Search

Today, with the dividend of mobile Internet traffic disappearing, mobile advertising is focusing on powerful algorithms of personalized recommendation. Sina Weibo has abundant advertising spaces and the accuracy of its recommendation directly affects user experience. UGC (user-generated content) is the selected or even original edited content for interaction through users’ subjective interests, which reflects users’ inner interests within a certain period of time. Adaptive DBSCAN clustering is performed on Weibo UGC, and the clustering center represents the user’s interest to a certain extent. By refining the text theme of advertisements to be recommended, the similarity between clustering center and the recommended theme is obtained, and the advertising category with high similarity is recommended to user. The experiment proves that the method is effective.

Epilepsy is characterized by oxidative stress in the brain. As the crucial reductive biothiol, cysteine (Cys) directly regulates the occurrence of oxidative stress in the brain. Observations suggest that the decreased cysteine in plasma could potentially serve as a redox biomarker in temporal lobe epilepsy. However, due to the complexity of the brain and lack of appropriate in situ detecting tools, the concentration change and regulation of Cys in epileptic brains remain unclear. Here, we report a near-infrared imaging probe (named Mito-CP) for tracking endogenous Cys in brains of pentylenetetrazole (PTZ)-induced epileptic seizures with high sensitivity and selectivity. Using this probe, we achieved an in situ visualization of the increased Cys in PC12 cells under dithiothreitol stimulation. In addition, Mito-CP was able to real-time monitor Cys fluctuation in lipopolysaccharide-mediated oxidative stress in zebrafish. Ultimately, we directly visualized the precipitous reduction of Cys in epileptic brains for the first time. Mito-CP also revealed the fluctuation of Cys in epileptic brains during the treatment by an antiepileptic drug, curcumin. This work provides an effective tool for Cys imaging in brains and will help to expand the understanding of the pathogenesis of epilepsy.

With the rapid spread of COVID-19, how to deal with the public opinion caused by COVID-19 epidemic situation and correctly guide public emotion has become an urgent problem to be solved. This paper collects more than 90 000 Weibo and more than 600000 Weibo comments from January 1 to February 29, 2020. Through web crawler technology, social network analysis, SnowNLP emotion analysis and text clustering, this paper analyzes the public opinion related topics of COVID-19, shows the evolution of public opinion in time, and finds that the change of Weibo emotion in this period is roughly divided into four stages. That is, the initial panic and anxiety about the unknown virus, the temporary relaxation after official clarification, the emotional ups and downs during the worst period of the epidemic and finally the confident and stable period of fighting the epidemic, at the same time, the emotional changes of key users will affect the overall emotional changes in the same direction during this period.

By analyzing the big data of Chinese-language films, we can predict high-quality films that are both “good and popular” before they are released. Firstly, collect data of 1876 Chinese-language films released in 2009-2017, and then select indicators that affect box office and score from four aspects: film people, plot, release time and synthesize, propose a prerelease prediction model. Finally, the model is trained by using 6 methods such as Random Forest to predict the “high quality or not” of the films. The results show that the best prediction performance comes from Random Forest, with an accuracy rate of 78.29% and an AUC value of 0.831. The model can provide decision-making reference for investors to invest in high quality films and bring more high quality films to the market.

The accurate analysis of tourist volume plays an important role in the scientific management of tourism resources and the formulation of policies by tourism decision makers. The data generated by tourists' online search behavior provide a new perspective for tourism prediction. Based on the six elements of tourism, combined with text analysis, the conceptual framework of Baidu index keywords related to tourist volume is established. The statistical test and correlation analysis of Baidu index keywords were carried out, and the keyword sequences with predictive ability were selected. Then several prediction models are established to predict the number of visitors and optimize the model. Finally, the prediction accuracy of different models is analyzed by using goodness of fit (R2) and mean absolute percentage error (MAPE). It is found that the variable weight combination prediction model based on GBDT has the best effect, $\mathrm{R}^{2}$ and MAPE are 0.9943 and 1.78%, respectively.

