Search

BACKGROUND There is an increasing body of research on the development of machine learning algorithms in the evaluation of online health educational resources for specific readerships. Machine learning algorithms are known for their lack of interpretability compared with statistics. Given their high predictive precision, improving the interpretability of these algorithms can help increase their applicability and replicability in health educational research and applied linguistics, as well as in the development and review of new health education resources for effective and accessible health education. OBJECTIVE Our study aimed to develop a linguistically enriched machine learning model to predict binary outcomes of online English health educational resources in terms of their easiness and complexity for international tertiary students. METHODS Logistic regression emerged as the best performing algorithm compared with support vector machine (SVM) (linear), SVM (radial basis function), random forest, and extreme gradient boosting on the transformed data set using L2 normalization. We applied recursive feature elimination with SVM to perform automatic feature selection. The automatically selected features (n=67) were then further streamlined through expert review. The finalized feature set of 22 semantic features achieved a similar area under the curve, sensitivity, specificity, and accuracy compared with the initial (n=115) and automatically selected feature sets (n=67). Logistic regression with the linguistically enhanced feature set (n=22) exhibited important stability and robustness on the training data of different sizes (20%, 40%, 60%, and 80%), and showed consistently high performance when compared with the other 4 algorithms (SVM [linear], SVM [radial basis function], random forest, and extreme gradient boosting). RESULTS We identified semantic features (with positive regression coefficients) contributing to the prediction of easy-to-understand online health texts and semantic features (with negative regression coefficients) contributing to the prediction of hard-to-understand health materials for readers with nonnative English backgrounds. Language complexity was explained by lexical difficulty (rarity and medical terminology), verbs typical of medical discourse, and syntactic complexity. Language easiness of online health materials was associated with features such as common speech act verbs, personal pronouns, and familiar reasoning verbs. Successive permutation of features illustrated the interaction between these features and their impact on key performance indicators of the machine learning algorithms. CONCLUSIONS The new logistic regression model developed exhibited consistency, scalability, and, more importantly, interpretability based on existing health and linguistic research. It was found that low and high linguistic accessibilities of online health materials were explained by 2 sets of distinct semantic features. This revealed the inherent complexity of effective health communication beyond current readability analyses, which were limited to syntactic complexity and lexical difficulty.

BACKGROUND Current health information understandability research uses medical readability formulas to assess the cognitive difficulty of health education resources. This is based on an implicit assumption that medical domain knowledge represented by uncommon words or jargon form the sole barriers to health information access among the public. Our study challenged this by showing that, for readers from non-English speaking backgrounds with higher education attainment, semantic features of English health texts that underpin the knowledge structure of English health texts, rather than medical jargon, can explain the cognitive accessibility of health materials among readers with better understanding of English health terms yet limited exposure to English-based health education environments and traditions. OBJECTIVE Our study explores multidimensional semantic features for developing machine learning algorithms to predict the perceived level of cognitive accessibility of English health materials on health risks and diseases for young adults enrolled in Australian tertiary institutes. We compared algorithms to evaluate the cognitive accessibility of health information for nonnative English speakers with advanced education levels yet limited exposure to English health education environments. METHODS We used 113 semantic features to measure the content complexity and accessibility of original English resources. Using 1000 English health texts collected from Australian and international health organization websites rated by overseas tertiary students, we compared machine learning (decision tree, support vector machine [SVM], ensemble tree, and logistic regression) after hyperparameter optimization (grid search for the best hyperparameter combination of minimal classification errors). We applied 10-fold cross-validation on the whole data set for the model training and testing, and calculated the area under the operating characteristic curve (AUC), sensitivity, specificity, and accuracy as the measurement of the model performance. RESULTS We developed and compared 4 machine learning algorithms using multidimensional semantic features as predictors. The results showed that ensemble tree (LogitBoost) outperformed in terms of AUC (0.97), sensitivity (0.966), specificity (0.972), and accuracy (0.969). Decision tree (AUC 0.924, sensitivity 0.912, specificity 0.9358, and accuracy 0.924) and SVM (AUC 0.8946, sensitivity 0.8952, specificity 0.894, and accuracy 0.8946) followed closely. Decision tree, ensemble tree, and SVM achieved statistically significant improvement over logistic regression in AUC, specificity, and accuracy. As the best performing algorithm, ensemble tree reached statistically significant improvement over SVM in AUC, specificity, and accuracy, and statistically significant improvement over decision tree in sensitivity. CONCLUSIONS Our study shows that cognitive accessibility of English health texts is not limited to word length and sentence length as had been conventionally measured by medical readability formulas. We compared machine learning algorithms based on semantic features to explore the cognitive accessibility of health information for nonnative English speakers. The results showed the new models reached statistically increased AUC, sensitivity, and accuracy to predict health resource accessibility for the target readership. Our study illustrated that semantic features such as cognitive ability–related semantic features, communicative actions and processes, power relationships in health care settings, and lexical familiarity and diversity of health texts are large contributors to the comprehension of health information; for readers such as international students, semantic features of health texts outweigh syntax and domain knowledge.

