Components of CSI 300 are used as a stock pool to construct a quantitative stock selection model that can continuously beat the market. The first step starts with the fundamentals and 50 long-term dominant stocks are filtered out through a multi-factor scoring model. These companies are in good operating conditions, but they may be affected by market shocks in the short term. Therefore, in the second step, the support vector machine is introduced to analyze the long-term dominant stocks, and the top 10 stocks with the highest probability of rise are selected to hold. The cumulative return rate of the model during the period of 2015-2017 reaches 73.03%, the annualized rate of return is 20.05%, and the Sharpe ratio is 0.54, far exceeding the performance of CSI 300 over the same period.

A stock prediction method based on multiple recurrent neural network and deep learning is proposed in this paper. Aiming at predicting the rise and fall of stocks, the method in this paper extracts the features of news corpus information by distributed vector representation methods. Considering the natural of time-series stock related information and the persistence of news impact, multiple recurrent neural networks are used to collaborate process the features and stock trading information to obtain the history information embedding. Finally, the outputs of all recurrent neural networks are concatenated together to predict a stock. The data of Shanghai A-share market stock are used to do case study, which indicates that our method significantly outperforms the other baselines.

This paper proposes a cross-media bursty events detection method based on bursty words clustering. According to the events analysis, as Microblogs has a huge number of posts, users post or retweet Microblogs in anytime, it may spend fewer time detecting busty events than other platforms. However, many microblogs are advertisements and worthless, which leads to a lower precision. On the contrary, as an official media, news is highly authentic and authoritative, and contents of news are more standard. Therefore, events detection has a higher accuracy. However, due to the small number of news, the efficiency of busty events detection is low. At present, all of the existing detection methods only mine the data of one media, which face with a dilemma between efficiency and accuracy. In this paper, the proposed model fuses the data of two medias, microblog and newssin order to meet the needs of efficiency and improve the accuracy of emergency detection.

Relation extraction is an important part of many information extraction systems that mines structured facts from texts. Recently, deep learning has achieved good results in relation extraction. Attention mechanism is also gradually applied to networks, which improves the performance of the task. However, the current attention mechanism is mainly applied to the basic features on the lexical level rather than the higher overall features. In order to obtain more information of high-level features for relation predicting, this paper proposes high-level semantic attention-based piecewise convolutional neural networks (PCNN_HSATT), which adds an attention layer after the piecewise max pooling layer in order to get significant information of sentence global features. Furthermore, this paper puts forward a data augmentation method by utilizing an external dictionary HIT IR-Lab Tongyici Cilin (Extended) to reply the sparse challenge in Chinese entity relation extraction corpus. Experimental results on COAE2016 and ACE2005 Chinese datasets are 78.41% and 73.94% respectively. Compared with the best existing method SVM, the results improve 10.45% and 0.67% respectively, which demonstrates that this approach is effective in Chinese entity relation extraction task.

Stance detection aims to analysize from the text whether the text author is in favor of, against or neutral to the given target, which plays an important part in public opinion analysis. Target dependent stance detection is a challenging task because of the idiomatic phrase and limited contextual information. It is noted that existing methods can not sufficiently model the whole semantic of target and those methods can not jointly model the target and context. To tackle these problems, a two-stage attention model for stance detection is proposed in this paper. Firstly, apply attention mechanism to model target, then match the context with the target representation to obtain attention signal, and finally form the target blended text representation for text classification. Experimental results show the proposed model improves accuracy and F-score by 0.4% and 1.0% respectively on Xinjiang anti-terrorist corpus, and obtain the state of the performance for 4 targets on the NLPCC-2016 stance detection task dataset.

Multi-task learning utilizes the similarity between different tasks to help decision making. Compared with single-task learning, multi-task learning can use more information, which can make up for the deficiency of single-task learning in the use of information. In this paper, a NTCIR-ECA dataset, which contains Chinese and English text data is used as the date in the experiment. The emotional cause analysis is regarded as the research task and a multi-task deep learning model (MTDLM) which combines multi-task learning and deep learning is presented. Finally, this model is used to do the emotional cause analysis in different languages. The experimental results show that in the case of unbalanced data, the optimal F value of the MTDLM model for English language emotion recognition is 39%, superior to single task learning (F value is 0) and traditional baseline model (LR, F value is 33%). The validity of the model is thus verified.

