Manual Markup Research Articles

Methodology for extracting narratives from social media big data

Purpose: of the article is to present the experience in developing and testing the methodology for extracting a system of narratives on a socially significant phenomenon from authentic social network big data (using the example of narratives about COVID-19 vaccination in the Russian social network VKontakte during the pandemic).Methods: of automated data analysis were used by the tools of the PolyAnalyst analytical platform: topic modeling (PLSA method), text indexing algorithms with the sentence identification stage, clustering, data aggregation, data normalization, calculation of a quantitative index. The calculation of the measure of proximity of keywords using the Python, partial manual markup and data validation were also carried out.Results: 4.5 million messages relevant to the topic of COVID-19 vaccination published in VKontakte from 01.01.2020 to 01.03.2023 were reduced to 237 stable narratives. A popularity index was calculated for each narrative. For example, the following narrative turned out to be the most popular: “Employers put pressure on people to get vaccinated” (it was supported by 76,118 texts). As a result of the study, a dataset was obtained, including 237 narratives.Conclusions and Relevance: the developed toolkit is universal: the methodology can be adapted to any relevant topic, requiring only adjustments to the input parameters of thematic modeling. The obtained dataset is planned to be introduced into scientific circulation as an up-to-date material for studying public opinion on vaccination in Russia. The results contribute to international research on public opinion and communication in crises and can serve as a basis for practical actions aimed at improving the quality of public communications and decision-making at all levels of government.

Keyword Extraction from Kazakh News Dataset with BERT

Keywords provide a concise and precise description of the document's content. Due to the importance of the keyword and the difficulty of manual markup, automatic keyword extraction makes this process easy and fast. In this paper, Keyword Extraction from Kazakh News Dataset was presented. Model performance results were obtained by using the BERT base - uncased and BERT-base-multilingual-uncased pre-trained language model for the newly compiled Kazakh News Dataset-KND. Compiled Kazakh news data set consists of 7060 data. Data were collected from the web pages anatili.kazgazeta.kz, Bilimdinews.kz, and zhasalash.kz using the BeautifulSoap and Requests libraries. These web pages mostly contain news, history, and literary texts. The dataset includes the publication name or news title, the author of the publication or news subject, and the URL of the Kazakh news site. In the evaluation of the training results, it was observed that the BERT base-multilingual-uncased F-score performance was higher than the BERT model.

Differentiation of Livestock Internal Organs Using Visible and Short-Wave Infrared Hyperspectral Imaging Sensors.

Automatic identification and sorting of livestock organs in the meat processing industry could reduce costs and improve efficiency. Two hyperspectral sensors encompassing the visible (400–900 nm) and short-wave infrared (900–1700 nm) spectra were used to identify the organs by type. A total of 104 parenchymatous organs of cattle and sheep (heart, kidney, liver, and lung) were scanned in a multi-sensory system that encompassed both sensors along a conveyor belt. Spectral data were obtained and averaged following manual markup of three to eight regions of interest of each organ. Two methods were evaluated to classify organs: partial least squares discriminant analysis (PLS-DA) and random forest (RF). In addition, classification models were obtained with the smoothed reflectance and absorbance and the first and second derivatives of the spectra to assess if one was superior to the rest. The in-sample accuracy for the visible, short-wave infrared, and combination of both sensors was higher for PLS-DA compared to RF. The accuracy of the classification models was not significantly different between data pre-processing methods or between visible and short-wave infrared sensors. Hyperspectral sensors, particularly those in the visible spectrum, seem promising to identify organs from slaughtered animals which could be useful for the automation of quality and process control in the food supply chain, such as in abattoirs.

Hybrid Active Contour Mammographic Mass Segmentation and Classification

This research implements a novel segmentation of mammographic mass. Three methods are proposed, namely, segmentation of mass based on iterative active contour, automatic region growing, and fully automatic mask selection-based active contour techniques. In the first method, iterative threshold is performed for manual cropped preprocessed image, and active contour is applied thereafter. To overcome manual cropping in the second method, an automatic seed selection followed by region growing is performed. Given that the result is only a few images owing to over segmentation, the third method uses a fully automatic active contour. Results of the segmentation techniques are compared with the manual markup by experts, specifically by taking the difference in their mean values. Accordingly, the difference in the mean value of the third method is 1.0853, which indicates the closeness of the segmentation. Moreover, the proposed method is compared with the existing fuzzy C means and level set methods. The automatic mass segmentation based on active contour technique results in segmentation with high accuracy. By using adaptive neuro fuzzy inference system, classification is done and results in a sensitivity of 94.73%, accuracy of 93.93%, and Mathew’s correlation coefficient (MCC) of 0.876.