BACKGROUND Chronic Obstructive Pulmonary Disease (COPD) is the third leading cause of death in China and has caused serious affect to health and life quality. However, the status stage of a patient is difficult to be accurately assessed because of dynamic changes in the condition and complex risk factors. A rapid and accurate methods to predict disease stage of COPD patients is of great significance. OBJECTIVE This study aims to explore an enhanced recurrent convolutional neural networks model for predicting correct staging of patients with COPD in China for assistant disease prevention and treatment. METHODS Data was collected from The First Affiliate Hospital of Guangzhou Medical University, which had standardized disease registration and follow-up management for 5108 patients with COPD. Our enhanced recurrent convolutional neural network consists of a bidirectional LSTM layer, a convolutional layer, a max-pooling layer, and an output layer. RESULTS The model proposed was evaluated on the real-world clinical dataset of 5108 COPD patients to predict the state stage of the disease. The performance of the proposed model achieved 93.2% in terms of accuracy, outperforming a list of baseline models. CONCLUSIONS This paper proposes an enhanced recurrent convolutional neural network model which is experimented on a real-world clinical dataset containing around 5,000 patients with COPD. The proposed model achieves the best performance on all evaluation metrics indicating its feasibility in predicting the state stage of diseases.

In a document reranking task, pre-trained language models such as BERT have been successfully applied due to their powerful capability in extracting informative features from queries and candidate answers. However, these language models always generate discriminative features and pay less attention to generalized features which contain shared information of query-answer pairs to assist question answering. In this paper, we propose a BERT-based model named FABERT by integrating both discriminative features and generalized features produced by a gradient reverse layer into one answer vector with an attention mechanism for document reranking. Extensive experiments on the MS MARCO passage ranking task and TREC Robust dataset show that FABERT outperforms baseline methods including a feature projection method which projects existing feature vectors into the orthogonal space of generalized feature vector to eliminate common information of generalized feature vectors.

The task of entity linking aims to correctly link mentions in a text fragment to a reference knowledge base. Most existing methods apply single neural network model to learn semantic representations on all granularities in contextual information, which neglecting the trait of different granularities. Also, these solely representation-based methods measure the semantic matching based on the abstract vector representation that frequently miss concrete matching information. To better capture contextual information, this paper proposes a new neural network model called Hybrid Semantic Matching (HSM) for the entity linking task. The model captures two different aspects of semantic information via representation and interaction-based neural semantic matching models. Furthermore, to consider the global consistency of entities, a recurrent random walk is applied to propagate entity linking evidences among related decisions. Evaluation was conducted on three publicly available standard datasets. Results show that our proposed HSM model is more effective compared with a list of baseline models.