Learning to rank has been attracted much attention in the domain of information retrieval and machine learning. A series of learning to rank algorithms have been proposed based on three types of methods, namely, pointwise, pairwise and listwise. Especially, ranking performance can be improved effectively by one of the listwise methods named group ranking. This paper explores how to combine the loss functions from these methods to improve group ranking performance. The basic idea is to incorporate the different loss functions and enrich the objective loss function based on neural networks. Firstly, a group learning to rank method based on Jeffrey’s divergence is presented. Secondly, a framework for loss function incorporation based on group ranking method and the other loss function is presented. The performance of the proposed method is compared on LETOR3.0 dataset, which demonstrates that with a good weighting scheme. Finally, experimental results show that the proposed method significantly outperforms the baselines which use single loss function, and it is comparable to the state-of-the-art algorithms in most cases.

For the issue of multi-label classification, most existing methods only take the neighborhood information into consideration and ignore the structural similarity, leading to the low accuracy of classification. Therefore, a deep autoencoder for multi-label classification is proposed in this paper. In order to capture the global network structure, this method uses orbit counting algorithm (Orca) to calculate structural similarity of each node, which is the input information of the representations in the latent space. Then, the highly-nonlinear network structure can be well preserved by jointly optimizing the global structure and the neighborhood structure in the proposed model. Finally, SVM is used to classify the nodes according to the nodes vectors obtained from the latent space. Three real-world networks are used to conduct the experiment and the results show that the new model outperforms the state-of-the-art methods in multi-label classification.

A large number of electronic health records (EHRs) have been accumulated since the wide adoption of medical information systems in China. However, most of these records are written in natural language, which cannot be processed by computers directly. Thus, it is important to transform unstructured EHRs into structured ones. In this paper, a recurrent capsule network is proposed for clinical relation extraction in EHRs, where entity pairs and their contexts are captured by piece-wise recurrent neural network layers, and capsule layers are finally employed for relation classification. Experimental results show that this model performs better than the existing supervised methods, achieving a F1-score of 96.51%.

Bilingual parallel corpus is an important resource for multilingual natural language processing. It has been widely used in the fields of machine translation, machine-assisted translation, translation knowledge extraction and cross-language information retrieval. In this paper, the automatic alignment of Chinese-Indonesian parallel corpus and the automatic extraction of comparable corpus are proposed. Firstly, a paragraph alignment method based on the combination of anchor point and dictionary is proposed. On this basis, the length alignment model based on confidence interval is used to achieve sentence alignment. At the same time, in order to quickly improve the construction efficiency of the Chinese-Indonesian parallel corpus, a comparable corpus extraction method based on the similarity of cross-language documents is proposed. The experimental results show that the accuracy of parallel corpus alignment method and comparable corpus extraction method is significantly higher than that of traditional methods, which indicates that the proposed method is effective and feasible.

It is important to study the changes of the private cars’ trip rates under different policies, which may provide theoretical basis for the policy making to reduce private cars’ trip rate effectively. Questionnaire is designed based on SP survey and RP survey combined with the actual situation of Guangzhou, and the situation of the simultaneous change of multi-factor is also considered in the design of the questionnaire. The sample data are randomly selected among people who have private cars or plan to buy private cars recently in Guangzhou. A binomial logit selection model for different travel conditions is established after cluster analysis of the investigated population, and then the change of residents’ travel rate under the change of policy is calculated by the model. Finally, it is concluded that the average number of commuters who choose private cars for commuting trips is mostly affected by various factors, and commuters are more sensitive to fuel costs than bus departure intervals and parking fees.