Automatic construction of the dialog tree based on unmarked text corpora in Russian

In this paper, we propose a method for automatically determining the structure of the tree and the key topics of nodes in the process of building a dialog tree based on unmarked text corpora. Building a dialog tree is one of the time-consuming tasks when creating an automatic dialog system and in most cases is performed on the basis of manual markup, which takes a lot of time and resources. The method of hierarchical clustering of dialogs takes into account the semantic proximity of messages, allows one to allocate a different number of nodes at each level of the hierarchy and limit the dialog tree in width and depth. The algorithm for constructing annotations of nodes of the dialog tree takes into account the hierarchy of topics by building thematic chains. The method is based on the complex use of natural language processing methods (tokenization, lemmatization, part-of-speech tagging, word embeddings, etc.), analysis of the main components to reduce the dimension and methods of cluster analysis. Experiments on constructing the structure of the dialog tree and annotating nodes have shown the great possibilities of the proposed method for constructing an automatic dialog tree. The recognition accuracy on the example of the reference dialog tree containing 13 nodes at the first level, 381 nodes at the second level and 299 nodes at the third level was 0.8, 0.7 and 0.5, respectively. Automatic construction of dialog trees can be in demand when developing automatic dialog systems and for improving the quality of generating answers to user questions.

Illumination estimation challenge: The experience of the first 2 years

AbstractIllumination estimation is the essential step of computational color constancy, one of the core parts of various image processing pipelines of modern digital cameras. Having an accurate and reliable illumination estimation is important for reducing the illumination influence on the image colors. To motivate the generation of new ideas and the development of new algorithms in this field, two challenges on illumination estimation were conducted. The main advantage of testing a method on a challenge over testing it on some of the known datasets is the fact that the ground‐truth illuminations for the challenge test images are unknown up until the results have been submitted, which prevents any potential hyperparameter tuning that may be biased. The First illumination estimation challenge (IEC#1) had only a single task, global illumination estimation. The second illumination estimation challenge (IEC#2) was enriched with two additional tracks that encompassed indoor and two‐illuminant illumination estimation. Other main features of it are a new large dataset of images (about 5000) taken with the same camera sensor model, a manual markup accompanying each image, diverse content with scenes taken in numerous countries under a huge variety of illuminations extracted by using the SpyderCube calibration object, and a contest‐like markup for the images from the Cube++ dataset. This article focuses on the description of the past two challenges, algorithms which won in each track, and the conclusions that were drawn based on the results obtained during the first and second challenge that can be useful for similar future developments.

Automatic Pulse Classification for Artefact Removal Using SAX Strings, a CENTER-TBI Study.

High-resolution, waveform-level data from bedside monitors carry important information about a patient's physiology but is also polluted with artefactual data. Manual mark-up is the standard practice for detecting and eliminating artefacts, but it is time-consuming, prone to errors, biased and not suitable for real-time processing.In this paper we present a novel automatic artefact detection technique based on a Symbolic Aggregate approXimation (SAX) technique which makes it possible to represent individual pulses as 'words'. It does that by coding each pulse with a specified number of letters (here six) from a predefined alphabet of characters (here six). The word is then fed to a support vector machine (SVM) and classified as artefactual or physiological.To define the universe of acceptable pulses, the arterial blood pressure from 50 patients was analysed, and acceptable pulses were manually chosen by looking at the average pulse that each 'word' generated. This was then used to train a SVM classifier. To test this algorithm, a dataset with a balanced ratio of clean and artefactual pulses was built, classified and independently evaluated by two observers achieving a sensitivity of 0.972 and 0.954 and a specificity of 0.837 and 0.837 respectively.

