Search

Question answering (QA) based on machine reading comprehension has been a recent surge in popularity, yet most work has focused on extractive methods. We instead address a more challenging QA problem of generating a well-formed answer by reading and summarizing the paragraph for a given question.For the generative QA task, we introduce a new neural architecture, LatentQA, in which a novel stochastic selector network composes a well-formed answer with words selected from the question, the paragraph and the global vocabulary, based on a sequence of discrete latent variables. Bayesian inference for the latent variables is performed to train the LatentQA model. The experiments on public datasets of natural answer generation confirm the effectiveness of LatentQA in generating high-quality well-formed answers.

OBJECTIVE To determine the expression levels and clinical significance of serum N-terminal pro-brain natriuretic peptide (NT-proBNP), hydrogen sulfide (H₂S) and interleukin-6 (IL-6) in children with Kawasaki disease (KD). STUDY DESIGN Descriptive comparative study. PLACE AND DURATION OF STUDY Department of Pediatric Medicine, Baoding Children's Hospital, from July 2017 to July 2018. METHODOLOGY Ninety-five KD children were chosen as the case group, and were classified into CAL group (23 patients) and NCAL group (72 patients, according to the presence of a coronary artery lesion (CAL). Forty-six non-KD children with an upper respiratory infection in the same time period were chosen as the control group. Electrochemiluminescence method was used to detect serum NT-proBNP levels. The spectrophotometer method was used to test H2S levels, and an enzyme-linked immunosorbent assay was used to test serum IL-6 levels and to analyse the correlation. RESULTS In the acute phase and recovery phase, serum NT-proBNP and IL-6 levels were higher in the case group than the control group, while H2S levels were lower than those in the control group (p

BACKGROUND Ewing’s sarcoma (ES) is regarded as a skeletal tumor, with few instances of extra-skeletal ES. A primary ES in the ureter is extremely rare. CASE SUMMARY We report the case of a 69-year-old woman who presented with intermittent flank pain and hematuria and was found to have a mass in the left ureter. Pathology of the excised mass indicated ES. The clinical treatment and pathologic characteristics in this case, and a review of the literature describing ES in the urinary system, are presented. CONCLUSION Due to the rarity and malignancy of ES in ureter, early diagnosis and prompt surgical treatment are critical.

Cold-start is the most difficult and time-consuming phase when building a question answering based chatbot for a new business scenario because of the collection of sufficient training data. In this paper, we propose AliMe DA, a practical data augmentation (DA) framework that consists of data production, denoising and consumption, to alleviate this problem. We show how our DA approach can be used to substantially enhance annotation productivity and also improve downstream model performance. More importantly, we provide best practices for data augmentation, including how to choose and employ appropriate methods at each stage of our framework, and share our observation on the applicable scene of data augmentation in the era of pre-trained language models.

Large pre-trained language models achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, they almost exclusively focus on text-only representation, while neglecting cell-level layout information that is important for form image understanding. In this paper, we propose a new pre-training approach, StructuralLM, to jointly leverage cell and layout information from scanned documents. Specifically, we pre-train StructuralLM with two new designs to make the most of the interactions of cell and layout information: 1) each cell as a semantic unit; 2) classification of cell positions. The pre-trained StructuralLM achieves new state-of-the-art results in different types of downstream tasks, including form understanding (from 78.95 to 85.14), document visual question answering (from 72.59 to 83.94) and document image classification (from 94.43 to 96.08)., Accepted by ACL2021 main conference

Recently, question answering (QA) based on machine reading comprehension has become popular. This work focuses on generative QA which aims to generate an abstractive answer to a given question instead of extracting an answer span from a provided passage. Generative QA often suffers from two critical problems: (1) summarizing content irrelevant to a given question, (2) drifting away from a correct answer during generation. In this paper, we address these problems by a novel Rationale-Enriched Answer Generator (REAG), which incorporates an extractive mechanism into a generative model. Specifically, we add an extraction task on the encoder to obtain the rationale for an answer, which is the most relevant piece of text in an input document to a given question. Based on the extracted rationale and original input, the decoder is expected to generate an answer with high confidence. We jointly train REAG on the MS MARCO QA+NLG task and the experimental results show that REAG improves the quality and semantic accuracy of answers over baseline models.

