CNN had higher frequency-weighted mean F1 scores than both cardiologists and MUSE in all 5 categories (CNN frequency-weighted F1 score for rhythm, 0.812; conduction, 0.729; chamber diagnosis, 0.598; infarct, 0.674; and other diagnosis, 0.875). For 32 of 38 classes (84.2%), the CNN had AUCs of at least 0.910 and demonstrated comparable F1 scores and higher sensitivity than cardiologists, except for atrial fibrillation (CNN F1 score, 0.847 vs cardiologist F1 score, 0.881), junctional rhythm (0.526 vs 0.727), premature ventricular complex (0.786 vs 0.800), and Wolff-Parkinson-White (0.800 vs 0.842). Compared with MUSE, the CNN had higher F1 scores for all classes except supraventricular tachycardia (CNN F1 score, 0.696 vs MUSE F1 score, 0.714). The LIME technique highlighted physiologically relevant ECG segments. The results of this cross-sectional study suggest that readily available ECG data can be used to train a CNN algorithm to achieve comparable performance to clinical cardiologists and exceed the performance of MUSE automated analysis for most diagnoses, with some exceptions. The LIME explainability technique applied to CNNs highlights physiologically relevant ECG segments that contribute to the CNN's diagnoses.

Business process modelling is an important part of system design. When designing or redesigning a business process, stakeholders specify, negotiate, and agree on business requirements to be satisfied, including non-functional requirements that concern the quality of the business process. This thesis addresses the question of how to specify and compute the quality of a business process, given the model that stakeholders use. The motivation for this thesis is the increasing importance of the quality of business processes. Knowing the quality of specific business processes enables stakeholders to judge if these processes need improvement. Knowing the quality of the constructs of those processes (viz., events, inputs, activities, and outputs) and the way they are structured enables a more detailed analysis of their shortcomings and provides a basis for the design of improvements. The research challenge of this thesis is grounded in the assumption that: “Organisations need an appropriate means to effectively compute achievement of their goals and objectives by their business processes.” Given this challenge, the main research question on which this thesis focuses is: “Can the quality of a business process be computed quantitatively at different levels of granularity?” The research objective is: “To develop frameworks, factors, and metrics for computing non-functional requirements (quality) of business processes quantitatively at different levels of granularity.” The outcomes of this thesis are: 1) BPIMM, a language-independent business process integrating meta-model, based on the concepts of seven mainstream business process modelling languages: BPMN, EPC, RAD, UML AD, SADT, IDEF0, and IDEF3. 2) BPC-QC (Business Process Concept - Quality Computation), an approach to quality computation at the lowest level of granularity of a business process. The approach consists of: i. BPC-QEF (Business Process Concept - Quality Evaluation Framework), a language-independent generic framework and algorithm to compute the quality of the constructs of a business process: event, input, activity, and output. ii. A set of business process quality dimensions and factors. The following quality dimensions are distinguished: performance, efficiency, reliability, recoverability, permissibility, and availability. Each dimension categorises different quality aspects in terms of factors. A non-exhaustive set of sixteen quantitative factors is provided. iii. Quality metrics for each of the quality factors, to facilitate a quantitative computation of the quality of a specific construct of a business process. 3) BP-QC (Business Process - Quality Computation), an approach to compute the quality at the highest level of granularity of a business process. The approach consists of: i. BP-CQCF (Business Process - Compositional Quality Computation Framework), a language-independent generic framework and algorithm to compute the quality of a business process as a whole, given the quality of its constructs. ii. A set of generic business process modelling patterns to decompose a business process into more succinct parts, namely: sequential, parallel with synchronisation, exclusive, inclusive, simple loop, and complex loop. iii. A set of over one hundred computational formulae. For each combination of modelling pattern and a quality factor, there is a formula to compute the quality. 4) AAV (Approach to Application and Validation), an evaluation plan to evaluate BPIMM, BPC-QC and BP-QC in practice, together with expert stakeholders. The plan consists of the units of measure, a measurement model, and a case study procedure. To evaluate the applicability of the contributions of this thesis to real world business needs, four case studies have been conducted in different environments: a Dutch educational institution, a global financial institution, an international financial service provider, and a Dutch research project on crisis management. Each of these case studies concerns a different, single business process. This thesis shows that: 1) A quality computation approach can be adopted independent of a business process modelling language. 2) Quantitative quality factors can be introduced specifically for the constructs of a business process. 3) Quantitative metrics and computational formulae can be developed for specific quality factors, allowing the computation of different aspects of the quality of a business process quantitatively at different levels of granularity. 4) An evaluation plan can be developed to evaluate the applicability of the contributions of this thesis (viz., BPIMM, BPC-QC, and BP-QC). The contributions of this thesis are designed to be beneficial to the areas of business and management, requirements engineering, software engineering, and business process modelling. In the areas of requirements engineering and software engineering, these contributions are intended to help practitioners to consider non-functional requirements at the earliest stage. In the area of business process modelling, information systems, service computing, and cloud computing, the contributions can be used for quality-driven modelling, design, and redesign. To conclude, knowing the quality value of a business process at different levels of granularity provides a basis for its improvement.