Automatic color matching is an essential technology for dyeing and printing in the textiles industry to ensure customer quality standards and increase production efficiency. Computer-assisted automatic color matching has played an increasingly important role in textiles dyeing and printing as the cycle time of fashion decreases, with consumers constantly demanding new styles. Machine learning has been widely applied in various fields, however there are still difficulties such as the time-consuming and labor-intensive procedure of manual feature extraction required by conventional machine learning models. For dyeing and printing, the large amount of dyeing recipe data involving dye concentrations, spectra, and substrate materials have made it difficult to build artificial neural network models in complex data fitting. This paper proposes an automatic color matching prediction model, CMR-Color, by incorporating three neural network models including typical CNN, MLP, and ResNet to improve the capability of extracting high-dimensional features from spectral data. It selects the dyes required and respective concentration values of each component in the recipe. Moreover, it uses feature fusion to improve the prediction performance of automatic color matching. The experiment datasets of 72,132 recipes are from a world-class listed company in the textiles industry. The results show that the CMR-Color model achieves the best performance in all three evaluation metrics compared with four state-of-the-art baseline methods, verifying its effectiveness in color matching prediction for textiles dyeing and printing.

Text-guided image manipulation aims to modify the visual attributes of images according to textual descriptions. Existing works either mismatch between generated images and textual descriptions or may pollute the text-irrelevant image regions. In this paper, we propose a dilated residual aggregation network (denoted as DRA) for text-guided image manipulation, which exploits a long-distance residual with dilated convolutions (RD) to aggregate the encoded visual content and style features and the textual features of the guiding descriptions. In particular, the dilated convolutions increase the receptive field without sacrificing spatial resolutions of intermediate features, benefiting to reconstructing the texture details matching with the textual descriptions. Furthermore, we propose an attention-guided injection module (AIM) to inject textual semantics into feature maps of DRA without polluting the text-irrelevant image regions by combining triplet attention mechanism and central biasing instance normalization. Quantitative and qualitative experiments conducted on the CUB-200-2011 and Oxford-102 datasets demonstrate the superior performance of the proposed DRA.

In recent years, the prevalence of diabetes has been increasing rapidly worldwide. With the advancement of information technology, automated question-answering services for healthcare, which are commonly based on annotated corpus in health domain, have positive effects on health knowledge spread and daily health management for high-risk populations. This paper proposes to construct a large scale diabetes corpus of frequently-asked questions for automated question-answering services and evaluations. Concentrating on the characteristics of diabetes-related factors that reflect conditions of diabetes, this work establishes an annotated dataset containing professional question & answer pairs about diabetes and their annotated question target categories. The corpus is applicable for various question-answering applications, supporting users to retrieve needed information, arrange diets, adhere to scientific medication as well as prevent and control disease complications.

UNSTRUCTURED With the rapid growth of information technology, the necessity for processing substantial amounts of health data using advanced information technologies is increasing. A large amount of valuable data exists in natural text such as diagnosis text, discharge summaries, online health discussions, and eligibility criteria of clinical trials. Health natural language processing, as an interdisciplinary field of natural language processing and health care, plays a substantial role in a wide scope of both methodology development and applications. This editorial shares the most recent methodology innovations of health natural language processing and applications in the medical domain published in this JMIR Medical Informatics special theme issue entitled "Health Natural Language Processing: Methodology Development and Applications".

This book constitutes the refereed proceedings of the 15th Information Retrieval Technology Conference, AIRS 2019, held in Hong Kong, China, in November 2019.The 14 full papers presented together with 3 short papers were carefully reviewed and selected from 27 submissions. The scope of the conference covers applications, systems, technologies and theory aspects of information retrieval in text, audio, image, video and multimedia data.

Classroom dialogue, characterized by its interactive features and verbal encounters between teachers and students or among students, has been commonly used in teaching and learning. This paper conducted a systematic review of the coding frameworks used in the examination of classroom dialogue. We discussed over three main issues relating to the development of coding frameworks: linguistic ethnography versus sociocultural approach, coding units and levels, and objects of coding. The review indicates that there are six themes that a dialogic framework should encapsulate in its categories, which are prior knowledge, personal information, analysis, generalization, speculation and uptakes. With this knowledge, scholars and practitioners would become more competent in designing or selecting frameworks.