Metric Classification of Traumatic Brain Injury Epileptiform Activity from Electroencephalography Data

This paper considers the problem of epileptiform activity recognition in EEG. We conducted experiments on male rat before and after Traumatic Brain Injury (TBI). Experts in neurology performed a manual markup of signals as Epileptiform Discharges (ED) and Sleep Spindles (SS). We developed a proprietary Event Detection Algorithm based on time-frequency analysis of wavelet spectrograms. Feature space was based on Power Spectrum Density (PSD) and Frequency of signals, and each feature was assessed for importance of epileptic activity prediction. We used resulted predictors for training logistic regression model, which estimated features weights in probability of epilepsy function. Validation of proposed model was done by multiple train-test division. We shown that the accuracy of prediction is around 80%. Proposed Epilepsy Prediction Model, as well as Event Detection Algorithm, can be applied to identification of epileptiform activity in long term EEG records of rats before and after TBI. and analysis of disease dynamics.

Epileptiform Activity Detection and Classification Algorithms of Rats with Post-traumatic Epilepsy

In this paper, the problem of epileptiform activity in EEG of rats before and after Traumatic Brain Injury is considered. Experts in neurology performed a manual markup of signals as Epileptiform Discharges and Sleep Spindles. A proprietary Event Detection Algorithm based on time-frequency analysis of wavelet spectrograms was created. Feature space from PSD and Frequency of a detected event was created, and each feature was assessed for importance of epileptic activity prediction. Resulted predictors were used for training logistic regression model, which estimated features weights in probability of epilepsy function. Validation of proposed model was done on Monte-Carlo simulation of cross-validations. It was showed that the accuracy of prediction is around 80%. Proposed Epilepsy Prediction Model, as well as Event Detection Algorithm, can be applied to identification of epileptiform activity in long term records of rats and analysis of disease dynamics.

SDHK meets NER

Mass digitization of historical text sources opens new avenues for research in the humanities and social sciences, but also presents a host of new methodological challenges. Historical text collections become more accessible, but new research tools must also be put in place in order to fully exploit the new research possibilities emerging from having access to vast document collections in digital format. This paper highlights some of the conditions to consider when place names in an older source material, in this case medieval charters, are to be matched to geographical data. The Swedish National Archives make some 43,000 medieval letters available in digital form through an online search facility. The volume of the material is such that manual markup of names will not be feasible. In this paper, we present the material, discuss the promises for research of linking, e.g., place names to other digital databases, and report on an experiment where an off-the-shelf named-entity recognition system for modern Swedish is applied to this material.

An approach for EEG of post traumatic sleep spindles and epilepsy seizures detection and classification in rats

The electroencephalographic (EEG) features of post traumatic epilepsy (PTE) are analyzed in the paper. The proposed method allows detection and classification of sleep spindles and epilepsy seizures. The experiments were conducted on a laboratory rats before and after traumatic brain inquiry (TBI). In the introduction, the details of the experiment along with the information about manual markup are provided. In the first part, the new method of sleep spindles and epilepsy seizures detection is described. The method is based on the analysis of the wavelet spectrogram extrema. Moreover, the described procedure of background extraction and ridge segmentation helps to classify signals as epilepsy seizures and sleep spindles. In the second part, the information about the clustering is given. k-Means clustering of seizures and spindles was performed based on signals power and frequency. The results of the clustering, along with the research of TBI effect on the EEG, are provided in the third part. It was shown that PTE may be considered as the cause of the frequency variance among clusters of sleep spindles and epilepsy seizures.

Gold-standard ontology-based anatomical annotation in the CRAFT Corpus.

Gold-standard annotated corpora have become important resources for the training and testing of natural-language-processing (NLP) systems designed to support biocuration efforts, and ontologies are increasingly used to facilitate curational consistency and semantic integration across disparate resources. Bringing together the respective power of these, the Colorado Richly Annotated Full-Text (CRAFT) Corpus, a collection of full-length, open-access biomedical journal articles with extensive manually created syntactic, formatting and semantic markup, was previously created and released. This initial public release has already been used in multiple projects to drive development of systems focused on a variety of biocuration, search, visualization, and semantic and syntactic NLP tasks. Building on its demonstrated utility, we have expanded the CRAFT Corpus with a large set of manually created semantic annotations relying on Uberon, an ontology representing anatomical entities and life-cycle stages of multicellular organisms across species as well as types of multicellular organisms defined in terms of life-cycle stage and sexual characteristics. This newly created set of annotations, which has been added for v2.1 of the corpus, is by far the largest publicly available collection of gold-standard anatomical markup and is the first large-scale effort at manual markup of biomedical text relying on the entirety of an anatomical terminology, as opposed to annotation with a small number of high-level anatomical categories, as performed in previous corpora. In addition to presenting and discussing this newly available resource, we apply it to provide a performance baseline for the automatic annotation of anatomical concepts in biomedical text using a prominent concept recognition system. The full corpus, released with a CC BY 3.0 license, may be downloaded from http://bionlp-corpora.sourceforge.net/CRAFT/index.shtml. Database URL: http://bionlp-corpora.sourceforge.net/CRAFT/index.shtml