Commuter passengers are people who travel during the rush hours in the morning and evening and have a regular pattern of travel. Accurate identification of commuter crowds from bus credit card data is of great significance for taking measures to alleviate traffic congestion during the rush hours in morning and evening and for overall urban line network planning and adjustment. Based on the data of Zhuhai bus IC card, this paper proposes a public transportation identification method based on differential evolutionary algorithm to optimize support vector machine (SVM). Firstly, the commuter passenger survey data are combined with the actual swipe data to analyze the characteristic attributes of commuter passenger travel. Then the SVM algorithm is used to build the classification recognition model, and a differential evolution algorithm (DE) is used to optimize the parameters of the SVM to obtain the optimal identification model, whose recognition accuracy is as high as 94.28%, better than other algorithm models. Finally, the model is used to identify the commuters in the Zhuhai bus IC data. The results show that the number of public transportation personnel is 178,259, accounting for 21.47% of the total number of bus trips.

This paper investigates the influence of average degree and scale of network on the partial synchronization state of the complex networks. The research results show that average degree may have effects only when the networks are in partial synchronization states, whether the network model is random network, small-world network or uncorrelated configuration network. Increasing the average degree can improve the partial synchronization state for the three network models. However, this can make the complexity change in completely different ways at different coupling strength regions. As to the network scale, it may have effects on small coupling strength: increasing the network scale may worsen the synchronization state and decrease the complexity. For the nearest-neighbor network, with the increase of average degree, the synchronization state of the whole network may be better and the complexity may be larger; and with the increase of network scale, the synchronization state of the whole network may be worse and the complexity may be smaller. This research makes the partial synchronization state clearer and proposes useful suggestions for the construction of multi-function networks.

A low temperature coefficient and high PSRR bandgap reference is presented in this paper. To temperature drift, a dynamic threshold MOS transistor is used to provide compensation current. A compact low-pass filter is introduced, the noise at high frequencies is reduced, and then a high PSRR over a wide frequency range is achieved. The proposed reference is implemented in SMIC 0.18μm CMOS process. The simulation results show that the temperature coefficient is 1.54×10^-6 ℃^-1, and an output reference voltage is 1.154 V for temperatures from -40 ℃ to 130 ℃ with supply voltage 1.8 V.The PSRR is -76 dB at 10 Hz, -85 dB at 100 kHz, and -63 dB at 15 MHz. The reference source has good comprehensive performance, which can provide high-precision reference voltage for digital-to-analog conversion circuit, analog-to-digital conversion circuit and power management chip, and has great application value.

Based on the topology of the power grid and the power tracing technique, a critical node identification method is proposed. Firstly,according to the result of the power flow calculation, the power flow direction between nodes can be obtained and then the power grid can be traced,so as to get the link strength matrix of nodes and establish the weighted directed network model for the power grid.Accordingly, the outbound and inbound strengths of the nodes as well as an evaluation index of nodes can be defined according to their importance based on the node strength and load weight.Taking IEEE39 system and IEEE14 system as testing cases, the importance of the nodes in each system is ranked. According to the results of ranking, the nodes will be overload attacked. To verify whether the order of importance of the nodes is reasonable,the change of power flow entropy is calculated after the node is overload attacked. The results show that the proposed method is more reasonable and effective in identifying the critical nodes of the power grid.

With small sample data sets, the virtual sample generation technology has been proved to effectively improve the performance of machine learning algorithm. However, there is no definite conclusion for the optimal generation number. First of all, under the condition of the limit of standard variance of a given training sample, the information entropy theory is proposed to study the number of optimal virtual sample generation. In addition, the noise generated by virtual sample generation is taken into account and a general probability model and the analysis method of the number of optimal virtual samples are established at a given confidence level (0.95). A small sample data set is set up based on the historical monitoring fault data of a substation in Huzhou, Zhejiang, in 2016 and a four virtual sample generation experiment is designed. The results show that the two optimal virtual sample generation rules are effective, and the accuracy of the corresponding machine learning prediction is obviously improved.

A subgroup H is said to be nearly SS-embedded in a finite group G if there exists an s-permutable subgroup T of G such that HT is s-permutable in G and H∩T≤H_seG,where H_seG is the subgroup contained in H,generated by all those subgroups of H which are s-permutably embedded in G. Let M_d(P)={P₁,P₂,…,P_d} be a set of the maximal subgroups of a group P of prime power order such that ∩^d_i=1P_i=Φ(P). The influence of the set M_d(P) with the above property on the p-nilpotency of finite groups is investigated, and some recent related results are generalized.