Online examination has been used more and more widely in the education and other fields due to its advantages of efficiency, convenience, and fairness. This paper proposes a new web-based online examination system, which utilizes PHP, Ajax and other technologies to implement online examination functions, including testing question collection, user management, online testing, real-time score calculation, answer checking, and result analysis. Particularly, an automated test paper generation module was designed and integrated. The system has been applied to a course involving more than 1000 students per semester at Guangzhou University of Foreign Studies. It has been proved to save efforts of teachers and students, demonstrating its effectiveness in assisting teaching and learning.

Information and communication technologies is being heralded as a catalyst for educational innovations. Artificial intelligence (AI) assisted user modeling has attracted great increasing interests from the academia with a growing research articles available. In this article, a bibliometric analysis of scientific literature concerning AI assisted user modeling was carried out. 333 articles from Web of Science were retrieved and analyzed to comprehensively understand trends and developments of the research field. Specifically, we analyzed the articles in terms of article count and citation count, influential journals, subjects, authors, and keyword occurrence. Finally, special attention was paid to the study of leading countries/regions and institutions. Findings of this work are useful in helping scholars as well as practitioners better understand the development trend of research of AI assisted user modeling, as well as being more aware of the research hotspots.

Question text similarity calculation is a fundamental and essential research problem for community question answering services. Different question text collections have various characteristics. Some frequently answered questions may have distinct statistical patterns, while some questions are syntactically different but semantically similar. To measure question similarity more adaptively to different kinds of question text, this paper proposes a method for identifying similar question utilizing the combination of both statistic and semantic features based on XGBoost. The method extracts semantic and statistical features from question text. After that, a feature set generation method is proposed, along with a model fusion strategy. Based on the standard Yahoo! dataset containing 25,569 questions with answers, three experiments have been conducted to evaluate the performance of the method. Results show that it achieves a precision of 88.65% and a recall of 71.85% outperforming a list of baseline methods.

Knowledge graph embedding technique aims to represent elements in knowledge graph, such as entities and relations, with numerical embedding vectors in semantic spaces. In general, an existing knowledge graph has relatively stable number of entities and directional relations before being updated. Though existing research has utilized relations of entities for link predication in knowledge graph, the relational directivity feature has not been fully exploited. Therefore, this paper proposes a bi-directional relation aware network (BDRAN) for representation learning, mining information based on directivity of relations in existing knowledge graphs. BDRAN leverages an encoder to capture features of entities in different patterns with diverse directional relations in entity representation level and semantic representation level. Besides, decoder is used to simulate interactions between entities and relations for precise representation learning. Experiments are conducted with widely used standard datasets including WN18RR, FB15k-237, NELL-995 and Kinship. The results present the improvement of BDRAN on the datasets, demonstrating the effectiveness of our model for link prediction.

User intent classification plays a critical role in identifying the interests of users in question-answering and spoken dialog systems. The question texts of these systems are usually short and their conveyed semantic information are frequently insufficient. Therefore, the accuracy of user intent classification related to user satisfaction may be affected. To address the problem, this paper proposes a hybrid neural network named RBERT-C for text classification to capture user intent. The network uses the Chinese pre-trained RoBERTa to initialize representation layer parameters. Then, it obtains question representations through a bidirectional transformer structure and extracts essential features using a Convolutional Neural Network after question representation modeling. The evaluation is based on the publicly available dataset ECDT containing 3736 labeled sentences. Experimental result indicates that our model RBERT-C achieves a F1 score of 0.96 and an accuracy of 0.96, outperforming a number of baseline methods.

Classroom dialogue is a common strategy for teaching and learning. Technology-assisted classroom dialogue has drawn increasing attentions, where the classification of classroom dialogue is one of active research topics. However, existing studies mainly paid much attention to dialogue manners rather than dialogue contexts. This paper conducts a deep learning-based experiment on the classification of classroom dialogue context in text format. A hybrid neural network-based model namely CNN-BiLSTM-Attention is proposed for context classification of classroom dialogue text. The hybrid model consists of a Convolutional Neural Network (CNN) and a Bidirectional Long Short-Term Memory Network (BiLSTM) by leveraging an attention mechanism. The CNN-BiLSTM-Attention model is able to capture and learn both the local and global features of classroom dialogue texts for learning semantic information of dialogue contexts. To test the effectiveness of the model, an annotated classroom dialogue text dataset is built based on a well-established coding framework through collecting 155 lessons in Chinese language. Compared with eleven baseline methods, including commonly-used machine learning models and deep learning models, the evaluation results demonstrate that the CNN-BiLSTM-Attention model achieves the best performance with an overall F1-score of 0.7006.

