Search

Sentiment Analysis deals with analysing the reviews stated by its consumers for any product. If such analysis is performed at a deeper level, it enables us to identify the consumer’s sentiment towards each feature of the product. The sentiment expressed may not be same towards each feature. The analysis of this sort is called Aspect Based Sentiment Analysis (ABSA) and it has been sub-divided into four subtasks. In this paper, the detailed study of the approaches used for the first subtask of ABSA, i.e. Aspect Term Extraction (ATE) is presented. This paper discusses how ATE can be performed for the reviews in a rich morphological language, like Hindi. The models proposed for ATE of Hindi Reviews are Conditional Random Field (CRF) and Bidirectional Long-Short-Term-Memory (Bi-LSTM) models with novel architecture. The CRF based approach with novel feature, ‘Cluster-id’ improved F-measure from 41.07% to 42.71%. However, with 5-fold cross-validation, the CRF model attained an F-measure of 44.54%. By using our proposed Bi-LSTM based model with PoS vector, the F-measure obtained is 44.49%.

Speech summarization techniques take human speech as input and then output an abridged version as text or speech. Speech summarization has applications in many domains from information technology to health care, for example improving speech archives or reducing clinical documentation burden. This scoping review maps close to 2 decades of speech summarization literature, spanning from the early machine learning works up to ensemble models, with no restrictions on the language summarized, research method, or paper type. We reviewed a total of 110 papers out of a set of 188 found through a literature search and extracted speech features used, methods, scope, and training corpora. Most studies employ one of four speech summarization architectures: (1) Sentence extraction and compaction; (2) Feature extraction and classification or rank-based sentence selection; (3) Sentence compression and compression summarization; and (4) Language modelling. We also discuss the strengths and weaknesses of these different methods and speech features. Overall, supervised methods (e.g. Hidden Markov support vector machines, Ranking support vector machines, Conditional random fields) performed better than unsupervised methods. As supervised methods require manually annotated training data which can be costly, there was more interest in unsupervised methods. Recent research into unsupervised methods focusses on extending language modelling, for example by combining Uni-gram modelling with deep neural networks. This review does not include recent work in deep learning.

The current paper investigates the problem of multimodal named entity recognition from Twitter data. Named entity recognition (NER) is an important task in natural language processing and has been carefully studied in recent decades. NER from tweets is particularly challenging because of 1) tweets are limited in length, 2) contains noisy text, and 3) contains hashtags. Moreover often tweets are associated with images and hyperlinks. Existing works on tweet-NER mostly concentrate on multimodal deep learning based models neglecting the use of hand-crafted features and usage of hyperlinks. The current paper investigates the incorporation of hand-crafted features extracted from different modalities like images, hyperlinks while extracting named entities from tweet-text. A large set of hand-crafted features are extracted from different modalities (images, hyperlinks) and those are added with the features extracted by a hybrid deep-neural model, bi-directional LSTM and CNN, followed by a conditional random field to perform this task. Several variants of these models in association with different hand-crafted feature sets are designed. Extensive experimentations on a multimodal Twitter data (containing text, images and urls) illustrate that character level hand-crafted features significantly improve the performance of the systems. In a part of the paper, results of the proposed models are also shown on a standard NER dataset, CoNLL 2003 dataset.

The massive amount of data/information buried in textual bridge inspection reports open opportunities to leverage big data analytics for advanced information-rich bridge deterioration prediction. However, utilizing textual data for bridge deterioration prediction is challenging because of its inherently unstructured nature. To this end, this paper proposes an ontology-based information extraction (IE) framework that automatically recognizes and extracts key data/information from unstructured textual reports, and represents the extracted data/information in a structured way that is ready for data analytics. The proposed IE framework is composed of two primary components: (1) ontology-based sequence labelling for term identification, and (2) ontology-based dependency grammar for relationship association. This paper focuses on presenting the proposed sequence labelling methodology. The methodology utilizes ontology-based begin, inside, and outside (BIO) encoding for phrase-level segmentation and Conditional Random Field (CRF) for ontology-based labelling in both token and phrase levels. The experimental results showed that the proposed methodology has a precision of 97% and a recall of 91%.

This paper proposes a method for scene categorization by integrating region contextual information into the popular Bag-of-Visual-Words approach. The Bag-of-Visual-Words approach describes an image as a bag of discrete visual words, where the frequency distributions of these words are used for image categorization. However, the traditional visual words suffer from the problem when faced these patches with similar appearances but distinct semantic concepts. The drawback stems from the independently construction each visual word. This paper introduces Region-Conditional Random Fields model to learn each visual word depending on the rest of the visual words in the same region. Comparison with the traditional Conditional Random Fields model, there are two areas of novelty. First, the initial label of each patch is automatically defined based on its visual feature rather than manually labeling with semantic labels. Furthermore, the novel potential function is built under the region contextual constraint. The experimental results on the three well-known datasets show that Region Contextual Visual Words indeed improves categorization performance compared to traditional visual words.

Rockhead profile is an important part of geological profiles and can have significant impacts on some geotechnical engineering practice, and thus, it is necessary to establish a useful method to reverse the rockhead profile using site investigation results. As a general method to reflect the spatial distribution of geo-material properties based on field measurements, the conditional random field (CRF) was improved in this paper to simulate rockhead profiles. Besides, in geotechnical engineering practice, measurements are generally limited due to the limitations of budget and time so that the estimation of the mean value can have uncertainty to some extent. As the Bayesian theory can effectively combine the measurements and prior information to deal with uncertainty, CRF was implemented with the aid of the Bayesian framework in this study. More importantly, this simulation procedure is achieved as an analytical solution to avoid the time-consuming sampling work. The results show that the proposed method can provide a reasonable estimation about the rockhead depth at various locations against measurement data and as a result, the subjectivity in determining prior mean can be minimized. Finally, both the measurement data and selection of hyper-parameters in the proposed method can affect the simulated rockhead profiles, while the influence of the latter is less significant than that of the former.

This paper introduces a driving danger-level prediction system that uses multiple sensor inputs and statistical modeling to predict the driving risk. Three types of features were collected for the research, specifically the vehicle dynamic parameter, the driver's physiological data and the driver's behavior feature. To model the temporal patterns that lead to safe/dangerous driving state, several sequential supervised learning algorithms were evaluated in the paper, including hidden Markov model, conditional random field and reinforcement learning. Experimental results showed that using reinforcement learning based method with the vehicle dynamic parameters feature outperforms the rest algorithms, and adding the other two features could further improve the prediction accuracy. Based on the result, a live driving danger-level prediction prototype system was developed. Compared to many previous researches that focused on monitoring the driver's vigilance level to infer the possibility of potential driving risk, our live system is non-intrusive to the driver, and hence it is very desirable for driving danger prevention applications. Subjective on-line user study of our prototype system gave promising results.