Judgments are often used in linguistic research. Not much is known, however, about the variation of such judgments within and between participants. From a usage-based perspective, variation might be expected: with judgments based in representations, and representations resulting from input and use, both inter- and intra-individual variation are likely. This study investigates the reliability of metalinguistic judgments, more specifically familiarity judgments, for Dutch prepositional phrases (e. g.,op de bank, ‘on the couch’). Familiarity judgments for 44 PPs offered in isolation and in a sentential context were given by 86 participants in two identical test sessions, using Magnitude Estimation. Aggregated scores (averaged over participants) are remarkably consistent (Pearson’sr=.97), and in part predicted by corpus frequencies. At the same time, there is considerable variation between and within participants. Context does not reduce this variation. We interpret both the stability and instability to be real reflections of language: a relatively stable system in a speech community consisting of speakers who are variable and forever changing. The results suggest that judgment data are informative at different levels of granularity. They call for more attention to individual variation and its underlying dynamics.

In traditional formal concept analysis, the attributes in the formal context are considered fixed. However, in the real world data set, attributes always have different levels of granularity, correspondingly, the derived concept lattice may reveal different information and patterns. Therefore, the capability to change the level of granularity of an attribute in formal concept analysis to capture relevant patterns in data is a natural requirement. In this paper, a theoretical study has been undertaken in multi-scale formal contexts, where attributes with different levels of granularity possess different attribute values. Two types of formal concepts, i.e., object-oriented and property-oriented multi-scale concepts, are introduced and studied in detail. The collection of object-oriented concept lattices and property-oriented concept lattices can be obtained at different granularity levels of attributes. It has been shown that the set of extents in the derived concept lattices increases when we choose to use a finer level of granularity. Moreover, a corresponding bidirectional approach to concept construction(i.e., from coarser to finer and from finer to coarser, respectively) is exhibited, and some characterization theorems have been obtained.

Cervical cancer remains a critical global health issue, responsible for over 600,000 new cases and 300,000 deaths annually. Pathological imaging of cervical cancer is a crucial diagnostic tool. However, distinguishing specific areas of cellular differentiation remains challenging because of the lack of clear boundaries between cells at various stages of differentiation. To address the limitations of conventional clinical and deep learning (DL) methods, we developed a mobile attention classification network (MacNet) with multiscale features, aiming to increase the accuracy of differentiation classification and quantitatively analyze cervical cancer cell differentiation. We investigated the application of MacNet for classifying non-background images into 3 stages of cervical cancer differentiation. The feature maps are processed through the Mobile Convolution Neural Network with Mobile Attention (MCMA) module, which integrates mobile convolutional blocks and mobile attention blocks. MacNet harnesses the benefits of the image pyramid structure and self-attention mechanism, enabling multiscale feature extraction and emulation of clinical pathologist analysis. The final prediction is generated by the adaptive fusion module, which aggregates features into a unified output. Comparative evaluations demonstrated that MacNet outperforms existing models. The proposed method achieved the best classification accuracy of 92.34% among all 7 DL-based models. Specifically, the result achieved by MacNet was 2.62% greater than that of Inception Version 3, 7.9% greater than that of vision transformer, 8.08% greater than that of the visual geometry group network, 3.21% greater than that of Densely Connected Convolutional Network, 2.85% greater than that of shifted window transformer (Swin transformer), 5.4% greater than that of Cross Stage Partial DarkNet, and 5.41% greater than that of Residual Neural Network. At the same time, MacNet also achieved superior results in recall, precision, and F1 score. We have proposed a lightweight neural network method that innovatively combines attention mechanisms with convolutional neural networks (CNNs) to efficiently utilize multiscale information from histopathological images. This integration enables the precise quantitative display of different stages of differentiation in cervical cancer. By doing so, our method not only enhances diagnostic accuracy but also provides clinicians with a more effective tool for faster and more reliable diagnosis, representing a significant advancement in the field of pathological imaging.