Bioluminescence-Based Tumor Quantification Method for Monitoring Tumor Progression and Treatment Effects in Mouse Lymphoma Models

Although bioluminescence imaging (BLI) shows promise for monitoring tumor burden in animal models of cancer, these analyses remain mostly qualitative. Here we describe a method for bioluminescence imaging to obtain a semi-quantitative analysis of tumor burden and treatment response. This method is based on the calculation of a luminoscore, a value that allows comparisons of two animals from the same or different experiments. Current BLI instruments enable the calculation of this luminoscore, which relies mainly on the acquisition conditions (back and front acquisitions) and the drawing of the region of interest (manual markup around the mouse). Using two previously described mouse lymphoma models based on cell engraftment, we show that the luminoscore method can serve as a noninvasive way to verify successful tumor cell inoculation, monitor tumor burden, and evaluate the effects of in situ cancer treatment (CpG-DNA). Finally, we show that this method suits different experimental designs. We suggest that this method be used for early estimates of treatment response in preclinical small-animal studies.

Bioluminescence-Based Tumor Quantification Method for Monitoring Tumor Progression and Treatment Effects in Mouse Lymphoma Models

Although bioluminescence imaging (BLI) shows promise for monitoring tumor burden in animal models of cancer, these analyses remain mostly qualitative. Here we describe a method for bioluminescence imaging to obtain a semi-quantitative analysis of tumor burden and treatment response. This method is based on the calculation of a luminoscore, a value that allows comparisons of two animals from the same or different experiments. Current BLI instruments enable the calculation of this luminoscore, which relies mainly on the acquisition conditions (back and front acquisitions) and the drawing of the region of interest (manual markup around the mouse). Using two previously described mouse lymphoma models based on cell engraftment, we show that the luminoscore method can serve as a noninvasive way to verify successful tumor cell inoculation, monitor tumor burden, and evaluate the effects of in situ cancer treatment (CpG-DNA). Finally, we show that this method suits different experimental designs. We suggest that this method be used for early estimates of treatment response in preclinical small-animal studies.

Neural network based automatic fingerprint recognition system for overlapped latent images

Automatic Fingerprint Recognition System (AFRS) is getting advanced as major distinct in the field of Biometrics. There are a number of difficult issues that need to be addressed in order to develop the scope for AFRS. In this regard designing challenges are non linear distortion, low quality image , segmentation, sensor noise, skin conditions, overlapping, inter class similarity, intra class variations and template aging. In crime scenes, the latent images can be merged with some background images or more number of fingerprint images from same person or different person can be overlapped. During investigation several possibilities are them to acquire damaged or un separated fingerprint image. The suspected criminals can’t be identified and recognized using such kind of images. In forensics, the matching accuracy of latent is extremely critical even if it involves some degree of manual intervention by latent examiners including manual markup. An overlapped fingerprint image must be able to split for fingerprint identification and recognition. This paper developed an algorithm to separate overlapping latent images. The proposed AFRS analyzes and design a fingerprint recognition system for overlapped latent images. The planned work is to formulate with accurate and fast data retrieval using one-to-N fingerprint identification for overlapped images. Extensive experiments are performed on the SLF databases, NIST SD27, FVC DB1, DB2 databases and evaluate rank-1 identification rate. The results show that the proposed system can separate overlapped fingerprint more accurately and robustly and it consequently improve the fingerprint recognition accuracy of AFRS.

English

Latent fingerprint matching has played a critical role in identifying suspects and criminals. However, compared torolled and plain fingerprint matching, latent fingerprint identification accuracy is much lower due to complex background noise,poor ridge quality and overlapping structured noise in latent images. Accordingly, manual markup of various features (e.g.,region of interest, singular points and minutiae) is typically necessary to extract reliable features from latents. To reduce thismarkup cost and to improve the consistency in feature markup, fully automatic and highly accurate latentmatching algorithms are needed. In this paper, we propose an automatic region segmentation algorithm whose goal is to separate the fingerprint region (region of interest) from background. It utilizes both ridge orientation and frequency features. The orientation tensor is used to obtain the symmetric patterns of fingerprint ridge orientation, and local Fourier analysis method is used to estimate the local ridge frequency of the latent fingerprint. Candidate fingerprint (foreground) regions are obtained for each feature type; an intersection of regions from orientation and frequency features localizes the true latent fingerprint regions. To verify the viability of the proposed region segmentation algorithm, we evaluated the segmentation results in two aspects: a comparison with the ground truth foreground and matching performance based on segmented region.