Vision-language pre-training (VLP) on large-scale image-text pairs has achieved huge success for the cross-modal downstream tasks. The most existing pre-training methods mainly adopt a two-step training procedure, which firstly employs a pre-trained object detector to extract region-based visual features, then concatenates the image representation and text embedding as the input of Transformer to train. However, these methods face problems of using task-specific visual representation of the specific object detector for generic cross-modal understanding, and the computation inefficiency of two-stage pipeline. In this paper, we propose the first end-to-end vision-language pre-trained model for both V+L understanding and generation, namely E2E-VLP, where we build a unified Transformer framework to jointly learn visual representation, and semantic alignments between image and text. We incorporate the tasks of object detection and image captioning into pre-training with a unified Transformer encoder-decoder architecture for enhancing visual learning. An extensive set of experiments have been conducted on well-established vision-language downstream tasks to demonstrate the effectiveness of this novel VLP paradigm., ACL2021 main conference

Existing work in multilingual pretraining has demonstrated the potential of cross-lingual transferability by training a unified Transformer encoder for multiple languages. However, much of this work only relies on the shared vocabulary and bilingual contexts to encourage the correlation across languages, which is loose and implicit for aligning the contextual representations between languages. In this paper, we plug a cross-attention module into the Transformer encoder to explicitly build the interdependence between languages. It can effectively avoid the degeneration of predicting masked words only conditioned on the context in its own language. More importantly, when fine-tuning on downstream tasks, the cross-attention module can be plugged in or out on-demand, thus naturally benefiting a wider range of cross-lingual tasks, from language understanding to generation. As a result, the proposed cross-lingual model delivers new state-of-the-art results on various cross-lingual understanding tasks of the XTREME benchmark, covering text classification, sequence labeling, question answering, and sentence retrieval. For cross-lingual generation tasks, it also outperforms all existing cross-lingual models and state-of-the-art Transformer variants on WMT14 English-to-German and English-to-French translation datasets, with gains of up to 1~2 BLEU., Accepted by ACL 2021 (long paper)

The antibiotics residues in freshwater lakes are being highlighted around the world because of high potential threat to environment and human health. Understanding the current state of antibiotics and potential sources in lakes are important. The potential sources of antibiotics (Sewage treatment plants (STPs)), livestock and poultry farms and fishponds in the East Dongting was studied. Compared with other surface water bodies, the concentration of antibiotic in the East Dongting Lake was at a moderate or low level. Ten of 12 antibiotics likely originated from veterinary applications in livestock and poultry farms, especially in swine farms, and concentrations at these sources (ND-1240.41 ng L−1) were 1–3 orders of magnitude higher than in the effluent of local sewage treatment plants and fishponds. Based on a principal component analysis with multiple linear regression (PCA-MLR) model, we estimated source contributions of 79.95% for livestock and poultry farms, 0.27% for STPs, and 19.79% for aquaculture source and livestock and poultry farms. Overall, the predominance of sulfonamides and livestock and poultry farms in the East Dongting Lake has been identified, which can provide important information for regulating their veterinary use and environmental management.