Objective: Clinical records contain significant medical information that can be useful to researchers in various disciplines. However, these records also contain personal health information (PHI) whose presence limits the use of the records outside of hospitals. The goal of de-identification is to remove all PHI from clinical records. This is a challenging task because many records contain foreign and misspelled PHI; they also contain PHI that are ambiguous with non-PHI. These complications are compounded by the linguistic characteristics of clinical records. For example, medical discharge summaries, which are studied in this paper, are characterized by fragmented, incomplete utterances and domain-specific language; they cannot be fully processed by tools designed for lay language. Methods and results: In this paper, we show that we can de-identify medical discharge summaries using a de-identifier, Stat De-id, based on support vector machines and local context (F-measure=97% on PHI). Our representation of local context aids de-identification even when PHI include out-of-vocabulary words and even when PHI are ambiguous with non-PHI within the same corpus. Comparison of Stat De-id with a rule-based approach shows that local context contributes more to de-identification than dictionaries combined with hand-tailored heuristics (F-measure=85%). Comparison with two well-known named entity recognition (NER) systems, SNoW (F-measure=94%) and IdentiFinder (F-measure=36%), on five representative corpora show that when the language of documents is fragmented, a system with a relatively thorough representation of local context can be a more effective de-identifier than systems that combine (relatively simpler) local context with global context. Comparison with a Conditional Random Field De-identifier (CRFD), which utilizes global context in addition to the local context of Stat De-id, confirms this finding (F-measure=88%) and establishes that strengthening the representation of local context may be more beneficial for de-identification than complementing local with global context.

To process data like text and speech, Natural Language Processing (NLP) is a valuable tool. As on of NLP’s upstream tasks, sequence labeling is a vital part of NLP through techniques like text classification, machine translation, and sentiment analysis. In this paper, our focus is on sequence labeling where we assign semantic labels within input sequences. We present two novel frameworks, namely SA-CRFLV-I and SA-CRFLV-II, that use latent variables within random fields. These frameworks make use of an encoding schema in the form of a latent variable to be able to capture the latent structure in the observed data. SA-CRFLV-I shows the best performance at the sentence level whereas SA-CRFLV-II works best at the word level. In our in-depth experimental results, we compare our frameworks with 4 well-known sequence prediction methodologies which include NER, reference parsing, chunking as well as POS tagging. The proposed frameworks are shown to have better performance in terms of many well-known metrics.

Traditional object recognition algorithms are based on a commonly adopted closed-set hypothesis, assuming that the knowledge given in training is complete. However, real situations are often open and nonstatic, in which case the models only obtain incomplete knowledge during the training phase. This paper proposes a new type of conditional random field (CRF) model to solve a special case of incomplete knowledge, in which the visual appearance of certain objects changes significantly between training and testing, and as a result, certain unary features (features of individual objects) extracted from red green blue-depth (RGB-D) images are no longer reliable. Mirror nodes are introduced into the architecture based on the standard CRF model to build the mirrored conditional random field (Mirror-CRF) model, which integrates two types of object nodes: original nodes and mirror nodes. The mirror nodes have no unary features, only pairwise features, which describe relationships between two objects and are more reliable than unary features for object recognition in the case of appearance variation. The experimental results show that the Mirror-CRF model reduces the influence of significant changes in the appearance of certain objects and improves the object recognition ability under the condition of incomplete knowledge.

Many deep learning detectors have been proposed to address the scale variation issue in pedestrian detection. However, it is still far from being fully solved because most of these detectors handle the issue by directly fusing multiple extracted features with the weighted average or concatenation. Therefore, this paper designs a novel and effective anchor-free network (namely MPAF-Net) for multi-scale pedestrian detection. Specifically, we first improve Res2Net as our backbone network by designing a novel bottleneck block to make the proposed detector has a strong multi-scale representation ability for pedestrians. Furthermore, based on Conditional Random Fields (CRFs), the multi-scale features obtained from three different feature pyramid levels of backbone network are refined with a designed message passing mechanism. By doing so, the complementary information from the features at other scales can be dynamically passed to enhance the scale-specific feature. Finally, pedestrian detection is simplified as a straightforward center and scale prediction task by performing convolution on the concatenation of refined multi-scale convolution features. This way, the proposed MPAF-Net can be trained in an end-to-end fashion. Experimental results show that MPAF-Net achieves state-of-the-art performance on CityPersons, Caltech, and CrowdHuman benchmarks.

Recently, most companies market their products on the web to recognize their customers’ requirements and to improve their services’ quality according to the customers’ feedback and opinions. A huge amount of reviews and opinions are posted daily on products. Obtaining and quickly analyzing these opinions become a difficult task. These opinions might lead to a tendency or disinclination to a specific point of view. To represent the products’ opinions from customers’ perspectives, opinion retrieval becomes a demanding and essential task for automatically extracting, analyzing, and summarizing customers’ reviews. Usually, online products are offered by several suppliers in e-commerce. Therefore, to keep up the competitiveness among suppliers, the need for innovative requirements is required. This paper proposed an enhanced opinion retrieval approach depending on the explicit feature based opinion mining. The proposed approach expands the initial products’ requirements using extended heuristics and linguistic patterns of the Arabic opinions. Besides the relevant score, several factors, like features’ weight, the opinion importance, and the sentiment polarity are used to rank the retrieved results. The experimental results show the capability of the proposed approach to automatically extract more innovative features compared to the conditional random field (CRF) results.

Named Entity Recognition (NER) is a vital step in medical information extraction, especially Electronic Health Records (EHRs). Proper extraction of medical entities such as disease and medications can automate the process of EHR coding as well as considerably improve the filtering of EHR resulting in better extraction of medical information. NER systems are generally trained and evaluated on relatively small standard datasets. However, they are applied on real-world applications, they exposed to different collection of texts, varying in topic, entity distribution, and text type (e.g. abstract vs. full text). This mismatch between the internal structure and application can cause drop in performance and consequently, unreliability. In this paper, we propose Med-Flair, an NER tagger covering mainly multiple entity types, medications and diseases. Med-Flair is mainly based on the Flair NLP framework, in addition, it’s integrated by adding Bidirectional A Long Short Term Memory network (BiLSTM) and Conditional Random Fields (CRF) for sequence tagger. To validate the performance of Med-Flair, it is tested on 4 benchmark datasets, two for medications entities and two for diseases entities. Med Flair successfully achieves high performance, as it achieves 92%, 88%, 92% and 95% of F1-score which are mostly highest compared to state of the art deep neural network architectures such as BioBERT, DTranNER, BERT and BioNerFlair.