In this paper we present a system for automatic Arabic text diacritization using three levels of analysis granularity in a layered back off manner. We build and exploit diacritized language models (LM) for each of three different levels of granularity: surface form, morphologically segmented into prefix/stem/suffix, and character level. For each of the passes, we use Viterbi search to pick the most probable diacritization per word in the input. We start with the surface form LM, followed by the morphological level, then finally we leverage the character level LM. Our system outperforms all of the published systems evaluated against the same training and test data. It achieves a 10.87% WER for complete full diacritization including lexical and syntactic diacritization, and 3.0% WER for lexical diacritization, ignoring syntactic diacritization.

Data warehousing is widely used in industry for reporting and analysis of huge volumes of data at different levels of detail. In general, data warehouses use standard dimensional schema designs to organize their data. However, current data warehousing schema designs fall short in their ability to model the multi-granular data found in various real-world application domains. For example, modern farm equipment in a field produces massive amounts of data at different levels of granularity that has to be stored and queried. A study of the commonly used data warehousing schemas exposes the limitation that the schema designs are intended to simply store data at the same single level of granularity. This paper on the other hand, presents several extended dimensional data warehousing schema design alternatives to store both detail and aggregated data at different levels of granularity. The paper presents three solutions to design the time dimension tables and four solutions to design the fact tables. Moreover, each of these solutions is evaluated in different combinations of the time dimension and the fact tables based on a real world farming case study.KeywordsMulti-granular datadata warehousingdata warehousing schema design

Geo‐SOMis a useful geovisualization technique for revealing patterns in spatial data, but is ineffective in supporting interactive exploration of patterns hidden in differentGeo‐SOMsizes. Based on the divide and group principle in geovisualization, the article proposes a new methodology that combinesGeo‐SOMand hierarchical clustering to tackle this problem.Geo‐SOMwas used to “divide” the dataset into several homogeneous subsets; hierarchical clustering was then used to “group” neighboring homogeneous subsets for pattern exploration in different levels of granularity, thus permitting exploration of patterns at multiple scales. An artificial dataset was used for validating the method's effectiveness. As a case study, the rush hour motorcycle flow data inTaipeiCity,Taiwan were analyzed. Compared with the best result generated solely byGeo‐SOM, the proposed method performed better in capturing the homogeneous zones in the artificial dataset. For the case study, the proposed method discovered six clusters with unique data and spatial patterns at different levels of granularity, while the originalGeo‐SOMonly identified two. Among the four hierarchical clustering methods,Ward's clustering performed the best in pattern discovery. The results demonstrated the effectiveness of the approach in visually and interactively exploring data and spatial patterns in geospatial data.

Due to the rapid development of social media networks, as well as the recent advances on smartphone technologies, users stopped being only consumers and became data producers. It enabled a new form of sensing, engaging a variety of studies using Location-Based Social Networks (LBSN) data. For instance, in device-to-device(D2D) communication systems, the user location is required to make mobile data offloading possible. However, getting the right location is a challenging task, and the analyses of LBSN datasets provide large-scale studies at a low-cost. On the other hand, depending the the level of granularity used on the analyses, it may result in the success or failure of the prediction. Therefore, in this paper, we show the importance of combining the different levels of granularities for studying user behavior in order to obtain a robuster analysis.

It is challenging to detect emotions on user-generated contents (UGC) because it tends to have sparse emotional semantics, multiple emotions in the same text, and a fast update of emotional expression. Word embedding can extract high-level features in words to enrich their semantics. The convolutional neural networks (CNNs) can make model training more efficient to adapt to fast-changing emotional expressions, but its original pooling operation can not simultaneously filter out multiple emotional features that are beneficial to classification results, which is not suitable for UGC. In this paper, we propose a self-attentive convolutional neural networks (SACNNs) trained on top of pre-trained word vectors for emotion detection on UGC, which reserves different kinds of emotion information, avoids the loss of emotional aspects in the pooling process of CNNs structure, and increases the interpretability of the model by visualizing the extraction process of features. Meanwhile, the convergence speed of the model training is accelerated, and the model is updated in real time to detect emotions with rich and novel expressions. The proposed model combines the CNNs and the self-attention mechanism, the self-attention mechanism can select key emotional features after convolution. We evaluate the proposed model with two datasets from NLPCC 2014 and SemEval 2018 Task 1. The experimental results show that our model obtain significant performance than baselines in multi-label classification on UGC. In addition, the experimental results and the rationality of the self-attention mechanism are analyzed in detail, and the influential convolutional filter windows are visualized based on attention weights.