This single-center, randomized, single-blind, parallel-controlled study aimed to analyze the changes in resting-state functional connectivity (RSFC) in young patients with a suicide attempt caused by depression before and after cognitive-behavioral therapy (CBT) combined with fluoxetine or fluoxetine alone by functional magnetic resonance imaging (fMRI). Before treatment, functional connectivity of the right subgenual anterior cingulate cortex (R-sgACC), left subgenual anterior cingulate cortex (L-sgACC) and right precuneus (R-PCu) was lower in depressed patients with a suicide attempt than that of healthy controls. After treatment, compared with the fluoxetine group, functional connectivity between the R-sgACC and left posterior cerebellar lobe in the CBT group was increased, while this group also showed increased RSFC between the L-sgACC and right anterior cingulate cortex/ medial prefrontal cortex. On the contrary, the functional connectivity between the R-PCu and right parietal lobe was reduced (P 0.001). It was also found there were some changes in different brain regions in pre- and post-treatment within both the CBT and MG group. The functional connectivity of the R-sgACC and the left posterior cerebellum lobe was negatively correlated with the SSI score. The functional connectivity of the R-PCu and right middle frontal cortex was negatively correlated with the HAMD score before treatment. After treatment, functional connectivity between the R-PCu and right superior frontal gyrus was positively correlated with the SSI scores in the CBT group. After 8 weeks of combined CBT, the strength of the functional connectivity in the bilateral sgACC and bilateral PCu was significantly changed.

Taking the cycloid pin gear reducer as the research object, the bearing force of the cycloid pin gear reducer shaft is calculated under different modification amount. Based on the Hertz theory, the contact stress and deformation of the bearing under different modification amounts are calculated. Based on the transmission principle of cycloid pin wheel reducer, the model of cycloid wheel and bearing is established and the finite element analysis is carried out. The analysis showed that the Hertz theoretical solution is consistent with the finite element analysis solution, which showed the correctness of the finite element results. The stress and deformation of the bearing obtained under different modification amounts of the cycloid wheel are not changed much. It can indicate that the cycloid wheel modification has no significant influence on the bearing force. Therefore, the research content of this paper has certain reference value for the design and optimization of cycloid pinwheel reducer.

The cycloidal wheel is a key component of the RV reducer, and its reliability directly affects the overall performance of the RV reducer. Based on the stress-intensity distribution interference theory, this paper establishes a mathematical calculation model for the reliability of the cycloidal wheel. Using software to write the program, analyze the distribution type of contact force and contact fatigue strength of cycloidal wheel, and then use software based on theoretical calculation, and calculate the reliability based on Monte Carlo method. Then through the parameter sensitivity analysis, it is concluded that the degree of influence of the parameters that change with the actual operating factors on the reliability during the actual operation provides a basis for the key calculation of the reliability analysis of the tooth surface contact fatigue of the cycloidal gear.

Osteosarcoma is the most frequent primary malignant bone tumor that occurs in children and adolescents. The present study aimed to identify novel therapeutic strategies for osteosarcoma, by assessing the antitumor activity of the cannabinoid WIN-55,212-2 and its combined effect with adriamycin (ADM) against the MG-63 human osteosarcoma cell line. To evaluate the antiproliferative action of these molecules, a Cell Counting kit-8 (CCK-8) assay was used. The ability of cannabinoid to inhibit the migration, invasion and angiogenic activity of MG-63 cells were assessed by scratch, Transwell® chamber and angiogenesis assays, respectively, in vitro. To examine the alterations in expression of targeted genes, quantitative polymerase chain reaction and western blot analysis were used. The administration of cannabinoid combined with ADM was demonstrated to inhibit the growth of MG-63 cells, resulting in a cell viability of 32.12±3.13%, which was significantly lower (P