Text mining has become an increasingly significant role in processing medical information. The research of text mining enhanced medical has attracted much attention in view from the substantial expansion of literature. This study aims to systematically review the existing academic research outputs of the field from Web of Science and PubMed by using techniques such as geographic visualization, collaboration degree, social network analysis, and topic modeling analysis. Specifically, publication statistical characteristics, geographical distribution, collaboration relations, and research topic are quantitatively analyzed. This study contributes to the text mining enhanced medical research field in a number of ways. First, it provides the latest research status for researchers who are interested in the field through literature analysis. Second, it helps scholars become more aware of the research subfields through hot topic identification. Third, it provides insights to researchers engaging in the field and motivates attention on the relevant research.

With the rapid growth of information technology, the necessity for processing substantial amounts of health data using advanced information technologies is increasing. A large amount of valuable data exists in natural text such as diagnosis text, discharge summaries, online health discussions, and eligibility criteria of clinical trials. Health natural language processing, as an interdisciplinary field of natural language processing and health care, plays a substantial role in a wide scope of both methodology development and applications. This editorial shares the most recent methodology innovations of health natural language processing and applications in the medical domain published in this JMIR Medical Informatics special theme issue entitled "Health Natural Language Processing: Methodology Development and Applications".

We aimed to develop machine learning classifiers as a risk-prevention mechanism to help medical professionals with little or no knowledge of the patient’s languages in order to predict the likelihood of clinically significant mistakes or incomprehensible MT outputs based on the features of English source information as input to the MT systems. A MNB classifier was developed to provide intuitive probabilistic predictions of erroneous health translation outputs based on the computational modelling of a small number of optimised features of the original English source texts. The best performing multinominal Naïve Bayes classifier (MNB) using a small number of optimised features (8) achieved statistically higher AUC (M = 0.760, SD = 0.03) than the classifier using high-dimension natural features (135) (M = 0.631, SD = 0.006, p &lt, 0.0001, SE = 0.004) and the automatically optimised classifier (22) (M = 0.7231, SD = 0.0084, p &lt, 0.0001, SE = 0.004). Furthermore, MNB (8) had statistically higher sensitivity (M = 0.885, SD = 0.100) compared with the full-feature classifier (135) (M = 0.577, SD = 0.155, p &lt, 0.0001, SE = 0.005) and the automatically optimised classifier (22) (M = 0.731, SD = 0.139, p &lt, 0.0001, SE = 0.0023). Finally, MNB (8) reached statistically higher specificity (M = 0.667, SD = 0.138) compared to the full-feature classifier (135) (M = 0.567, SD = 0.139, p = 0.0002, SE = 0.026) and the automatically optimised classifier (22) (M = 0.633, SD = 0.141, p = 0.0133, SE = 0.026).

Answer type classification is a vital step of question answering systems to detect the most suitable target answer type. Highly accurate identification and classification of an answer type can help identify users’ question targets and filter out irrelevant candidate answers to improve system performances. This paper proposes a novel hybrid approach, named as ATICM, for automated answer type identification and classification by utilizing both syntactic and semantic analysis. We firstly propose to integrate four strategies to identify question target features by using dependency relations and rules. Afterwards, we leverage semantic relations to expand the extracted features. Our experiment datasets are publicly available UIUC and TREC10 annotated question datasets. The result shows the ATICM approach achieves an accuracy of 93.9% on the UIUC dataset and 92.8% on the TREC10 dataset. The performance outperforms the state-of-the-art baseline methods, demonstrating its effectiveness in answer type classification.