Natural language processing is a useful processing technique of language data, such as text and speech. Sequence labeling represents the upstream task of many natural language processing tasks, such as machine translation, text classification, and sentiment classification. In this paper, the focus is on the sequence labeling task, in which semantic labels are assigned to each unit of a given input sequence. Two frameworks of latent variable conditional random fields (CRF) models (called LVCRF-I and LVCRF-II) are proposed, which use the encoding schema as a latent variable to capture the latent structure of the hidden variables and the observed data. Among the two designed models, the LVCRF-I model focuses on the sentence level, while the LVCRF-II works in the word level, to choose the best encoding schema for a given input sequence automatically without handcraft features. In the experiments, the two proposed models are verified by four sequence prediction tasks, including named entity recognition (NER), chunking, reference parsing and POS tagging. The proposed frameworks achieve better performance without using other handcraft features than the conventional CRF model. Moreover, these designed frameworks can be viewed as a substitution of the conventional CRF models. In the commonly used LSTM-CRF models, the CRF layer can be replaced with our proposed framework as they use the same training and inference procedure. The experimental results show that the proposed models exhibit latent variable and provide competitive and robust performance on all three sequence prediction tasks.

Entity mentions usually contain other mention in the task of named entity recognition (NER). Nested entities pose challenge to the task of NER. Existing methods fail to sufficiently capture the boundaries information between nested entities, which limits the performance of the task. In this paper, we propose a dispatched attention neural model with multi-task learning for the task. In particular, given an input sentence, a bi-directional Long Short Term Memory (BiLSTM) encodes it as common contextualized hidden representation. Then position and syntax information are leveraged into attention network for capturing mention span features. The attention representation of each task is dispatched to subsequent task to exchange boundaries information for nested mentions. Finally, Conditional Random Fields (CRFs) are used to extract nested mentions in an inside-out order for each task. Results on ACE2005 and GENIA datasets show that the proposed model outperforms state-of-the-art systems, showing its effectiveness in detecting nested mentions.

Identifying named entities (NEs) present in electronic newspapers in regional languages is an important step in machine translation and summarization systems. In this paper, we propose a statistical named entity recognition system based on machine learning for the identification and classification of named entities present in Marathi language text. In our system, named entities are identified and classified using conditional random fields (CRFs). As being a morphologically rich language, statistical algorithms achieves good NE identification and classification accuracy but needs extra knowledge to improve accuracy. Experiments conducted on the FIRE-2010 corpus show that our system submitted for the challenge achieves the precision, recall and F1-measure of 82.33%, 70.68% and 75.51% under the CRF algorithm.

Advances in techniques for automated classification of point cloud data introduce great opportunities for many new and existing applications. However, with a limited number of labelled points, automated classification by a machine learning model is prone to overfitting and poor generalization. The present paper addresses this problem by inducing controlled noise (on a trained model) generated by invoking conditional random field similarity penalties using nearby features. The method is called Atrous XCRF and works by forcing a trained model to respect the similarity penalties provided by unlabeled data. In a benchmark study carried out using the ISPRS 3D labeling dataset, our technique achieves 85.0% in term of overall accuracy, and 71.1% in term of F1 score. The result is on par with the current best model for the benchmark dataset and has the highest value in term of F1 score. Additionally, transfer learning using the Bergen 2018 dataset, without model retraining, was also performed. Even though our proposal provides a consistent 3% improvement in term of accuracy, more work still needs to be done to alleviate the generalization problem on the domain adaptation and the transfer learning field.

Standard approaches to named entity recognition (NER) are based on sequential labeling methods, such as conditional random fields (CRFs), which label each word in a sentence and extract entities from them that correspond to named entities. With the extensive deployment of deep learning methods for sequential labeling tasks, state-of-the-art NER performance has been achieved on long short-term memory (LSTM) architectures using only basic features. In this paper, we address Korean NER tasks and propose an extension of a bidirectional LSTM CRF by investigating character-based representation. Our extension involves deploying a hybrid representation using ConvNet and LSTM for the sequential modeling of characters, namely a character-based LSTM-ConvNet hybrid representation. Using morphemes as processing units for bidirectional LSTM, we apply a proposed hybrid representation composed of morpheme vectors. Experimental results showed that the proposed LSTM-ConvNet hybrid representation yielded improvements over each single representation on standard Korean NER tasks.

Dialogue breakdown is a significant problem in conversational agents. Timely breakdown detection helps the agents quickly recover from mistakes, minimizing the impact on user experience. In this paper, we focus on two problems: variations in determining a response that breakdowns a conversation i.e., subjectivity, and variations in breakdown types due to designs of conversational agents, i.e., variationality. To address the subjectivity, which decreases the agreement rate among annotators, our methods detect a dialogue breakdown by ensembling detectors trained by different sets of annotators that are grouped using a clustering algorithm. To address the variationality, our methods apply two types of detector architectures to capture global and local breakdowns. The long short-term memory detector considers the global context and the convolutional neural networks detector is sensitive to the local characteristics. The ensemble of all detectors makes a final judgment. The results of the Japanese task in the Dialogue Breakdown Detection Challenge 3 (DBDC3) confirm that our approach significantly outperforms the baseline, which uses the conventional conditional random field. Detailed error analysis reveals that our encoders based on a convolutional neural network and a long short-term memory have different characteristics. It also confirms the effects of annotator clustering.

In this paper, a two-step automatic change detection chain for rapid flood mapping based on Sentinel-1 Synthetic Aperture Radar (SAR) data is presented. First, a reference image is selected from a set of potential image candidates via a Jensen-Shannon (JS) divergence-based index. Second, saliency detection is applied on log-ratio data to derive the prior probabilities of changed and unchanged classes for initializing the following expectation-maximization (EM) based generalized Gaussian mixture model (GGMM). The saliency-guided GGMM is capable of capturing the primary pixel-based change information and handling highly imbalanced datasets. A fully-connected conditional random field (FCRF) model, which takes long-range pairwise potential connections into account, is integrated to remove the ambiguities of the saliency-guided GGMM and to achieve the final change map. The whole process chain is automatic with an efficient computation. The proposed approach was validated on flood events at the Evros River, Greece and the Wharfe River and Ouse River in York, United Kingdom. Kappa coefficients (k) of 0.9238 and 0.8682 were obtained respectively. The sensitivity analysis underlines the robustness of the proposed approach for rapid flood mapping.