To observe the preventive and therapeutic effects of diosgenin on retinoic acid-induced osteoporosis in rats.A total 50 Sprague-Dawley rats were randomly divided into 5 groups: control group, model group (osteoporosis rats), low (10 mg kg(-1)), middle (30 mg kg(-1)), and high-dose diosgenin (90 mg kg(-1))-treated groups. The osteoporosis rats model was induced by retinoic acid. The BMD and physical parameters of femoral including length, wet weight, and dry weight in each group were measured. The hematoxylin-eosin staining was used for bone histomorphology analysis. Besides, the bone calcium (Ca) and phosphorus (P) contents were measured. In order to detect the biochemical index in different treatment groups, the serum tartrate-resistant acid phosphatase (TRAP), alkaline phosphatase (ALP), estradiol, and osteocalcin were compared among different groups.The osteoporosis rat model was successfully induced by retinoic acid. Compared with the model group, the lessening of femoral length and weight and the loss of BMD were significantly improved in diosgenin groups. Both contents of Ca and P were much more increased when induced by retinoic acid (p 0.05). The estradiol and osteocalcin levels in the middle and high-dose treatment groups were significantly higher than that of the model group, while the ALP and TRAP levels were much lower than the model group (p 0.05).Diosgenin can prevent the loss of bone in experimental rats. The mechanism may be that it improves the level of estrogenic hormone of estradiol and inhibits the high bone turnover.

Pollution caused by antibiotics has been highlighted in recent decades as a worldwide environmental and health concern. Compared to traditional physical, chemical and biological treatments, constructed wetlands (CWs) have been suggested to be a cost-efficient and ecological technology for the remediation of various kinds of contaminated waters. In this review, 39 antibiotics removal-related studies conducted on 106 treatment systems from China, Spain, Canada, Portugal, etc. were summarized. Overall, the removal efficiency of CWs for antibiotics showed good performance (average value = over 50%), especially vertical flow constructed wetlands (VFCWs) (average value = 80.44%). The removal efficiencies of sulfonamide and macrolide antibiotics were lower than those of tetracycline and quinolone antibiotics. In addition, the relationship between the removal efficiency of antibiotics and chemical oxygen demand (COD), total suspended solids (TSS), total nitrogen (TN), total phosphorus (TP) and ammonia nitrogen (NH3-N) concentrations showed an inverted U-shaped curve with turning points of 300 mg L−1, 57.4 mg L−1, 40 mg L−1, 3.2 mg L−1 and 48 mg L−1, respectively. The coexistence of antibiotics with nitrogen and phosphorus slightly reduced the removal efficiency of nitrogen and phosphorus in CWs. The removal effect of horizontal subsurface flow constructed wetlands for antibiotic resistance genes (ARGs) had better performance (over 50%) than that of vertical wetlands, especially for sulfonamide resistance genes. Microorganisms are highly sensitive to antibiotics. In fact, microorganisms are one of the main responsible for antibiotic removal. Moreover, due to the selective pressure induced by antibiotics and drug-resistant gene transfer from resistant bacteria to other sensitive strains through their own genetic transfer elements, decreased microbial diversity and increased resistance in sewage have been consistently reported. This review promotes further research on the removal mechanism of antibiotics and ARGs in CWs.

On the basis of human factors analysis of driver error accidents, this paper establishes the accident-causing index system due to driver errors and puts forward the use of trapezoidal fuzzy hierarchy - the grey correlation analysis method to analyze the accident-causing index model due to driver errors. Then this paper uses related data and proposed analysis method to obtain the weights order of accident-causing factors. The results reveal the main cause of accidents due to driver errors, and provide the reference for further research to prevent the traffic accidents.

Twitter (and similar microblogging services) has become a central nexus for discussion of the topics of the day. Twitter data contains rich content and structured information on users' topics of interest and behavior patterns. Correctly analyzing and modeling Twitter data enables the prediction of the user behavior and preference in a variety of practical applications, such as tweet recommendation and followee recommendation. Although a number of models have been developed on Twitter data in prior work, most of these only model the tweets from users, while neglecting their valuable retweet information in the data. Models would enhance their predictive power by incorporating users' retweet content as well as their retweet behavior. In this paper, we propose two novel Bayesian nonparametric models, URM and UCM, on retweet data. Both of them are able to integrate the analysis of tweet text and users' retweet behavior in the same probabilistic framework. Moreover, they both jointly model users' interest in tweet and retweet. As nonparametric models, URM and UCM can automatically determine the parameters of the models based on input data, avoiding arbitrary parameter settings. Extensive experiments on real-world Twitter data show that both URM and UCM are superior to all the baselines, while UCM further outperforms URM, confirming the appropriateness of our models in retweet modeling.