Transfer learning techniques have been broadly applied in applications where labeled data in a target domain are difficult to obtain while a lot of labeled data are available in related source domains. In practice, there can be multiple source domains that are related to the target domain, and how to combine them is still an open problem. In this paper, we seek to leverage labeled data from multiple source domains to enhance classification performance in a target domain where the target data are received in an online fashion. This problem is known as the online transfer learning problem. To achieve this, we propose novel online transfer learning paradigms in which the source and target domains are leveraged adaptively. We consider two different problem settings: homogeneous transfer learning and heterogeneous transfer learning. The proposed methods work in an online manner, where the weights of the source domains are adjusted dynamically. We provide the mistake bounds of the proposed methods and perform comprehensive experiments on real-world data sets to demonstrate the effectiveness of the proposed algorithms.

Automatic extraction of large-scale and accurate drug–disease pairs from the medical literature plays an important role for drug repurposing. However, many existing extraction methods are mainly in a supervised manner. It is costly and time-consuming to manually label drug–disease pairs datasets. There are many drug–disease pairs buried in free text. In this work, we first leverage a pattern-based method to automatically extract drug–disease pairs with treatment and inducement relationships from free text. Then, to reflect a drug–disease relation, a network embedding algorithm is proposed to calculate the degree of correlation of a drug–disease pair. In the experiments, we use the method to extract treatment and inducement drug–disease pairs from 27 million medical abstracts and titles available on PubMed. We extract 138,318 unique treatment pairs and 75,396 unique inducement pairs. Our algorithm achieves a precision of 0.912 and a recall of 0.898 in extracting the frequent treatment drug–disease pairs, and a precision of 0.923 and a recall of 0.833 in extracting the frequent inducement drug–disease pairs. Besides, our proposed information network embedding algorithm can efficiently reflect the degree of correlation of drug–disease pairs. Our algorithm can achieve a precision of 0.802, a recall of 0.783 in the fine-grained evaluation of extracting frequent pairs.

In a question-answering system, learner generated content including asked and answered questions is a meaningful resource to capture learning interests. This paper proposes an approach based on question topic mining for revealing learners' concerned topics in real community question-answering systems. The authors' approach firstly preprocesses all questions associated with learners. Afterwards, it analyzes each question with text features and generates a weight feature matrix using a revised TF/IDF method. In order to decrease the sparsity issue of data distribution, the authors employ three concept-mapping strategies including named entity recognition, synonym extension, and hyponym replacement. Applying an SVM classifier, their approach categorizes user questions into representative topics. Three experiments are conducted based on a TREC dataset and an actual dataset containing 1,120 questions posted by learners from a commercial question-answering community. Results demonstrate the effectiveness of the method compared with conventional classifiers as baselines.

Imbalanced dataset is a common issue in many applications. The one-class Support Vector Machine (SVM) is found to be an effective algorithm to construct classification models over the underlying imbalanced dataset. In some cases, feature extraction is hard and one would prefer using pre-defined kernels to train the model. In traditional practice, a valid kernel has to satisfy the Mercer's condition, which may restrict the design of kernel functions or matrices. In this paper, an indefinite kernel extension is applied to the one-class SVM model in order to relieve such limitation. To illustrate its performance, the algorithm is applied to perform virtual screening of drugs.

This study used Amazon Mechanical Turk to crowdsource public opinions about sharing medical records for clinical research. The 1,508 valid respondents comprised 58.7% males, 54% without college degrees, 41.5% students or unemployed, and 84.3% under 40 years old. More than 74% were somewhat willing to share de-identified records. Education level, employment status, and gender were identified as significant predictors of willingness to share one’s own or one’s family’s medical records (partially identifiable, completely identifiable, or de-identified). Thematic analysis applied to respondent comments uncovered barriers to sharing, including the inability to track uses and users of their information, potential harm (such as identity theft or healthcare denial), lack of trust, and worries about information misuse. Our study suggests that implementing reliable medical record de-identification and emphasizing trust development are essential to addressing such concerns. Amazon Mechanical Turk proved cost-effective for collecting public opinions with short surveys.