The automatic detection and accurate localization of landmarks is a crucial task in medical imaging. It is necessary for tasks like diagnosis, surgical planning, and post-operative assessment. A common approach to localize multiple landmarks is to combine multiple independent localizers for individual landmarks with a spatial regularizer, e.g., a conditional random field (CRF). Its configuration, e.g., the CRF topology and potential functions, often has to be manually specified w.r.t. the application. In this paper, we present a general framework to automatically learn the optimal configuration of a CRF for localizing multiple landmarks. Furthermore, we introduce a novel “missing” label for each landmark (node in the CRF). The key idea is to define a pool of potentials and optimize their CRF weights and the potential values for missing landmarks in a learning framework. Potentials with a low weight are removed, thus optimizing the graph topology. This allows to easily transfer our framework to new applications, and to integrate different localizers. Further advantages of our algorithm are its low test runtime, low amount of training data, and interpretability. We illustrate its feasibility in a detailed evaluation on three medical datasets featuring high degrees of pathologies and outliers.

Conditional random field model can make best use of limited site investigation data to properly characterize the spatial variation of soil properties. This paper aims to propose a simplified approach for generating conditional random fields of soil undrained shear strength. A numerical method is adopted to validate the effectiveness of the proposed approach. With the proposed approach, the analytical posterior statistics of spatially varying undrained shear strength conditioned on the known values at measurement locations can be obtained. The conditional random field model of undrained shear strength is constructed using the field vane shear test data at a site of the west side highway in New York and the probability of slope failure is estimated by subset simulation. A clay slope under undrained conditions is investigated as an example to illustrate the proposed approach. The effects of borehole location and borehole layout scheme on the slope reliability are addressed. The results indicate that the proposed approach not only can well incorporate the limited site investigation data into modelling of the actual spatial variation of soil parameters by conditional random fields, but also can capture the depth-dependent nature of soil properties. The realizations of conditional random fields generated by the proposed approach can be well constrained to the site investigation data.

The paper presents a method for reliability analysis of slopes in unsaturated soils. Conditional random fields are simulated by consideration possible fluctuations of the measured soil properties. To predict the unsaturated soil behaviour, suctions are estimated and implemented in a finite element analysis. Numerical results of a case study demonstrate that the use of conditional random fields, affects not only the standard deviation of the safety factor distribution, but can also cause considerable variation in the mean value. It is concluded that unsaturated conditions can result in higher slope reliability, but with potentially more extensive failure zones.

Context: Function Point Analysis (FPA) provides an objective, comparative measure for size estimation in the early stage of software development. When practicing FPA, analysts typically abide by the following steps: data function (DF) extraction, transactional function extraction, function type classification and adjustment factor determination. However, due to lack of approach and tool support, these steps are usually conduct by human efforts in practice. Related approaches can hardly be applied in the FPA due to the following three challenges, i.e., FPA rule-driven extraction, domain-specific parsing, and expensive labeled resources. Objective: In this paper, we aim to automate the extraction of DFs, which is the starting and fundamental step in FPA. Method: We propose an automated approach named DEX to extract data functions from textual requirements. Specifically, DEX introduces the popularly-used conditional random field (CRF) model to predict the boundary of a data function. Besides, DEX employs the bootstrapping-based algorithm and DF-oriented language model to further boost the performance. Results: We evaluate DEX from two aspects: evaluation on a real industrial dataset and a manual review by domain experts. The evaluation on the real industrial dataset shows that DEX could achieve 80% precision, 84% recall, and 82% F1, and outperforms three state-of-the-art baselines. The expert review suggests that DEX could increase 16% precision and 13% recall, compared with those produced by engineers. Conclusion: DEX could achieve promising results under a small number of labeled requirements and outperform the state-of-the-art approaches. Moreover, DEX could help engineers produce more accurate and complete DFs in the industrial environment.

Point cloud segmentation is the foundation of 3D environmental perception for modern intelligent systems. To solve this problem and image segmentation, conditional random fields (CRFs) are usually formulated as discrete models in label space to encourage label consistency, which is actually a kind of postprocessing. In this paper, we reconsider the CRF in feature space for point cloud segmentation because it can capture the structure of features well to improve the representation ability of features rather than simply smoothing. Therefore, we first model the point cloud features with a continuous quadratic energy model and formulate its solution process as a message-passing graph convolution, by which it can be easily integrated into a deep network. We theoretically demonstrate that the message passing in the graph convolution is equivalent to the mean-field approximation of a continuous CRF model. Furthermore, we build an encoder-decoder network based on the proposed continuous CRF graph convolution (CRFConv), in which the CRFConv embedded in the decoding layers can restore the details of high-level features that were lost in the encoding stage to enhance the location ability of the network, thereby benefiting segmentation. Analogous to the CRFConv, we show that the classical discrete CRF can also work collaboratively with the proposed network via another graph convolution to further improve the segmentation results. Experiments on various point cloud benchmarks demonstrate the effectiveness and robustness of the proposed method. Compared with the state-of-the-art methods, the proposed method can also achieve competitive segmentation performance., 20 pages + 8 pages (supplemental material)

The exploration of legal documents in the Brazilian Judiciary context lacks reliable annotated corpus to support the development of new Natural Language Process (NLP) applications. Therefore, this paper presents a step toward exploring legal decisions with Named Entity Recognition (NER) in the Brazilian Supreme Court (STF) context. We aim to present a case study on the fine-grained annotation task of legal decisions, performed by law students as annotators where two levels of nested legal entities were annotated. Nested entities mapped in a preliminary study composed of four coarser legal named entities and twenty-four nested ones (fine-grained). The final result is a corpus of 594 decisions published by the STF annotated by the 76 law students, those with the highest average inter-annotator agreement score. We also present two baselines for NER based on Conditional Random Fields (CRFs) and Bidirectional Long-Short Term Memory Networks (BiLSTMs). This corpus is the first of its kind, the most extensive corpus known in Portuguese dedicated for legal named entity recognition, open and available to better support further research studies in a similar context.