We regard answer extraction of Question Answering (QA) system as a classification problem, classifying answer candidate sentences into positive or negative. To confirm the feasibility of this new approach, we first extract features concerning question sentences and answer words from question answer pairs (QA pair), then we conduct experiments based on these features, using Maximum Entropy Model (MEM) as a Machine Learning (ML) technique. The first experiment conducted on the class-TIME_YEAR achieves 81.24% in precision and 78.48% in recall. The second experiment expanded to two other classes-OBJ_SUBSTANCE and LOC_CONTINENT also shows good performance.KeywordsQuery WordCombine FeatureQuestion AnsweringInterrogative WordMaximum Entropy ModelThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Abstract. In the wake of a volcanic eruption, the rapid assessment of building damage is paramount for effective response and recovery planning. Uncrewed aerial vehicles, UAVs, offer a unique opportunity for assessing damage after a volcanic eruption, with the ability to collect on-demand imagery safely and rapidly from multiple perspectives at high resolutions. In this work, we established a UAV-appropriate tephra fall building damage state framework and used it to label ∼50 000 building bounding boxes around ∼2000 individual buildings in 2811 optical images collected during surveys conducted after the 2021 eruption of La Soufrière volcano, St Vincent and the Grenadines. We used these labelled data to train convolutional neural networks (CNNs) for (1) building localisation (average precision equals 0.728) and (2) damage classification into two levels of granularity: no damage vs. damage (F1 score = 0.809) and moderate damage vs. major damage (F1 score = 0.838) (1 is the maximum obtainable for both metrics). The trained models were incorporated into a pipeline along with all the necessary image processing steps to generate spatial data (a georeferenced vector with damage state attributes) for rapid tephra fall building damage mapping. Using our pipeline, we assessed tephra fall building damage for the town of Owia, finding that 22 % of buildings that received 50–90 mm of tephra accumulation experienced at least moderate damage. The pipeline is expected to perform well across other volcanic islands in the Caribbean where building types are similar, though it would benefit from additional testing. Through cross-validation, we found that the UAV look angle had a minor effect on the performance of damage classification models, while for the building localisation model, the performance was affected by both the look angle and the size of the buildings in images. These observations were used to develop a set of recommendations for data collection during future UAV tephra fall building damage surveys. This is the first attempt to automate tephra fall building damage assessment solely using post-event data. We expect that incorporating additional training data from future eruptions will further refine our model and improve its applicability worldwide. To facilitate continued development and collaboration all trained models and the pipeline code can be downloaded from GitHub.

Accurate identification of small tea buds is a key technology for tea harvesting robots, which directly affects tea quality and yield. However, due to the complexity of the tea plantation environment and the diversity of tea buds, accurate identification remains an enormous challenge. Current methods based on traditional image processing and machine learning fail to effectively extract subtle features and morphology of small tea buds, resulting in low accuracy and robustness. To achieve accurate identification, this paper proposes a small object detection algorithm called STF-YOLO (Small Target Detection with Swin Transformer and Focused YOLO), which integrates the Swin Transformer module and the YOLOv8 network to improve the detection ability of small objects. The Swin Transformer module extracts visual features based on a self-attention mechanism, which captures global and local context information of small objects to enhance feature representation. The YOLOv8 network is an object detector based on deep convolutional neural networks, offering high speed and precision. Based on the YOLOv8 network, modules including Focus and Depthwise Convolution are introduced to reduce computation and parameters, increase receptive field and feature channels, and improve feature fusion and transmission. Additionally, the Wise Intersection over Union loss is utilized to optimize the network. Experiments conducted on a self-created dataset of tea buds demonstrate that the STF-YOLO model achieves outstanding results, with an accuracy of 91.5% and a mean Average Precision of 89.4%. These results are significantly better than other detectors. Results show that, compared to mainstream algorithms (YOLOv8, YOLOv7, YOLOv5, and YOLOx), the model improves accuracy and F1 score by 5-20.22 percentage points and 0.03-0.13, respectively, proving its effectiveness in enhancing small object detection performance. This research provides technical means for the accurate identification of small tea buds in complex environments and offers insights into small object detection. Future research can further optimize model structures and parameters for more scenarios and tasks, as well as explore data augmentation and model fusion methods to improve generalization ability and robustness.