The SARS-CoV-2 variant VUI/202012/01 has been reported to spread unexpectedly fast in the United Kingdom. It is estimated that its transmissibility may increase by 70%. In this study, the top five variants circulating in the UK including D614G+L18F+A222V, D614G+A222V, D614G+S477N, VUI/202012/01 and D614G+69-70del+439K were analyzed for their infective and neutralizing characteristics. The pseudotyped viruses were constructed for the five variants and 12 single mutants composed those variants. We found that the VUI/202012/01 variant does enhance its infectivity due to the cumulative effect of multiple mutations including 69-70del and 144/145del mutations in NTD, A570D in RBD, and S982A in S2. Meanwhile, mutations N501Y, N439K and S477N in RBD can cause a significant decrease in the neutralization activity for some mAbs. Although VUI/202012/01 did not affect the neutralization effect of convalescent sera, it affected the neutralization activity of animal immunized sera by RBD protein or recombinant spike DNA to some extent.

Due to the improvement of Language Modeling, the emerging NLP assistant tools aiming for text generation greatly reduce the human workload on writing documents. However, the generation of legal text faces greater challenges than ordinary texts because of its high requirement for keeping logic reasonable, which can not be guaranteed by Language Modeling right now. To generate reasonable legal documents, we propose a novel method CoLMQA, which (1) combines Language Modeling and Question Answering, (2) generates text with slots by Language Modeling, and (3) fills the slots by our proposed Question Answering method named Transformer-based Key-Value Memory Networks. In CoLMQA, the slots represent the text part that needs to be highly constrained by logic, such as the name of the law and the number of the law article. And the Question Answering fills the slots in context with the help of Legal Knowledge Base to keep logic reasonable. The experiment verifies the quality of legal documents generated by CoLMQA, surpassing the documents generated by pure Language Modeling.

A clear elucidation of a disease-related viscosity change in vivo is significant yet highly challenging as well. Fluorescence imaging in the second near-infrared region (NIR-II, 1000-1700 nm) has gained increasing attention for observation in living organisms, but a viscosity-activatable fluorescent probe emitting at this region remains a vacancy. Herein, we report the first panel of a viscosity-activated NIR-II emissive fluorescent probe WD-X. By embedding different substituents into the WD-X platform and screening, we obtained an ideal probe, WD-NO2, which displayed the best combination of properties, including a 31-fold fluorescence enhancement in response to viscosity, insensitivity to environments (pH, polarity), and relatively high quantum yield (1.6% in glycerol). WD-NO2 was successfully applied to track the variation of viscosity in diabetes-induced liver injury in vivo.

Small organic phototherapeutic molecules of the second near-infrared (NIR-II) window (1000-1700 nm) serve as promising candidates for theranostics. However, developing such versatile agents for fluorescence-guided photodynamic/photothermal therapy remains a demanding task stirred by competitive energy dissipation pathways, including radiative decay, internal conversion, and intersystem crossing. To the best of current knowledge, the current paradigm for addressing the issue has deliberately approached the optimum balance among three deactivation processes through offsetting from each other, possibly leading to a comprehensively compromised theranostic efficacy. Few reports aim to modulate the three deactivation pathways excluding sacrificing any one of them. Herein, a molecular design strategy to construct a phototherapeutic organic fluorophore CCNU-1060, armed with NIR-II luorescence-guided phototherapeutic properties, is rationally developed. With a flexible motor, tetraphenylethene, bridged to the rigidified coplanar core boron-azadipyrromethene, the desired CCNU-1060 is subsequently encapsulated into an amphiphilic matrix to form CCNU-1060 nanoparticles (NPs), which match or transcend its precursor NJ-1060 NPs in the three energy dissipation processes. CCNU-1060 NPs are utilized to realize high-spatial vessel imaging and effective NIR-II fluorescence-guided phototherapeutic tumor ablation. This study unlocks a viewpoint of molecular engineering that simultaneously regulates multiple energy dissipation pathways for the construction of versatile phototherapy agents.