In this paper, a 4WS control method based on BP neural network was introduced. It used the BP neural network to simulate the map of vehicle and the nonlinear dynamic characteristics of the tire to avoid large errors that relying on mathematical simulation model of the problem. The 4WS measured data of Tokyo institute of Technology institute of Japan was used and used BP neural network method to identify the nonlinear characteristics of vehicle and tires. System controller’s design is not based on any theoretical method, but on the BP neural network’s self-learning ability. Experimental results show that this method has good controlling characteristics, and it can improve the vehicle’s active safety and manipulating stability effectively.

This article presents our solution to PAKDD’07 Data Mining Competition, whose task is to build a classifier to score the propensity of a credit card customer to take up a home loan with a finance company. After analyzing the task, we first describe the data preparation steps in detail. Then, a mixed resampling method is put forward to deal with the problem that model samples are redundant and class imbalance. Following that, a hybrid classifier that integrates Logistic Regression, Adaboost with Decision Stump and Voting Feature Intervals, is built. It is evaluated via cross-identification. Finally, some useful business insights gained from our solution are interpreted.

Stem cells hold great promise for brain and spinal cord injuries (SCI), but cell survival following transplantation to adult central nervous system has been poor. Salvianolic acid B (Sal B) has been shown to improve functional recovery in brain-injured rats. The present study was designed to determine whether Sal B could improve transplanted mesenchymal stem cell (MSC) survival in SCI rats.SCI rats were treated with Sal B. The Basso-Beatie-Bresnahan (BBB) scale was used to test the functional recovery. Sal B was used to protect MSC from being damaged by TNF-alpha in vitro. Bromodeoxyuridine-labeled MSC were transplanted into SCI rats with Sal B intraperitoneal injection, simultaneously. MSC were examined, and the functional recovery of the SCI rats was tested.Sal B treatment significantly reduced the lesion area from 0.26+/-0.05 mm2 to 0.15+/-0.03 mm2 (P0.01) and remarkably raised the BBB scores on d 28, post-injury, from 7.3+/-0.9 to 10.5+/-1.3 (P0.05), compared with the phosphate-buffered saline (PBS) control group. MSC were protected from the damage of TNF-alpha by Sal B. The number of surviving MSC in the MSC plus Sal B groups were 1143.3+/-195.6 and 764.0+/-81.3 on d 7 and 28, post-transplantation, more than those in the MSC group, which was 569.3+/-72.3 and 237.0+/-61.3, respectively (P0.05). Rats with MSC transplanted and Sal B injected obtained higher BBB scores than those with MSC transplanted alone (P0.05) and PBS (P0.01).Sal B provides neuroprotection to SCI and promotes the survival of MSC in vitro and after cell transplantation to the injured spinal cord in vivo.

Over the past few years, major web search engines have introduced knowledge bases to offer popular facts about people, places, and things on the entity pane next to regular search results. In addition to information about the entity searched by the user, the entity pane often provides a ranked list of related entities. To keep users engaged, it is important to develop a recommendation model that tailors the related entities to individual user interests. We propose a probabilistic Three-way Entity Model (TEM) that provides personalized recommendation of related entities using three data sources: knowledge base, search click log, and entity pane log. Specifically, TEM is capable of extracting hidden structures and capturing underlying correlations among users, main entities, and related entities. Moreover, the TEM model can also exploit the click signals derived from the entity pane log. We further provide an inference technique to learn the parameters in TEM, and propose a principled preference learning method specifically designed for ranking related entities. Extensive experiments with two real-world datasets show that TEM with our probabilistic framework significantly outperforms a state of the art baseline, confirming the effectiveness of TEM and our probabilistic framework in related entity recommendation.

