Human emergency behaviour and psychological stress response in emergencies are important scientific issues in basic emergency management research. The analysis of the dynamic characteristics of large-scale human behaviour based on electronic footprint data provides a new method for quantitative research on this problem. Previous studies usually assumed that human behaviors were randomly distributed in time, but few studies have studied the psychological stress response of human groups under the influence of emergencies and carried out prediction methods through social media data. Based on the data from five emergencies and daily events in the Qzone, this paper explores the statistical characteristics of human communication behaviors such as time, space and social interaction. The research results reveal the psychological evolution of human groups when they encounter public security emergencies by analysing the causes of individual psychological stress responses in the group. We find that the time interval between people’s posting behaviour and interactive comment behaviour in mobile QQ space before and after an emergency can be approximately described by a power-law distribution. The time interval distribution of Posting and reply is an obvious heavy-tailed distribution. These behavioural characteristics are consistent with people’s psychological stress characteristics. Individual psychological stress responses gradually evolve into social-psychological responses with changes in behavioural characteristics. The greater the social-psychological stress response is, the more panic the public will be, which will cause the outbreak of group irrational behaviour. The research results are theoretically helpful in understanding the impact of emergencies on human communication behaviour patterns and reveal the psychological stress process of mass panic in large-scale emergencies.

In a dynamic network, the characteristics of local nodes include first and higher-order proximity among the nodes as well as different attributes attached to each node. This complexity impose significant challenge for dynamic network modeling. As a result, few dynamic network studies have considered high-order proximity among local nodes. In this paper, we adopt the network embedding method to map high-order proximity of local nodes into low-dimensional, dense and real-valued vectors. Morevoer, we incorporate it into a model-based evolutionary clustering method through regularity conditions. Such a unified framework can increase the effectiveness and robustness of dynamic community detection while pertaining a good explanatory and visualization ability. Experiments based on synthetic and real world data sets show that our model can produce better community detection results than other popular models such as DECS and Genlouvain in dense networks. This result is consistent with the advantage of network embedding method in dense networks.

Dynamic network analysis is a promising research area with a wide range of applications. Criminal network investigation is one of them. People may be interested in using dynamic network analysis to detect criminal communities in a dynamic social network, track the evolution of those communities, identify critical criminal members, or predict links between criminal members and others. One difficulty in applying dynamic network analysis to real-world data is that real-world dynamic networks may vary sparse, which can cause overfitting problem and compromise the performance of the proposed model. Another problem is that each node in a complex real-world network may have multiple features, making it complicated to compute the distance between nodes. We propose a higher-order multiple-feature-based community evolution model (HFCE) to address those two issues. The model uses a higher-order representation of the neighbouring information among nodes to alleviate the sparsity problem. It also introduces first-order similarity regularization to clarify the distance between nodes with multiple features. Experiment results show that the HFCE model outperforms five other popular dynamic network models (ESPRA, AFECT, GenLouvain, ECD and DYNMOGA) in terms of community on the real-world sparse dataset detection and link prediction precision. The HFCE model can also effectively track the evolution of the communities and identify the important nodes in the network over time, which makes it a desirable model in criminal network investigation.

The low-power wide-area network (LPWAN) technologies, such as LoRa, Sigfox, and NB-IoT, bring new renovation to the wireless communication between end devices in the Internet of things (IoT), which can provide larger coverage and support a large number of IoT devices to connect to the Internet with few gateways. Based on these technologies, we can directly deploy IoT devices on the candidate locations to cover targets or the detection area without considering multihop data transmission to the base station like the traditional wireless sensor networks. In this paper, we investigate the problems of the minimum energy consumption of IoT devices for target coverage through placement and scheduling (MTPS) and minimum energy consumption of IoT devices for area coverage through placement and scheduling (MAPS). In the problems, we consider both the placement and scheduling of IoT devices to monitor all targets (or the whole detection area) such that all targets (or the whole area) are (or is) continuously observed for a certain period of time. The objectives of the problems are to minimize the total energy consumption of the IoT devices. We first, respectively, propose the mathematical models for the MTPS and MAPS problems and prove that they are NP-hard. Then, we study two subproblems of the MTPS problem, minimum location coverage (MLC), and minimum energy consumption scheduling deployment (MESD) and propose an approximation algorithm for each of them. Based on these two subproblems, we propose an approximation algorithm for the MTPS problem. After that, we investigate the minimum location area coverage (MLAC) problem and propose an algorithm for it. Based on the MLAC and MESD problems, we propose an approximation algorithm to solve the MAPS problem. Finally, extensive simulation results are given to further verify the performance of the proposed algorithms.