Nucleotide-based methods are conventionally used to classify the hepatitis E virus (HEV) genotypes. A serological enzyme immunoassay (EIA) using open reading frame 3 (ORF3) C-terminal peptides was developed to conveniently and accurately classify and evaluate the genotypes of HEV. The sera of mice immunized with HEV genotype 1, 3, and 4 reacted highly specifically to the peptides of the corresponding genotypes. Most (84.2%) clinical sera infected with HEV genotype 4 were positive for anti-HEV antibodies when tested with the ORF3 peptides of genotype 4, but were negative for genotypes 1 and 3. Monkey and clinical serial sera infected with HEV reacted strongly to the homologous genotype ORF3 peptides. The indirect EIAs were more sensitive, with stronger reactivity, than commercial anti-HEV immunoglobulin G assays when serial sera from monkeys infected with HEV genotype 1 or 4 were tested. All our results indicate that the serological typing EIA assays described in this study are more effective and convenient for the classification of HEV genotypes than molecular approaches, and can be used to screen large numbers of serum samples and differentiate genotypes for the diagnosis of HEV infections.

Pre-existing neutralizing antibodies (NAbs) directed against vaccine vectors have attracted considerable research attention. Therefore, our aim was to establish a high-throughput economical neutralization assay to investigate the epidemiology of adenovirus type 2 (Ad2)-specific immunity in China and developed countries, including in a Chinese Human immunodeficiency virus (HIV)-1-infected population, and to guide the application of Ad2-vectored vaccines. We established a FluoroSpot-based anti-Ad2-virus neutralization assay using a recombinant replication-deficient Ad2 that expresses enhanced green fluorescent protein and standardized the critical parameters, including the choice of cell line, cell concentration, viral infective dose, and incubation time. The sera of 561 healthy individuals from China and developed countries and from 230 HIV-1-infected Chinese individuals were screened with this assay for Nabs against Ad2. The prevalence of anti-Ad2 NAbs was high in both China (92.2%) and developed countries (86.9%). Of the Ad2-seropositive individuals, 64.6% in China and 77.4% in developed countries had high NAb titers (> 810). The frequency of anti-Ad2 NAbs was higher in Anhui (97.5%) than in Beijing (88.7%). Their prevalence differed significantly according to age in Beijing, but not in Anhui Province, but by sex in neither province. Ad2 seroprevalence was as high among HIV-1-infected individuals (88.7%) as among healthy individuals (92.2%) in China. In conclusion, a simple, intuitive, high-throughput, economical fluorescence-based neutralization assay was developed to determine anti-Ad2 NAbs titers. Ad2 exposure was high in both healthy and HIV-1-infected populations in China, so vectors based on Ad2 may be inappropriate for human vaccines.

Treatment with chimeric antigen receptor (CAR)-modified T cells targeting CD19 has proved successful in patients with relapsed/refractory B cell malignancies. However, long-term follow-up indicates that remission in a substantial proportion of patients is not sustainable. Most patients that experience recurrence have tumors and lost the CAR-T cells. To maintain the activity of CAR-T cells, Raji-B-NDG mice were treated sequentially with CAR-T-19 cells and homologous cells expressing human CD19 to promote expansion of CAR-T cells. Sequential treatment of mice with CAR-T-19 cells followed by Raji tumor cells led to marked prolongation of survival. The best case scenario after sequential treatment was a survival time of more than 200 days; the average survival time of mice in the non-sequential treatment group was 80 days. We treated mice with autologous CD19-modified T cells after initial treatment with CAR-T-19 cells. The overall survival and recurrence-free survival times of mice receiving sequential treatment were significantly longer. The percentages of CAR+ T cells in peripheral blood increased. Sequential therapy with autologous CAR-T-19 and aT19 cells provides a new strategy for generating memory CAR-T cells, which may lead ultimately to increased clinical efficacy.

H7N9 viral infections pose a great threat to both animal and human health. This avian virus cannot be handled in level 2 biocontainment laboratories, substantially hindering evaluation of prophylactic vaccines and therapeutic agents. Here, we report a high-titer pseudoviral system with a bioluminescent reporter gene, enabling us to visually and quantitatively conduct analyses of virus replications in both tissue cultures and animals. For evaluation of immunogenicity of H7N9 vaccines, we developed an in vitro assay for neutralizing antibody measurement based on the pseudoviral system; results generated by the in vitro assay were found to be strongly correlated with those by either hemagglutination inhibition (HI) or micro-neutralization (MN) assay. Furthermore, we injected the viruses into Balb/c mice and observed dynamic distributions of the viruses in the animals, which provides an ideal imaging model for quantitative analyses of prophylactic and therapeutic monoclonal antibodies. Taken together, the pseudoviral systems reported here could be of great value for both in vitro and in vivo evaluations of vaccines and antiviral agents without the need of wild type H7N9 virus.