Gliomas are the most common types of brain tumors. Their shapes are irregular and their boundaries are ambiguous, producing a tumor that is hard to detect. To improve the segmentation performance, we propose a P-Norm-Based Glioma Segmentation by encapsulating it within a Deep Convolutional Autoencoder (DCA). In this paper, we first develop a new robust Deep P-norm Generative Model (DPGM) that can extract and classify the feature by the encoder section and segment this latter by the decoder section. After that, we utilise a Deep Discriminative Model (DDM) which applies the conditional random fields to refine the segmentation result. Effectiveness is gained by gradient descent training on the set of pre-P-norm filtering and thresolding convolutional operations using a novel geometric P-norm-based loss function. This latter will be optimized by the RANSAC layer, which is geared toward optimizing the strength of the correct model relative to the most convincing false model. We evaluate the accuracy and robustness of our method within MRI sequences (T1, T2, T2c & FLAIR) from the Multimodal Brain Tumor Image Segmentation Challenge (BRATS 2013 Leaderboard, BRATS 2013 Challenge, BRATS 2015 and BRATS 2017). The experiments demonstrate a high-quality segmentation results of our DPGM-DDM architecture and the average Dice is as high as 97.2%, 97.6%, and 98.4% respectively for complete, core, and enhancing tumor regions. Overall, our deep proposed architecture is effective in segmenting gliomas in multimodal images or fair images, and has the potential in routine examinations of gliomas in daily clinical practice.

Online reviews are an important asset for users who are deciding to buy a product, see a movie, or go to a restaurant and for managers who are making business decisions. The reviews from e-commerce websites are usually attached to ratings, which facilitates learning from the reviews by users. However, many reviews that spread across forums or social media are written in plain text, which is not rated, and these reviews are called non-rated reviews in this paper. From the perspective of sentiment analysis at the aspects level, this study develops a predictive framework for calculating ratings for non-rated reviews. The idea behind the framework began with an observation: the sentiment of an aspect is determined by its context; the rating of the review depends on the sentiment of the aspects and the number of positive and negative aspects in the review. Viewing term pairs that co-occur with aspects as their context, we conceived of a variant of a Conditional Random Field model, called SentiCRF, for generating term pairs and calculating their sentiment scores from a training set. Then, we developed a cumulative logit model that uses aspects and their sentiments in a review to predict the ratings of the review. In addition, we met the challenge of class imbalance when calculating the sentiment scores of term pairs. We also conceived of a heuristic re-sampling algorithm to tackle class imbalance. Experiments were conducted on the Yelp dataset, and their results demonstrate that the predictive framework is feasible and effective at predicting the ratings of reviews.

Part-of-speech (POS) tagging is a fundamental task in natural language processing (NLP). A robust POS tagger plays an important role in most NLP problems and applications, including syntactic parsing, semantic parsing, machine translation, and question answering. Although a lot of efficient POS taggers has been developed for general, conventional text, little work has been done for social media text. In this paper, we present an empirical study on POS tagging for Vietnamese social media text, which shows several challenges compared with tagging for general text. Social media text does not always conform to formal grammars and correct spelling. It also uses abbreviations, foreign words, and emoticons frequently. A POS tagger developed for conventional text would perform poorly on such noisy data. We address this problem by proposing a tagging model based on Conditional Random Fields (CRFs) with various kinds of features for Vietnamese social media text. We also investigate the effect of features extracted from word clusters under the Brown and canonical correlation analysis (CCA) based clustering in semi-supervised settings. We introduce an annotated corpus for POS tagging, which consists of more than four thousand sentences from Facebook, the most popular social network in Vietnam. Using this corpus, we performed a series of experiments to evaluate the proposed model. Our model achieved 88.26% and 88.92% tagging accuracy in supervised and semi-supervised scenarios, respectively, which are nearly 12% improvement over vnTagger, a state-of-the-art and most widely used Vietnamese POS tagger developed for general, conventional text. In addition, the semi-supervised model outperformed, in terms of accuracy, the version of vnTagger trained on the same Facebook dataset, showing the usefulness of word cluster features. 1

Road detection is one of the key challenges for autonomous vehicles. Two kinds of sensors are commonly used for road detection: cameras and LIDARs. However, each of them suffers from some inherent drawbacks. Thus, sensor fusion is commonly used to combine the merits of these two kinds of sensors. Nevertheless, current sensor fusion methods are dominated by either cameras or LIDARs rather than making the best of both. In this paper, we extend the conditional random field (CRF) model and propose a novel hybrid CRF model to fuse the information from camera and LIDAR. After aligning the LIDAR points and pixels, we take the labels (either road or background) of the pixels and LIDAR points as random variables and infer the labels via minimization of a hybrid energy function. Boosted decision tree classifiers are learned to predict the unary potentials of both the pixels and LIDAR points. The pairwise potentials in the hybrid model encode (i) the contextual consistency in the image, (ii) the contextual consistency in the point cloud, and (iii) the cross-modal consistency between the aligned pixels and LIDAR points. This model integrates the information from the two sensors in a probabilistic way and makes good use of both sensors. The hybrid CRF model can be optimized efficiently with graph cuts to get road areas. Extensive experiments have been conducted on the KITTI-ROAD benchmark dataset and the experimental results show that the proposed method outperforms the current methods.

Diagnosis of process operating modes is an important aspect of process monitoring. Due to its ability to model process transitions, the Hidden Markov Model (HMM) is widely used as a tool for operating mode diagnosis. However, it suffers from certain drawbacks due to its inherent assumptions. Hence, to address these issues and improve the operating mode diagnosis performance, we introduce the Conditional Random Field (CRF), which is a discriminiative probabilistic model based approach. Further, to deal with the missing measurement problem that commonly occurs in industrial datasets, a marginalized CRF framework is proposed in this paper and the related inference algorithms are developed under this newly designed framework. Validation studies performed on a simulated continuous stirred tank reactor (CSTR) system and an experimental hybrid tank system demonstrate that the proposed CRF based algorithms have superior performances compared to the existing approaches.

Natural language understanding tasks are key to extracting structured and semantic information from text. One of the most challenging problems in natural language is ambiguity and resolving such ambiguity based on context including temporal information. This paper, focuses on the task of extracting temporal roles from text, e.g. CEO of an organization or head of a state. A temporal role has a domain, which may resolve to different entities depending on the context and especially on temporal information, e.g. CEO of Microsoft in 2000. We focus on the temporal role extraction, as a precursor for temporal role disambiguation. We propose a structured prediction approach based on Conditional Random Fields (CRF) to annotate temporal roles in text and rely on a rich feature set, which extracts syntactic and semantic information from text. We perform an extensive evaluation of our approach based on two datasets. In the first dataset, we extract nearly 400k instances from Wikipedia through distant supervision, whereas in the second dataset, a manually curated ground-truth consisting of 200 instances is extracted from a sample of The New York Times (NYT) articles. Last, the proposed approach is compared against baselines where significant improvements are shown for both datasets.