This paper presents a text feature extraction model based on stacked variational autoencoder (SVAE). A noise reduction mechanism is designed for variational autoencoder in input layer of text feature extraction to reduce noise interference and improve robustness and feature discrimination of the model. Three kinds of deep SVAE network architectures are constructed to improve ability of representing learning to mine feature intension in depth. Experiments are carried out in several aspects, including comparative analysis of text feature extraction model, sparse performance, parameter selection and stacking. Results show that text feature extraction model of SVAE has good performance and effect. The highest accuracy of SVAE models of Fudan and Reuters datasets is 13.50% and 8.96% higher than that of PCA, respectively.

Alow-dimensional embedding can be easily applied in the downstream tasks for network mining and analysis. In the meantime, the popular models of random walk-based network embedding are viewed as the form of matrix factorization, whose computational cost is very expensive. Moreover, mapping different types of nodes into one metric space may result in incompatibility. To cope with the two challenges above, a weighted meta-path embedding framework (WMPE) is proposed in this paper. On one hand, a nearly-linear approximate embedding approach is leveraged to reduce the computational cost. On the other hand, the meta-path and its weight are learned to integrate the incompatible semantics in the form of weighted combination. Experiment results show that WMPE is effective and outperforms the state-of-the-art baselines on two real-world datasets.

A low-dimensional embedding of multiple nodes is great convenient for clustering, which is one of the most fundamental tasks for heterogeneous information networks (HINs). In the meantime, the random walk-based network embedding is proved to be equivalent to the method of matrix factorization whose computational cost is very expensive. Moreover, mapping different types of nodes into one metric space may result in incompatibility. To cope with the two challenges above, a weighted meta-path embedding based clustering method (called WMPEClus) is proposed in this paper. On the one hand, in order to solve the incompatibility problem, the original network is transformed into several subnetworks with independent semantics specified by meta-paths which are automatically generated by our method. On the other hand, an approximate commute embedding approach, avoiding eigen-decomposition to reduce computational cost, is leveraged to the representation learning of the nodes in each subnetwork. At last, a unified probabilistic generation model is designed to aggregate the vectorized representations learned in different metric spaces for clustering. Experiment results show that WMPEClus is effective in HIN clustering and outperforms the state-of-the-art baselines on two real-world datasets.

Network Embedding (NE) is one of the most popular learning methods in complex networks. It aims at learning the low-dimensional representations of nodes in networks and has been applied in a variety of network analytic tasks. Most existing methods of NE are designed by merely using the local, high-order or global proximity to preserve the network structure; hence they are incapable of fully capturing the structural identity of nodes, which is a concept of symmetry defined by the network structure and their relationship to other nodes. There are two limitations to existing NE models. First, the local and global node dependency information is not considered simultaneously. Second, there is no adequate framework that can reveal the role property of each node. In this paper, we propose an intuitive and unified deep learning framework named DMER, short for Deep Mutual Encode for Embedding, to learn node embeddings from structural identity. In our model, Graph Convolution Network (GCN) is adopted to model the dependency relations between nodes from a global perspective. An Auto-Encoder (AE) framework is proposed to reconstruct the features of nodes, and it can conclusively reveal the structural identity from network structure. By integrating the GCN and AE components with a shared and constrained mechanism, the proposed model implements mutual enhancement for node embedding from structural identity. Experimental results based on structural role classification and visualization demonstrate that our model achieves better performance compared with the state-of-the-art methods.

近年来,强化学习在电子游戏、棋类、决策控制等领域取得了巨大进展,也带动着金融交易系统的迅速发展.金融交易问题已经成为强化学习领域的研究热点,特别是股票、外汇和期货等方面具有广泛的应用需求和学术研究意义.以金融领域常用的强化学习模型的发展为脉络,对交易系统、自适应算法、交易策略等方面的诸多研究成果进行了综述.最后讨论了强化学习在金融领域应用中存在的困难和挑战,并对今后强化学习交易系统发展趋势进行展望.