BACKGROUND HIV-1 gp120/gp41 is heavily modified by n-linked carbohydrates that play important roles either in correct folding or in shielding vulnerable viral protein surfaces from antibody recognition. METHODS In our previous work, 25 potential N-linked glycosylation sites (PNGS) of a CRF07_BC isolate of HIV-1 were individually mutated, and the resulting effects on infectivity and antibody-mediated neutralization were evaluated. To further understand the functional role of these PNGS, we generated double and multiple mutants from selected individual PNGS mutants. The effects were then evaluated by examining infectivity and sensitivity to antibody-mediated neutralization by neutralizing monoclonal antibodies (nMAbs) and serum antibodies from HIV-1 positive donors. RESULTS Infectivity results showed that, among the 12 combined PNGS mutants, only 197M.1 (N197D/N301Q) lost infectivity completely, whereas all others (except for 197M.6) showed reduced viral infectivity. In terms of neutralization sensitivity to known nMAbs, we found that adding N463Q mutation to all the gp120 mutants containing N197D significantly increased neutralization sensitivity to VRC01 and VRC03, suggesting N197 and N463 have a strong synergistic effect in regulating the neutralizing sensitivity of HIV-1 to the anti-CD4bs nMAbs VRC01/VRC03. Structural analysis based on the available structures of gp120 alone and in complex with CD4 and various nMAbs elucidates a molecular rationale for this experimental observation. CONCLUSIONS The data indicate that N463 plays an important role in regulating the CD4bs MAbs VRC01/VRC03 sensitivity in the genetic background of N197D mutation of gp120, which should provide valuable information for a better understanding of the interplay between HIV-1 and VRC01/03.

Topic modeling on phrases is important in understanding documents by providing interpretable topics. But existing methods are not as efficient as the topic modeling methods on words, which may limit their potential application.Towards providing a more efficient method, we propose a novel topic model SparseTP, which (1) models the words and phrases by linking them in Markov Random Field when necessary; (2) provides a well-formed lower bound of the model for Gibbs sampling; (3) utilizes the sparse distribution of words and phrases on topics to speed up the inference. The experiments demonstrate that it can achieve the high efficiency without sacrificing the effectiveness.

Potential-induced degradation (PID) causes a sharp decrease in the efficiency and shunt resistance of the solar module. In this paper, PID-affected modules were studied detailedly under non-standard light conditions for the first time. The trend of efficiency degradation was obtained by measuring irradianceefficiency (I-η) and the ΔI-Δη. The result shows that the efficiency (η) of PID-affected module reduced by 1.28% which is rather higher than the 0.2% of PID-free module when the light intensity changes from 400 W/m2 to 200 W/m2. This means that efficiency of PID-affected module decreases more seriously in low light conditions.

In this paper, the I-V characteristics of PID-affected solar modules were investigated under illuminated and dark conditions. Based on the measured data, it is found that the I-V characteristic of PID-affected modules shows a regular change under different irradiations. Under dark conditions, the change of I-V characteristic of PID-affected modules disassembled from a power plant is also revealed. The dark I-V characteristic of PID-affected modules becomes soft. These results indicate that the efficiency of PID-affected modules has a significant degradation under low irradiance.

Predicting the gap between taxi demand and supply in taxi booking apps is completely new and important but challenging. However, manually mining gap rule for different conditions may become impractical because of massive and sparse taxi data. Existing works unilaterally consider demand or supply, used only few simple features and verified by little data, but not predict the gap value. Meanwhile, none of them dealing with missing values. In this paper, we introduce a Double Ensemble Gradient Boosting Decision Tree Model(DEGBDT) to predict taxi gap. (1) Our approach specifically considers demand and supply to predict the gap between them. (2) Also, our method provides a greedy feature ranking and selecting method to exploit most reliable feature. (3) To deal with missing value, our model takes the lead in proposing a double ensemble method, which secondarily integrates different Gradient Boosting Decision Tree(GBDT) model at the different data sparse situation. Experiments on real large-scale dataset demonstrate that our approach can effectively predict the taxi gap than state-of-the-art methods, and shows that double ensemble method is efficacious for sparse data.