One of the most active areas in computer vision is object detection, which has made significant improvement in recent years. Current state-of-the-art object detection methods mostly adhere to the framework of the regions with convolutional neural network (R-CNN). However, they only take advantage of the local appearance features inside object bounding boxes. Since these approaches ignore the contextual information around the object proposals, the outcome of these detectors may generate a semantically incoherent interpretation of the input image. In this paper, we propose a novel object detection system which incorporates the local appearance and the contextual information. Specifically, the contextual information comprises the relationships among objects and the global scene based contextual feature generated by a convolutional neural network. The whole system is formulated as a fully connected conditional random field (CRF) defined on object proposals. Then the contextual constraints among object proposals are modeled as edges naturally. Furthermore, a fast mean field approximation method is utilized to infer in this CRF model efficiently. The experimental results demonstrate that our algorithm achieves a higher mean average precision (mAP) on PASCAL VOC 2007 datasets compared with the baseline algorithm Faster R-CNN.

Site characterization, which aims to characterize the subsurface stratigraphic configuration and the associated geo-properties, has long been a significant challenge in geological and geotechnical practice. Due to the complexity and inherent spatial variability of the geological bodies and the limited availability of borehole data, uncertainty is unavoidable in the characterized subsurface stratigraphic configuration and the associated geo-properties. In previous studies, the stratigraphic uncertainty and the geo-properties uncertainty were characterized separately. This paper proposes a conditional random field approach for a coupled characterization of stratigraphic and geo-properties uncertainties. The spatial correlation of the stratum existence between different subsurface elements and the spatial correlation of geo-properties are characterized by two autocorrelation functions, determined with the maximum likelihood principle. With the knowledge of the spatial correlation of the stratum existence, the stratigraphic configuration can be sampled using a modified random field approach. Then, the spatial correlation of the geo-properties is updated based on the sampled stratigraphic configuration. With the updated spatial correlation of the geo-properties, the spatial distribution of the geo-properties can readily be simulated with the conditional random field theory. The effectiveness of the proposed approach is demonstrated through a case study of probabilistic site characterization of an offshore wind farm site in Taiwan. To extend the applicability of the proposed approach, a probabilistic evaluation of liquefaction potential at this site under a given seismic shaking level is performed.

Recent deep learning-based syllabification models generally give low error rates for high-resource languages with big datasets but sometimes produce high error rates for the low-resource ones. In this paper, two procedures: massive data augmentation and validation, are proposed to improve a deep learning-based syllabification, using a combination of bidirectional long short-term memory (BiLSTM), convolutional neural networks (CNN), and conditional random fields (CRF) for a low-resource Indonesian language. The massive data augmentation comprises four methods: transposing nuclei, swapping consonant-graphemes, flipping onsets, and creating acronyms. Meanwhile, the validation is implemented using a phonotactic-based scheme. A preliminary investigation on 50k Indonesian words informs that those augmentation methods significantly enlarge the dataset size by 12.8M valid words based on the phonotactic rules. An examination is then performed using 5-fold cross-validation. It reports that the augmentation methods significantly improve the BiLSTM-CNN-CRF model for 50k formal words and 100k named-entities datasets. A detailed investigation informs that augmenting the training set can reduce the word error rate (WER) coming from the long formal words and named entities., Creating acronyms; Indonesian orthographic syllabification; Flipping onsets; Transposing nuclei; Swapping consonant-graphemes

Recently, automatic computer-aided detection (CAD) of COVID-19 using radiological images has received a great deal of attention from many researchers and medical practitioners, and consequently several CAD frameworks and methods have been presented in the literature to assist the radiologist physicians in performing diagnostic COVID-19 tests quickly, reliably and accurately. This paper presents an innovative framework for the automatic detection of COVID-19 from chest X-ray (CXR) images, in which a rich and effective representation of lung tissue patterns is generated from the gray level co-occurrence matrix (GLCM) based textural features. The input CXR image is first preprocessed by spatial filtering along with median filtering and contrast limited adaptive histogram equalization to improve the CXR image's poor quality and reduce image noise. Automatic thresholding by the optimized formula of Otsu's method is applied to find a proper threshold value to best segment lung regions of interest (ROIs) out from CXR images. Then, a concise set of GLCM-based texture features is extracted to accurately represent the segmented lung ROIs of each CXR image. Finally, the normalized features are fed into a trained discriminative latent-dynamic conditional random fields (LDCRFs) model for fine-grained classification to divide the cases into two categories: COVID-19 and non-COVID-19. The presented method has been experimentally tested and validated on a relatively large dataset of frontal CXR images, achieving an average accuracy, precision, recall, and F1-score of 95.88%, 96.17%, 94.45%, and 95.79%, respectively, which compare favorably with and occasionally exceed those previously reported in similar studies in the literature.

In this paper, we introduce a simple unsupervised framework for pre-training hidden-unit conditional random fields (HUCRFs), i.e., learning initial parameter estimates for HUCRFs prior to supervised training.Our framework exploits the model structure of HUCRFs to make effective use of unlabeled data from the same domain or labeled data from a different domain. The key idea is to use the separation of HUCRF parameters between observations and labels: this allows us to pre-train observation parameters independently of label parameters. Pre-training is achieved by creating pseudo-labels from such resources. In the case of unlabeled data, we cluster observations and use the resulting clusters as pseudo-labels. Observation parameters can be trained on these resources and then transferred to initialize the supervised training process on the target labeled data. Experiments on various sequence labeling tasks demonstrate that the proposed pre-training method consistently yields significant improvement in performance. The core idea could be extended to other learning techniques including deep learning. We applied the proposed technique to recurrent neural networks (RNN) with long short term memory (LSTM) architecture and obtained similar gains.

Registration of three dimensional (3D) abdominal computed tomography (CT) scans is essential for computer-aided disease diagnosis and treatment, but the non-rigid respiratory movements of abdomen increase its difficulty. An automatic two-step multi-organ registration method is presented in this paper for abdominal CT scans. First, the lightweight squeeze-and-excitation (SE) attention blocks and the fully connected conditional random field (CRF)-based post-processing are integrated into a fully convolutional networks (FCN) based model, which can achieve more accurate segmentation results for abdominal multiple organs such as liver, kidneys, and spleen. Then, a non-rigid local correlation coefficient (LCC) similarity metric and an isotropic total variation regularization are combined to register the multi-organ regions, which can reduce computation time and avoid the over-smooth problem of deformation field. The proposed method is validated on three public abdominal CT databases, and the experimental results show that the segmentation and registration performances of our method outperform those of some competing methods. Moreover, compared with global registration strategy, the average registration time of the proposed method is shortened by 20.45%, and the average MSE, PSNR, SSIM, and DICE values are also improved by 7.21%, 22.37%, 15.80%, and 14.43%, respectively.