In this paper, the notions of FI-gr-injective modules and strongly FI-gr-injective modules are introduced, and the relationship between them and graded injective modules is explained. It is proved that the graded ring R is a gr-QF ring if and only if each graded module is a strongly FI-gr-injective module; suppose R is a left gr-coherent ring, then l.FP-gr-dim(R)≤1 if and only if each FI-gr-injective module is a graded injective module. In addition, it is also proved that l.gr-fiD(R)=sup{gr-pd(L)|L is an FP-gr-injective module}.

Based on density functional theory (DFT), the primary structures of cluster Ni₃CoP were optimized in singlet and triplet under B3LYP/lanl2dz level. According to the frontier molecular orbital theory (FMO), the mechanism of the hydrogen evolution reaction catalyzed by cluster Ni₃CoP was studied theoretically from aspects of the frontier molecular orbital images and the difference between the frontier molecular orbital energy levels. Results showed that cluster Ni₃CoP adsorbs H atom through electrons flowing from the highest occupied molecular orbital (HOMO) of cluster Ni₃CoP to the lowest unoccupied molecular orbital (LUMO) of H₂O. As for configurations in triplet, β-HOMO plays a prominent role in the reaction; however, their catalytic properties can descend apparently in the desorption of H atom to generate H₂. Nevertheless, structure 1⁽¹⁾, the only configuration in singlet, can not only perform relatively good catalytic capability in adsorbing H atom from H₂O but also improve the desorption process greatly compared with other configurations in triplet.

First-principles theoretical study of the elastic and thermodynamic properties of RbCaCl₃ in perovsklte structure is performed by using the pseudopotential plane-wave method and quasi-harmonic Debye model. The calculated lattice parameter, elastic constants, bulk modulus B and shear modulus G at 0 GPa and 0 K are conducted based on the optimized structure. The results are in agreement with the experimental values. Then the B/G value of the RbCaCl₃ crystal at zero pressure is computed, and from the values of the elastic constants at high pressure, the transition pressure is about 10.5 GPa judged by mechanical stability conditions of cubic crystals. The Debye temperature at 300 K and the values for relative volume, heat capacity, thermal expansion coefficient and Debye temperature at different pressures and temperatures are also obtained by applying the quasi-harmonic Debye model. It is found that the heat capacity at constant volume C_v is close to the Dulong-Petit limit at high temperature.

Anti-inflammatory and analgesia effects were observed by methods of acetic acid writhing, ear edema induced by xylene and hot-plate respectively. The results showed that the alcohol extracts from barks of Shima superba could significantly inhibite the number of writhing induced by acetic acid. The rates of anti-inflammatory and analgesia effects of 3 groups which treated mice with extracts from barks of S. superba (200 mg·kg^-1, 400 mg·kg^-1, 800 mg·kg^-1) were 37.9%, 52.2% and 66.7%. The rate of anti-inflammatory and analgesia effects of aspirin groups (200 mg·kg^-1) was 37.4%. The alcohol extracts from barks of S. superba could reduce the ear edema induced by xylene. The rates of inhibition ear edema of 3 groups which treated mice with extracts from barks of S. superba (200 mg·kg^-1, 400 mg·kg^-1, 800 mg·kg^-1) were 35.1%, 56.8% and 69.3%. The rate of inhibition ear edema of aspirin groups (200 mg·kg^-1) was 54.9%. The alcohol extracts from barks of S. superba could significantly enhance the pain threshold on hot-plate in mice. The rates of enhance the pain threshold of 3 groups which treated mice with extracts from barks of S. superba (200 mg·kg^-1, 400 mg·kg^-1, 800 mg·kg^-1) when 60 min after the last administration were 50.1%, 63.3% and 78.3%. The rate of enhance the pain threshold of aspirin groups (200 mg·kg^-1) was 63.6%. So the alcohol extracts from barks of S. superba have obvious effect on anti-inflammatory and analgesia.