Segmentation and Enhancement of Latent Fingerprints: A Coarse to Fine Ridge Structure Dictionary.

Latent fingerprint matching has played a critical role in identifying suspects and criminals. However, compared to rolled and plain fingerprint matching, latent identification accuracy is significantly lower due to complex background noise, poor ridge quality and overlapping structured noise in latent images. Accordingly, manual markup of various features (e.g., region of interest, singular points and minutiae) is typically necessary to extract reliable features from latents. To reduce this markup cost and to improve the consistency in feature markup, fully automatic and highly accurate ("lights-out" capability) latent matching algorithms are needed. In this paper, a dictionary-based approach is proposed for automatic latent segmentation and enhancement towards the goal of achieving "lights-out" latent identification systems. Given a latent fingerprint image, a total variation (TV) decomposition model with L1 fidelity regularization is used to remove piecewise-smooth background noise. The texture component image obtained from the decomposition of latent image is divided into overlapping patches. Ridge structure dictionary, which is learnt from a set of high quality ridge patches, is then used to restore ridge structure in these latent patches. The ridge quality of a patch, which is used for latent segmentation, is defined as the structural similarity between the patch and its reconstruction. Orientation and frequency fields, which are used for latent enhancement, are then extracted from the reconstructed patch. To balance robustness and accuracy, a coarse to fine strategy is proposed. Experimental results on two latent fingerprint databases (i.e., NIST SD27 and WVU DB) show that the proposed algorithm outperforms the state-of-the-art segmentation and enhancement algorithms and boosts the performance of a state-of-the-art commercial latent matcher.

Unsupervised Learning for Syntactic Disambiguation

We present a methodology framework for syntactic disambiguation in natural language texts. The method takes advantage of an existing manually compiled non-probabilistic and non-lexicalized grammar, and turns it into a probabilistic lexicalized grammar by automatically learning a kind of subcategorization frames or selectional preferences for all words observed in the training corpus. The dictionary of subcategorization frames or selectional preferences obtained in the training process can be subsequently used for syntactic disambiguation of new unseen texts. The learning process is unsupervised and requires no manual markup. The learning algorithm proposed in this paper can take advantage of any existing disambiguation method, including linguistically motivated methods of filtering or weighting competing alternative parse trees or syntactic relations, thus allowing for integration of linguistic knowledge and unsupervised machine learning.

Unsupervised Learning for Syntactic Disambiguation

We present a methodology framework for syntactic disambiguation in natural language texts. The method takes advantage of an existing manually compiled non-probabilistic and non-lexicalized grammar, and turns it into a probabilistic lexicalized grammar by automatically learning a kind of subcategorization frames or selectional preferences for all words observed in the training corpus. The dictionary of subcategorization frames or selectional preferences obtained in the training process can be subsequently used for syntactic disambiguation of new unseen texts. The learning process is unsupervised and requires no manual markup. The learning algorithm proposed in this paper can take advantage of any existing disambiguation method, including linguistically motivated methods of filtering or weighting competing alternative parse trees or syntactic relations, thus allowing for integration of linguistic knowledge and unsupervised machine learning.

Unsupervised Learning for Syntactic Disambiguation

We present a methodology framework for syntactic disambiguation in natural language texts. The method takes advantage of an existing manually compiled non-probabilistic and non-lexicalized grammar, and turns it into a probabilistic lexicalized grammar by automatically learning a kind of subcategorization frames or selectional preferences for all words observed in the training corpus. The dictionary of subcategorization frames or selectional preferences obtained in the training process can be subsequently used for syntactic disambiguation of new unseen texts. The learning process is unsupervised and requires no manual markup. The learning algorithm proposed in this paper can take advantage of any existing disambiguation method, including linguistically motivated methods of filtering or weighting competing alternative parse trees or syntactic relations, thus allowing for integration of linguistic knowledge and unsupervised machine learning.