Effective upkeep of aging infrastructure systems with limited funding and resources calls for efficient bridge management systems. Although data-driven models have been extensively studied in the last decade for extracting knowledge from past experience to guide future maintenance decision making, their performance and usefulness have been limited by the level of detail and accuracy of currently available bridge condition databases. This paper leverages an untapped resource for bridge condition data and proposes a new method to extract condition information from it at a high level of detail. To that end, a natural language processing approach was developed to formalize structural condition knowledge by formulating a sequence labeling task and modeling inspection narratives as a combination of words representing defects, their severity and location, while accounting for the context of each word. The proposed framework employs a deep-learning-based approach and incorporates context-aware components including a bi-directional Long Short Term Memory (LSTM) neural network architecture and a Conditional Random Field (CRF) classifier to account for the context of words when assigning labels. A dependency-based word embedding model was also used to represent the raw text while incorporating both semantic and contextual information. The sequence labeling model was trained using bridge inspection reports collected from the Virginia Department of Transportation bridge inspection database and achieved an F1 score of 94.12% during testing. The proposed model also demonstrated improvements compared with baseline sequence labeling models, and was further used to demonstrate the capability of detecting condition changes with respect to previous inspection records. Results of this study show that the proposed method can be used to extract and create a condition information database that can further assist in developing data-driven bridge management and condition forecasting models, as well as automated bridge inspection systems.

Recently, the conditional random field (CRF) model has been greatly considered in synthetic aperture radar (SAR) image segmentation. This model not only directly considers the posterior distribution of the label field conditioned on images but also gives the interactions between the observations. In this paper, we propose a new CRF-based algorithm for SAR image segmentation. We consider the statistical approach jointly in feature extraction and similarity measurement in the proposed conditional random field model. Using the benefit of the 2-D wavelet transform, we define the generalized Gaussian distribution (GGD) on the wavelet coefficients to extract texture-based features. Then, to improve the CRF potential functions a new unary function is proposed which exactly matches the statistical properties of the wavelet coefficients and produces more accurate parameters for different regions. As the advantage of this function, it is no longer necessary to apply the multinomial logistic regression (MLR) model used in previous CRFs. Moreover, using the Kullback–Leibler distance (KLD) between distribution functions, the similarity measure in our pairwise potential is proposed very effectively and efficiently. The superiority of this scheme is that the similarity measure can be entirely computed using the parameters of the GGD that are typically of small size compared with the feature vectors in the previous methods. Comprehensive experiments on both synthetic and real SAR images indicate that our proposed algorithm achieves accuracy improvement in SAR image segmentation.

A large amount of detailed data about bridge conditions and maintenance actions are buried in bridge inspection reports without being used. Information extraction and data analytics open opportunities to leverage this wealth of data for improved bridge deterioration prediction and enhanced maintenance decision making. This paper proposes a novel ontology-based, semi-supervised conditional random fields (CRF)-based information extraction methodology for extracting information entities describing existing deficiencies and performed maintenance actions from bridge inspection reports. The ontology facilitates the analysis of the text based on content and domain-specific meaning. The proposed semi-supervised CRF simultaneously captures the dependency structures as well as the distributions of labeled and unlabeled data in a concave machine-learning function. It learns from a small set of fixed labeled data and, at the same time, dynamically adapts itself to unseen instances by further learning from a large set of unlabeled data for both reduced human effort and high performance. The proposed algorithm achieved an average precision, recall and, F-1 measure of 94.1%, 87.7%, and 90.7%, respectively.

Probabilistic Graphical Models (PGMs) in general, and Undirected Graphical Models (UGMs) in particular, become suitable frameworks to capture and conveniently model the uncertainty inherent in a variety of problems. When applied to real world applications, such as scene object recognition, they turn into a reliable and widespread resorted tool. The effectiveness of UGMs is tight to the particularities of the problem to be solved and, especially, to the chosen learning strategy. This paper presents a review of practical, widely resorted learning approaches for Conditional Random Fields (CRFs), the discriminate variant of UGMs, which is completed with a thorough comparison and experimental analysis in the field of scene object recognition. The chosen application for UGMs is of particular interest given its potential for enhancing the capabilities of cognitive agents. Two state-of-the-art datasets, NYUv2 and Cornell-RGBD, containing intensity and depth imagery from indoor scenes are used for training and testing CRFs. Results regarding success rate, computational burden, and scalability are analyzed, including the benefits of using parallelization techniques for gaining in efficiency. Cutting-edge strategies for learning UGMs are applied to scene object recognition.Two state-of-the-art datasets, NYUv2 and Cornell-RGBD, are used as testbed.Parallelization techniques are explored.The conducted analysis covers recognition success, computational burden, and scalability.

Existing salient object detection research generally focused on designing diverse saliency features and integrating them heuristically. In this paper, a novel salient object detection method is proposed by employing supervised training and contextual modeling. Gradient boosting decision trees are explored to aggregate features on segmented regions using a supervised training manner. Feature representation of hierarchically segmented regions is exploited to capture salient objects at different levels so as to extract discriminative features better. A region-level pairwise conditional random field (CRF) method is constructed to further boost the accuracy of saliency estimation as well as to improve the perceptual consistency of saliency maps. Experimental results demonstrate that the proposed method could achieve state-of-the-art performance over all public datasets. The F-measure is improved by 3.9 % , 13.0 % , 4.3 % on the MSRA-B, DUT-OMRON and HKU-IS dataset respectively, and the mean absolute error (MAE) is reduced by 31.6 % , 26.4 % and 21.2 % respectively on these three datasets.

Identifying expansion forms for acronyms is beneficial to many natural language processing and information retrieval tasks. In this work, we study the problem of finding expansions in texts for given acronym queries by modeling the problem as a sequence labeling task. However, it is challenging for traditional sequence labeling models like Conditional Random Fields (CRF) due to the complexity of the input sentences and the substructure of the categories. In this paper, we propose a Latent-state Neural Conditional Random Fields model (LNCRF) to deal with the challenges. On one hand, we extend CRF by coupling it with nonlinear hidden layers to learn multi-granularity hierarchical representations of the input data under the framework of Conditional Random Fields. On the other hand, we introduce latent variables to capture the fine granular information from the intrinsic substructures within the structured output labels implicitly. The experimental results on real data show that our model achieves the best performance against the state-of-the-art baselines.