Social influence analysis on microblog networks, such as Twitter, has been playing a crucial role in online advertising and brand management. While most previous influence analysis schemes rely only on the links between users to find key influencers, they omit the important text content created by the users. As a result, there is no way to differentiate the social influence in different aspects of life (topics). Although a few prior works do support topic-specific influence analysis, they either separate the analysis of content from the analysis of network structure, or assume that content is the only cause of links, which is clearly an inappropriate assumption for microblog networks. To address the limitations of the previous approaches, we propose a novel Followship-LDA (FLDA) model, which integrates both content topic discovery and social influence analysis in the same generative process. This model properly captures the content-related and content-independent reasons why a user follows another in a microblog network. We demonstrate that FLDA produces results with significantly better precision than existing approaches. Furthermore, we propose a distributed Gibbs sampling algorithm for FLDA, and demonstrate that it provides excellent scalability on large clusters. Finally, we incorporate the FLDA model in a general search framework for topic-specific influencers. A user freely expresses his/her interest by typing a few keywords, the search framework will return a ranked list of key influencers that satisfy the user's interest.

Topic models are used to group words in a text dataset into a set of relevant topics. Unfortunately, when a few words frequently appear in a dataset, the topic groups identified by topic models become noisy because these frequent words repeatedly appear in "irrelevant" topic groups. This noise has not been a serious problem in a text dataset because the frequent words (e.g., the and is) do not have much meaning and have been simply removed before a topic model analysis. However, in a social network dataset we are interested in, they correspond to popular persons (e.g., Barack Obama and Justin Bieber) and cannot be simply removed because most people are interested in them. To solve this "popularity problem", we explicitly model the popularity of nodes (words) in topic models. For this purpose, we first introduce a notion of a "popularity component" and propose topic model extensions that effectively accommodate the popularity component. We evaluate the effectiveness of our models with a real-world Twitter dataset. Our proposed models achieve significantly lower perplexity (i.e., better prediction power) compared to the state-of-the-art baselines. In addition to the popularity problem caused by the nodes with high incoming edge degree, we also investigate the effect of the outgoing edge degree with another topic model extensions. We show that considering outgoing edge degree does not help much in achieving lower perplexity.

Knowing users' views and demographic traits offers a great potential for personalizing web search results or related services such as query suggestion and query completion. Such signals however are often only available for a small fraction of search users, namely those who log in with their social network account and allow its use for personalization of search results. In this paper, we offer a solution to this problem by showing how user demographic traits such as age and gender, and even political and religious views can be efficiently and accurately inferred based on their search query histories. This is accomplished in two steps; we first train predictive models based on the publically available myPersonality dataset containing users' Facebook Likes and their demographic information. We then match Facebook Likes with search queries using Open Directory Project categories. Finally, we apply the model trained on Facebook Likes to large-scale query logs of a commercial search engine while explicitly taking into account the difference between the traits distribution in both datasets. We find that the accuracy of classifying age and gender, expressed by the area under the ROC curve (AUC), are 77% and 84% respectively for predictions based on Facebook Likes, and only degrade to 74% and 80% when based on search queries. On a US state-by-state basis we find a Pearson correlation of 0.72 for political views between the predicted scores and Gallup data, and 0.54 for affiliation with Judaism between predicted scores and data from the US Religious Landscape Survey. We conclude that it is indeed feasible to infer important demographic data of users from their query history based on labelled Likes data and believe that this approach could provide valuable information for personalization and monetization even in the absence of demographic data.

This work focuses on applying data mining techniques to a practical example of cross-selling problem raised by PAKDD'07 Data Mining Competition. Firstly, we build decision tree from high dimensional data to identify the most important features according to information gain. Some comprehensible business insights are also gained from it. Secondly, a novel re-sampling technique is proposed to resolve the imbalanced dataset problem. Thirdly, we apply MLP + n-tuple classifier to build prediction model. Finally, experiments and results are presented to evaluate the performance of our approach.