《佛地经论》法界说的核心是清净法界.《佛地经论》通过“十种清净相”以及自性相、因相、果相、作业相、相应相、差别相之“六相”,对清净法界进行了说明,前者意在阐明清净法界本性清净之原因所在,后者强调清净法界之本质体现为“返本还源”,并且以“清净法界”和“四智”的构造对“清净法界”向“如来三身”之转变进行了详细说明.《佛地经论》的法界说具有将如来藏思想与唯识思想相杂糅的特征,它处在新旧唯识思想的转折点上,被后来的中国华严宗思想家所吸收,成为华严宗法界思想的重要来源之一.

针对目前话题归类模型中文本逻辑结构特征与文本组织结构特征利用不充分的问题,该文提出一种面向文本结构的混合分层注意力网络的话题归类模型(TSOHHAN).文本结构包括逻辑结构和组织结构,文本的逻辑结构包括标题和正文等信息;文本的组织结构包括字—词语—句层次.TSOHHAN模型采用竞争机制融合标题和正文以增强文本逻辑结构特征在话题归类中的作用;同时该模型采用字-词语-句层次的注意力机制增强文本组织结构特征在话题归类中的作用.在4个标准数据集上的实验结果表明,TSOHHAN模型能够提高话题归类任务的准确率.

Given a paragraph of text, the objective of aspect-level sentiment classification is to identify the sentiment polarity of a specific phrase. Most exisiting work employed LSTM model and attention mechanisms to predict the sentiment polarity of the question targets. Unfortunately, these approaches haven't fully utilize the independent modeling of these target phrases. We propose a model based on TextCNN and Transformer pretrained model. In our model, the representations are generated for the targets and the contexts separately. We use Transformer model to help represent a target and its context via attention learning, which improves the performance of aspect-level sentiment classification. Experiments on COAE2014 and COAE2015 task show the effectiveness of our new model.

知识图谱嵌入是一种将实体和关系映射到低维向量空间的技术.目前已有的嵌入表示方法在对具有不对等特征的知识图谱中的实体和关系建模时存在两大缺陷:一是假定头尾实体来自同一语义空间,忽略二者在链接结构和数量上的不对等;二是每个关系单独配置一个投影矩阵,忽略关系之间的内在联系,导致知识共享困难,泛化能力差.该文提出一种新的嵌入表示方法T ransRD,首先对头尾实体采用不对等转换矩阵进行投影,并用ADADEL T A算法自适应调整学习率;其次对关系按相关性分组,每组关系使用同一对投影矩阵的方式来共享公共信息,解决泛化能力差的问题.在公开的数据集WN18和FB15K以及MPBC_20(乳腺癌知识图谱的子集)上进行实验和结果分析并与现有的模型进行对比,结果表明T ransRD在各项指标上均取得大幅提升.

With the progress of mobile communication technology, mobile social networks that can reflect the real situation of human social activities in real life have gradually formed. An even challenging problem whether it is possible to determine the kinship merely based on mobile social networks is raised during research on mobile social networks. This enlightens us the following research. First, we take the mobile phone contacts, telephone network and SMS network as the object of social network research. Second, we tackle the challenge of kinship mining using novel feature extraction and selection methods. Third, after selecting the most discriminative data features, we use a variety of algorithms to compare the study of the kinship in this paper. Finally, we demonstrate a classification accuracy of 81.04% on a test set using XGBoost.

微博网络中的信息传播模型是分析用户行为,找出传播路径,确定领袖人物,发现舆情热点等研究的基础.虽然多种不同角度的信息传播模型已经得到广泛研究,但缺乏对信息动态传播过程的直观描述.本文基于Petri网的结构和特征,针对微博网络信息传播过程,提出了一种简单直观的描述模型,该模型首先对微博网络的信息动态传播过程中的对象进行结构化描述.本文根据微博网络的用户结构关系,并利用Petri网的相关理论,形式化解释和定义信息传播基本对象,从而更加直接描述了微博网络中的转发、评论、回复等多种用户行为.在此基础上,本文利用Petri网能够描述信息流动问题的特征,结合颜色Petri网和时延Petri网,从Petri网系统的角度表示信息动态传播路径,并研究网络的动态性质和传播条件,使得信息传播模型更加真实地模拟信息传播情况.最后本文分析信息传播算例和新浪微博真实数据实验,验证了模型的有效性和可行性,为舆情态势分析以及用户行为的研究提供帮助和支持,同时也为其它社交网络信息传播的用户行为描述提供了新的思路.