Based on the daily air quality index (AQI) data of 367 cities published by the Chinese Ministry of Environmental Protection from 2014 to 2016, a series of spatial data statistical models such as spatial autocorrelation model, kernel density estimation model and spatial interpolation method are comprehensively used. The results show that: ①AQI of cities in China decreased from 2014 to 2016, and the number of cities with air pollution decreased, indicating that the air quality of cities had been improved. The seasonal mean AQI of China is as follows: winter, spring, autumn, summer. The pattern of spatial heterogeneity is not obvious in summer and autumn, while the spatial pattern is obvious in spring and winter, showing a special seasonal distribution pattern of high level in the north and low level in the south, and high level in inland and low level along the coastal. ②The spatial distribution of AQI spatial distribution in China′s cities shows a continuous enhanced spatial agglomeration situation, forming a moderate polluted area with the core of the three provinces of Hebei, Shandong and Henan provinces and spreading to the middle north of Hubei province, and the medium pollution area with Urumqi and Artux city as the core. The Southern China area with the core of the Pearl River Delta city group is the stable area of better air quality. ③From the kernel density estimation map, it can be found that the spatial distribution pattern of China′s urban AQI is characterized by six main density core regions and three secondary core density regions, while the basic pattern has changed significantly in some areas in 2014-2016. This paper comprehensively analyzes the spatial distribution characteristics of urban air quality based on different time scales, aiming to provide scientific basis for developing relevant air quality control policies in various cities of China.

In order to evaluate the landscape stability in mountainous area of southwest China, this paper analyzed the landscape stability in middle and upper reaches of the Wujiang River from three aspects of landscape matrix stability, landscape patch stability and landscape density stability based on the Landsat remote-sensing images in 2000 and 2015. The results showed that the landscape stability decreased during the period from 2000 to 2015. Landscape patch stability showed the characteristics of bare land>shrubbery>grass>water>forest>plantation>built-up land. Landscape density stability showed the characteristics of bare land>shrubbery>grass>forest>water>built-up land>plantation. There were significant spatial heterogeneities for landscape matrix stability, landscape patch stability and landscape density stability in each watershed. Ecological projects and rapid economic development are dominant factors affecting landscape stability in middle and upper reaches of the Wujiang River.

To discuss the abnormalities in peripheral blood chromosomal karyotypes and amniotic fluid karyotypes, as well as their application in clinical analysis. Peripheral blood G-banding technique was used to analyze the chromosomal karyotypes of the genetic counseling patients from 181st Hospital of Chinese People’s Liberation Army and Shenzhen People’s Hospital. 1 868 samples of peripheral blood and 541 samples of amniotic fluid were collected for the preparation of chromosome specimens. 15 abnormal karyotypes of peripheral blood were found in peripheral blood chromosomes. 84 abnormal karyotypes were found in amniotic fluid chromosome with the screeing positive rate of 15.52%. There are 99 cases of abnormal chromosome karyotypes. The results showed that Chromosome examination can detect some related chromosome diseases timely, in order to identify its genetic diagnosis more clearly, avoid blind treatment, and reduce the birth of children with chromosomal abnormalities effectively.

The polymorphisms of HSP70 in the GIFT tilapia samples were detected by using PCR-SSCP technique and the correlation with cold tolerance capability of GIFT tilapia was analyzed by least square method in this study. The results showed that the electrophoresis diagram of the primer HP70-1, HP70-4, HP70-6 designed according to the HSP70 has the polymorphism of the belt type in different cold tolerance samples. The polymorphism of the gene was significantly correlated with its cold tolerance. Therefore, it is inferred that HSP70 genes are potential cold-resistant genes of tilapia.

From November 2016 to April 2017, pre-release training and reintroduction experiments of Syrmaticus ellioti were conducted in Maoershan National Nature Reserve,Guangxi. During the training and reintroduction period, the food selection of birds, including species and preferencs, was studied by artifical feeding and radio-tracking techology. The results indicated that fruits, buds and leaves of the plant, which belong to 18 families, 28 species, were found in their diet. Besides,four orders of insects,such as Lepidoptera, Orthoptera, Hymenoptera and Coleoptera, were also found eaten by this pheasant.