Background Social media plays a more and more important role in the research of health and healthcare due to the fast development of internet communication and information exchange. This paper conducts a bibliometric analysis to discover the thematic change and evolution of utilizing social media for healthcare research field. Methods With the basis of 4361 publications from both Web of Science and PubMed during the year 2008–2017, the analysis utilizes methods including topic modelling and science mapping analysis. Results Utilizing social media for healthcare research has attracted increasing attention from scientific communities. Journal of Medical Internet Research is the most prolific journal with the USA dominating in the research. Overly, major research themes such as YouTube analysis and Sex event are revealed. Themes in each time period and how they evolve across time span are also detected. Conclusions This systematic mapping of the research themes and research areas helps identify research interests and how they evolve across time, as well as providing insight into future research direction. Electronic supplementary material The online version of this article (10.1186/s12911-019-0757-4) contains supplementary material, which is available to authorized users.

Background: The identification of medical entities and relations from electronic medical records is a fundamental research issue for medical informatics. However, the task of extracting valuable knowledge from these records is challenging due to its high complexity. The accurate identification of entity and relation is still an open research problem in medical information extraction. Methods: A pattern-based method for extracting certain tumor-related entities and attributes from Chinese unstructured diagnostic imaging text is proposed. This method is a composition of three steps. Firstly, an algorithm based on keyword matching is designed to obtain the primary sites of tumors. Then a set of regular expressions is applied to identify primary tumor size information. Finally, a set of rules is defined to acquire metastatic sites of tumors. Results: Our method achieves a recall of 0.697, a precision of 0.825 and an F1 score of 0.755 using an overall weighted metric. For each of the extraction tasks, the F1 scores are 0.784, 0.822 and 0.740. Conclusions: The method proves to be stable and robust with different amounts of testing data. It achieves a comparatively high performance in the CHIP 2018 open challenge, demonstrating its effectiveness in extracting tumor-related entities from Chinese diagnostic imaging text.

Knowledge Graphs (KGs) contain rich semantic information and are of importance to many downstream tasks. In order to enhance practical utilization of KGs, KG completion task, which is also called link prediction, is a newly emerging hot research topic. During KG embedding model training, negative sampling is a fundamental method for obtaining negative samples. Inspired by an adversarial learning framework KBGAN, this paper proposes a new knowledge selective adversarial network, named as KSGAN, using a knowledge selector for high-quality negative sampling to benefit link prediction. The performances of our model KSGAN are evaluated on three standard knowledge completion datasets: FB15k-237, WN18 and WN18RR. The results show that KSGAN outperforms a list of baseline models on all the datasets, demonstrating the effectiveness of the proposed model.

This chapter presents a quantitative and visual analysis of social media and health research publications from Web of Science database during the year 2007–2017. The analysis is conducted using a bibliometric method, a social network analysis method, and a latent dirichlet allocation method to acquire the predominant subjects, journals, and countries, the collaboration relationship, and the major topics. Some interesting results are presented. For example, Journal of Medical Internet Research is the most influential journal. Public, Environmental & Occupational Health and Health Care Sciences & Services are the subjects with the most publications and citations, respectively. The USA is the most influential country with 1317 publications and an H-index of 53. Twenty topics are identified with potential themes as: Sex-related event, Analysis on medical-related content, Vaccine, Adverse drug reactions, Diet and weight control, Smoking cessation, Nursing, etc., which have received much more attention in scientific community during 2012–2017 compared with the period 2007–2011.

Table S1. Search strategy and keywords used for Web of Science. Table S2. Search strategy and keywords used for PubMed. (DOCX 16 kb)

Attention mechanism has been proved to be able to improve the quality of neural machine translation by selectively focusing on partial words of a source sentence during translation process. Attention mechanism usually focuses on local attention by using solely the linear index distance of words while ignores syntax structures of sentences. In this paper, we extend local attention through syntax distance constraint, and propose an attention mechanism based on a new syntactic branch distance, which simultaneously pays attention to words with similar linear index distances and syntax-related words. Based on the English-to-German translation task, experiment results showed that our model outperforms a recent baseline method with an improvement of 1.61 BLEU points, demonstrating the effectiveness of the proposed model.