网络的快速发展使得在线新闻媒体成为人们获取信息的重要来源.Web站点发布的信息是否能够反映当前关注的热点,是否能够及时发布事件的最新进展情况,对Web站点的可用性有重大影响.在利用条件随机场(CRF)模型识别主题线索句的基础上,得到与Web信息相关的同一主题的线索发展趋势.通过获得的线索发展趋势来推断主题线索的时间区间,进而估计出Web信息的有效区间,在此基础上结合时效性所包含的发布及时性和内容新鲜性两个方面来对Web信息时效性进行合理的评价.实验结果表明,所提方法在Web信息时效性评价上有较好的效果.

在西方的科学研究中,对研究者的立场和身份问题,一直存在着所谓"局内人信条".具体到宗教研究领域,研究者大体来说有两类,即信仰者和非信仰者.前者出于局内人视角,其研究往往关注的是研究本身是否对信仰者的信仰与宗教修行有所帮助,这种研究往往是可以超越时空限制的;后者出于局外人视角,其研究立足于现实,局限于一定的时空,更关注研究对象的客观性.所以,进行宗教研究时,明确研究者的身份以及研究立场至关重要,这便是所谓"局内人信条".在分析"局内人信条"理论的基础上,试图以佛教中的感应故事研究和禅宗公案研究为例,对宗教研究中的"局内人信条"进行探讨.

Information network is a kind of complex network with semantic information.The semantic community detection of information network is a new research direction.The complexity of community detection algorithm is increased by considering the node content.Therefore this paper proposes a label propagation algorithm which is suitable for dealing with large scale information network in linear time.Firstly,the latent Dirichlet allocation topic model is used to represent the node content.Secondly,the multiplicative model of content similarity and propagation influence is taken as the label propagation strategy.And the content and the network topology are combined naturally in the normalization.Thirdly,the algorithm updates the node label while the node and the vast majority of neighbors are not the same.Extensive experiments on 12 real-world datasets with varying sizes and characteristics validate the proposed method outperforms other baseline algorithms in quality.

With audit data requirements, taking for example the audit data of construction industry, the audit data integration requirements are analyzed and application architecture of business intelligence technology in the audit business is put forward. The modeling process and the populating process of data warehouse are detailed, and the scheme of data pre-processing is set out. The report display methods are proposed, such as proportion chart, tendency chart, index plots and etc. The application model and value of business intelligence technology in audit business are analyzed, which is based on data mining technology such as outlier detection. This research has been applied to National Twelfth Five-Year Plan Science and Technology Support Program and has achieved better analysis and prediction of audit.

The construction of sentiment lexicon plays an important role in text mining.In recent years,the lexicon annotating format gradually evolves from binary annotation to multiple annotation,and sentiment lexicons of a single specific domain have caught more and more attentions of researchers.However,manual annotation costs too much labor work and time,and it is also difficult to get accurate quantification of emotional intensity.Besides,the excessive emphasis on one specific field has greatly limited the applicability of domain sentiment lexicons[1].This paper implemented statistical training for large-scale Chinese corpus through neural network language model,and proposed an automatic method of constructing a multidimensional sentiment lexicon based on constraints of Euclidean distance group.In order to distinguish the sentiment polarities of those words which may express either positive or negative meanings in different contexts,we further presented a sentiment disambiguation algorithm to increase the flexibility of our lexicon.Lastly,we presented a global optimization framework that provides a unified way to combine several human-annotated resources for learning our 10-dimensional sentiment lexicon SentiRuc.Experiments show the superior performance of SentiRuc lexicon in category labeling test,intensity labeling test and sentiment classification tasks.It is worth mentioning that in intensity label test,SentiRuc outperforms the second place by 23 ％.