Many Web applications need the accurate event detection technique on microblog stream. But the accuracy of existing methods is still challenged by microblog’s short length and high noise. We develop a novel category-level transfer learning method TransDetector to deal with the task. TransDetector bases on two facts, that microblog is short but can be enriched by knowledge base semantically with transfer learning; and events can be detected more accurately on microblogs with richer semantics. The following contributions are made in TransDetector. (1) We propose a structure-guided category-level topics extraction method, which exploits the knowledge base’s hierarchical structure to extract categories’ highly correlated topics. (2) We develop a probabilistic model CTrans-LDA for category-level transfer learning, which utilizes the word co-occurrences and transfers the knowledge base’s category-level topics into microblogs. (3) Events are detected accurately on category-level word time series, due to richer semantics and less noise. (4) Experiment verifies the quality of category-level topics extracted from knowledge base, and the further study on the benchmark Edinburgh twitter corpus validates the effectiveness of our proposed transfer learning method for event detection. TransDetector achieves high accuracy, promoting the precision by 9% without sacrificing the recall rate.

Ideology detection from text plays an important role in identifying the political ideology of politicians who have expressed their beliefs on many issues. Most existing approaches based on bag-of-words features fail to capture semantic information. And other sentence modeling methods are inefficient to extract ideological target context which is significant for identifying the political ideology. In this paper, we propose a target-specific Convolutional and Bi-directional Long Short Term Memory neural network (CB-LSTM) which is suitable in intensifying ideological target-related context and learning semantic representations of the text at the same time. We conduct experiments on two commonly used datasets and a well-designed dataset extracted from tweets. The experimental results show that the proposed method outperforms the state-of-the-art methods.

As the number of academic papers and new technologies soars, it has been increasingly difficult for researchers, especially beginners, to enter a new research field. Researchers often need to study a promising paper in depth to keep up with the forefront of technology. Traditional Query-Oriented study method is time-consuming and even tedious. For a given paper, existent academic search engines like Google Scholar tend to recommend relevant papers, failing to reveal the knowledge structure. The state-of-the-art Map-Oriented study methods such as AMiner and AceMap can structure scholar information, but they're too coarse-grained to dig into the underlying principles of a specific paper. To address this problem, we propose a Study-Map Oriented method and a novel model called RIDP (Reference Injection based Double-Damping PageRank) to help researchers study a given paper more efficiently and thoroughly. RIDP integrates newly designed Reference Injection based Topic Analysis method and Double-Damping PageRank algorithm to mine a Study Map out of massive academic papers in order to guide researchers to dig into the underlying principles of a specific paper. Experiment results on real datasets and pilot user studies indicate that our method can help researchers acquire knowledge more efficiently, and grasp knowledge structure systematically.

Social networking service (SNS) applications are changing the way information spreads in online communities. As real social relationships are projected into SNS applications, word of mouth has been an important factor in the information spreading processes of those applications. By assuming each user needs a cost to accept some specific information, this paper studies the initial “seed user” selection strategy to maximize information spreading in a social network with a cost budget. The main contributions of this paper are: 1) proposing a graphic SEIR model (gSEIR) by extending the epidemic compartmental model to simulate the dynamic information spreading process between individuals in the social network; 2) proposing a formal definition for the influence maximization problem with limit cost (IMLC) in social networks, and proving that this problem can be transformed to the weighted set-cover problem (WSCP) and thus is NP-Complete; 3) providing four different greedy algorithms to solve the IMLC problem; 4) proposing a heuristic algorithm based on the method of Lagrange multipliers (HILR) for the same problem; 5) providing two parts of experiments to test the proposed models and algorithms in this paper. In the first part, we verify that gSEIR can generate similar macro-behavior as an SIR model for the information spreading process in an online community by combining the micro-behaviors of all the users in that community, and that gSEIR can also simulate the dynamic change process of the statuses of all the individuals in the corresponding social networks during the information spreading process. In the second part, by applying the simulation result from gSEIR as the prediction of information spreading in the given social network, we test the effectiveness and efficiency of all provided algorithms to solve the influence maximization problem with cost limit. The result show that the heuristic algorithm HILR is the best for the IMLC problem.