This paper proposes a document indexing method for keyword spotting based on semi-Markov conditional random fields (semi-CRFs), which provide a theoretical framework for fusing the information of different contexts. The candidate segmentation-recognition lattice is first augmented based on the linguistic context to improve recognition results. For fast retrieval and to save storage space, the lattice is then purged by a forward-backward pruning procedure. In the reduced lattice, we estimate character similarity scores based on the semi-CRF model. The parameters of semi-CRF model are estimated using a binary classification objective, i.e., the cross-entropy (CE) to discriminate candidate characters in the lattice. To locate mis-recognized character instances in the lattice, we use confusing similar characters as proxies and search for proxy-characters in the index file. The proxy-character driven search can significantly improve the performance compared with our previous character-synchronous dynamic search (CSDS) method. Experimental results on the online handwriting database CASIA-OLHWDB justify the effectiveness of the proposed method.

As causality extraction from cases is essential for emergency causal learning, it serves as a foundation for follow-up emergency management. However, there remain barriers to break for applying the previous causality extraction methods to emergency management. The experience of emergency management inspires us that the cause of disasters should have existed in the time before the effect. Therefore, causality relations can be seen as distinct temporal relations. By utilizing the temporal characteristics of causality, this paper redefines the causality extraction as a special kind of temporality extraction and presents a method for extracting causality from emergency cases based on conditional random fields (CRFs). Then the task turns to be a sequence labeling process which can be solved by involving a CRFs model. Several typhoon-related emergency cases are chosen as the experimental dataset. To seek the impact of different features on the model performance, two feature templates are also chosen to train the model. The experimental results show that our approaches can not only deal with marked causal relations, but also work effectively on unmarked causal relations. Besides, the CRFs model can even extract causal relations between sentences.

Using contextual information for scene labeling has gained substantial attention in the fields of image processing and computer vision. In this paper, a fusion model using flexible segmentation graph (FSG) is presented to explore multi-scale context for scene labeling problem. Given a family of segmentations, the representation of FSG is established based on the spatial relationship of these segmentations. In the scenario of FSG, the labeling inference process is formulated as a contextual fusion model, trained from the discriminative classifiers. Compared to previous approaches, which usually employ Conditional Random Fields (CRFs) or hierarchical models to explore contextual information, our FSG representation is flexible and efficient without hierarchical constraint, allowing us to capture a wide variety of visual context for the task of scene labeling. Our approach yields state-of-the-art results on the MSRC dataset (21 classes) and the LHI dataset (15 classes), and near-record results on the SIFT Flow dataset (33 classes) and PASCAL VOC segmentation dataset (20 classes), while producing a 320?240 scene labeling in less than a second. A remarkable fact is that our approach also outperforms recent CNN-based methods. HighlightsWe propose a novel flexible segmentation graph (FSG) representation to capture multi-scale visual context for scene labeling problem.In the scenario of FSG, we establish a contextual fusion model to formulate multi-scale context.We have evaluated our method on four datasets. Our model yields similar to or better labeling result than competing models, and is also computationally efficient.

This paper proposes a novel method for simulating a three-dimensional (3D) random field conditional on site data. It is based on the assumption of separable auto-correlation in the vertical and horizontal directions. This novel method adopts special simulation techniques so that it can handle incomplete site data (e.g., missing data at some depths). Moreover, it can simulate a 3D conditional random field in a computationally efficient way without the need to invert and store large matrices. The proposed method simulates a 3D conditional random field with two steps. The first step simulates the missing site data to make the site data complete. The second step simulates the conditional random field at locations not explored by the soundings/boreholes. A simulated example is adopted to illustrate the effectiveness of the proposed method.

Over the past few years, Twitter has experienced massive growth and the volume of its online content has increased rapidly. This content has been a rich source for several studies that focused on natural language processing (NLP) research. However, Twitter data pose numerous challenges and obstacles to NLP tasks. For the English language, Twitter has an NLP tool that provides tweet-specific NLP tasks, which present significant opportunities for English NLP research and applications. Part-of-speech (POS) tagging for English tweets is one of the tasks that is offered and facilitated by such a tool. In contrast, only a few attempts have been made to develop POS taggers for Arabic content on Twitter. In this paper, we consider POS tagging, which is one of the NLP tasks that directly affects the performance of other subsequent text processing tasks. We introduce three manually annotated datasets for the POS tagging of Arabic tweets: the ‘Mixed,’ ‘MSA,’ and ‘GLF’ datasets with 3000, 1000, and 1000 Arabic tweets, respectively. In addition, we present an exploratory analysis of the behavior of using hashtags in Arabic tweets, which is a phenomenon that affects the task of POS tagging. We also present two supervised POS taggers that are developed based on two approaches: Conditional Random Fields and Bidirectional Long Short-Term Memory (Bi-LSTM) models. We conclude that the Bi-LSTM-based POS tagger achieves the state-of-the-art results for the ‘Mixed’ dataset with 96.5% accuracy. However, the specific-dialect taggers trained on the ‘MSA’ and ‘GLF’ datasets achieve an accuracy of 95.6% and 95%, respectively. The results for the ‘Mixed’ dataset indicate the effectiveness of developing a joint POS tagger without the need for a dialect-specific POS tagger.

Background: Knowledge is evolving over time, often as a result of new discoveries or changes in the adopted methods of reasoning. Also, new facts or evidence may become available, leading to new understandings of complex phenomena. This is particularly true in the biomedical field, where scientists and physicians are constantly striving to find new methods of diagnosis, treatment and eventually cure. Knowledge Graphs (KGs) offer a real way of organizing and retrieving the massive and growing amount of biomedical knowledge. Objective: We propose an end-to-end approach for knowledge extraction and analysis from biomedical clinical notes using the Bidirectional Encoder Representations from Transformers (BERT) model and Conditional Random Field (CRF) layer. Methods: The approach is based on knowledge graphs, which can effectively process abstract biomedical concepts such as relationships and interactions between medical entities. Besides offering an intuitive way to visualize these concepts, KGs can solve more complex knowledge retrieval problems by simplifying them into simpler representations or by transforming the problems into representations from different perspectives. We created a biomedical Knowledge Graph using using Natural Language Processing models for named entity recognition and relation extraction. The generated biomedical knowledge graphs (KGs) are then used for question answering. Results: The proposed framework can successfully extract relevant structured information with high accuracy (90.7% for Named-entity recognition (NER), 88% for relation extraction (RE)), according to experimental findings based on real-world 505 patient biomedical unstructured clinical notes. Conclusions:In this paper, we propose a novel end-to-end system for the construction of a biomedical knowledge graph from clinical textual using a variation of BERT models.