LiFePO 4 /C and LiFe 1− x Ni x PO 4 /C ( x = 0, 0.02, 0.04, and 0.06) composites were prepared using solid-state reaction. The as-prepared composites were characterized by using X-ray diffraction, charge–discharge cycling, cyclic voltammograms, Raman spectroscopy, field emission scanning electron microscopy, high-resolution transmission electron microscopy, and other techniques. The experimental results show that all as-prepared composites have a single phase of orthorhombic olivine-type structure with the Pnmb space group. The as-prepared LiFe 1− x Ni x PO 4 /C composites exhibit high capacities and good cycling performance; e.g., the LiFe 0.98 Ni 0.02 PO 4 /C composite delivers 142 and 138 mAh g −1 at the 0.1 C rate for the first and fifth cycles, respectively. Such composites also show good rate capabilities; e.g., when discharged at the 2 C rate the LiFe 0.98 Ni 0.02 PO 4 /C composite delivers an initial capacity of 121 mAh g −1 , 85% of the discharge capacity at the 0.1 C rate. The reason why the LiFe 1− x Ni x PO 4 /C composites have better electrochemical performance compared to the LiFePO 4 /C composites is because nickel doping enhances the P O bond, stabilizes the structure, and thus the charge-transfer resistance and cathode particle resistance of the composites are decreased.

Microblogs contain the most up-to-date and abundant opinion information on current events. Aspect-based opinion mining is a good way to get a comprehensive summarization of events. The most popular aspect based opinion mining models are used in the field of product and service. However, existing models are not suitable for event mining. In this paper we propose a novel probabilistic generative model (ASEM) to simultaneously discover aspects and the specified opinions. ASEM incorporate a sequence labeling model(CRF) into a generative topic model. Additionally, we adopt a set of features for separating aspects and sentiments. Moreover, we novelly present a continuously learning model. It can utilize the knowledge of one event to learn another, and get a better performance. We use five real world events to do experiment. The experimental results show that ASEM extracts aspects and sentiments well, and ASEM outperforms other state-of-art models and the intuitive two-step method.

With more and more new social network services appearing, the volumes of data they created are continuous increasing at an astonishing speed. These data represent a snapshot of what real social network happening and evolving, and they contain the basic relationships and interacted behaviors among users. Core-based friend cycles are connected nodes around given "core node", and their interaction pattern with core node may reveal potential habits of users. This may be useful for online personalized advertising, online public opinion analysis, and other fields. To search core-based friend cycles by global method needs to scan the entire graph of social network every time, and thus its efficiency is low. This study (1) modeled the core-based friend cycles with core-based subgraphs;(2) provided algorithms to find structure and evolving interaction pattern of friend cycles around a given core node in online social network; (3) discussed and analyzed the design of incremental search algorithm theoretically; (4)applied the provided model to do informed prediction between node and its core-based friend cycles and received hit rate over 77.6%;(5) provided sufficient experiments and proven the newly proposed approach with good scalability and efficiency.

Extracting sentiments from unstructured text has emerged as an important problem in many disciplines, for example, to mine online opinions from the Internet. Many algorithms have been applied to solve this problem. Most of them fail to handle the large scale web data. In this paper, we present a parallel algorithm for BN (Bayesian Networks) structure leaning from large-scale dateset by using a MapReduce cluster. Then, we apply this parallel BN learning algorithm to capture the dependencies among words, and, at the same time, finds a vocabulary that is efficient for the purpose of extracting sentiments. The benefits of using MapReduce for BN structure learning are discussed. The performance of using BN to extract sentiments is demonstrated by applying it to real web blog data. Experimental results on the web data set show that our algorithm is able to select a parsimonious feature set with substantially fewer predictor variables than in the full data set and leads to better predictions about sentiment orientations than several usually used methods.