Traditional 3rd order Singular Value Decomposition (HOSVD) recommendation algorithm is based on mining the potential relationship among users, item labels and items. However, this method does not take the user's emotions into account. Based on the preference of user emotions extracted from the emoji expressions in the comments, this paper proposes a HOSVD recommendation algorithm that introduces the user's emotional preferences. This method classifies emoji expressions, weights them, and calculates the weighted sum of the emoji expressions of different types to represent the user's emotions. 3rd order tensor model is introduced to store three tuple data of user, user emotion and item, and HOSVD decomposition is applied personalized recommendations. In this paper, a numerical experiment is conducted on an empirical dataset of online Internet education. The results show that this method improves the accuracy and recall performance indexes better than the collaborative filtering algorithm based on items, PersonalRank algorithm and HOSVD algorithm based on user-tag-item triple information. When Top-1 recommendation is made, the accuracy and recall rate can reach 0.353 and 0.281. The work of this paper provides a reference for the personalized recommendation of mobile.

In order to investigate the guarantee structure of enterprises, this paper builds the two layer network model including short-term and long-term guarantee relationships based on multilayer network theory, and then conducts empirical analysis according to guarantee data of Chinese listed enterprises from 2014 to 2016. Empirical results show that, the distribution of total degrees of multilayer networks in different years obeys the power-law distribution, the degree correlation of multilayer networks is positively correlated, short-term guarantee networks and long-term guarantee networks all have scale-free feature in different years, the similarity between the short-term guarantee network and the long-term guarantee network is relatively low in the same year; the structure of the same layer guarantee network of two consecutive years is relatively stable, but the similarity between the same layer networks decreases with the increase of the lag phase, there is a weak positive correlation between betweenness centralities of different layer networks, but the top central node of a single-layer network is not that in other layer network, and the top central node of one layer network is changing with time.

In link prediction, the similarity indices based on random walk process often set the probability of particles transferring to adjacent nodes to be equal, but neglecting the influence of node degree on the transition probability. To save this problem, a link prediction algorithm of biased random walk with restart is proposed. Firstly, we redefine the transfer probability of particles by referring to biased random walk. Then we apply it to the random walk with restart to explore the effect of node degree on the transfer of particles. Finally, on the basis of biased random walk,the proposed index is compared with six classical similarity indices.The experimental results of six real data sets show that the prediction algorithm of biased random walk with restart has higher prediction accuracy than that the unbiased one, and is better than other similarity indices.

Random walk is a fundamental mechanism for studying dynamics on networks. Random walk on the weighted network has also been widely concerned. In this paper, the mean first-passage time and average trapping time are studied based on the spectral graph theory for random walks on weighted networks. This is different from the Ref. [11], which is concerned with the mean first return time. The main results can be stated as following: the mean first-passage time and average trapping time are related to the size of the network, the degree of trapping point, and the weighted parameter θ and the average degree of the network. Our simulation results are in good agreement with the analytical results.

This paper aims to study the topological properties of protein complex hypernetwork and to solve some practical problems based on topology characteristics of the hypernetwork, including the identification of key proteins of the network. Based on the obtained data set of protein complexes, a hypernetwork model of protein association relationship is constructed, in which each protein is represented by a node and each complex by a hyperedge. We study the topological featuresof the protein complex hypernetwork such as hyperdegree distribution, degree distribution, and sub-hypergraph centrality to identifying the key proteins. Furthermore, the results are verified by Online GEne Essentiality (OGEE) data set.

Data acquisition of four time series of day trading on the whole Chinese P2P lending market is completed for empirical analysis. Using BDS nonlinear test method, the nonlinear dependence of the Chinese P2P lending market is tested and confirmed,In addition, an empirical analysis about whether there exists long memory in Chinese P2P lending market is conducted, using the classical R/S analysis method and the modified R/S analysis method. Results show that time series of P2P lending is nonlinear and a maybe low-dimensional chaotic process. It is found that there doesn’t show any long memory characteristics in each of P2P time series. Still staying in the primary stage of development and evolution at present, the P2P lending market is on transition stage from simple linear system to complicated open giant system.

In recent years, the price of non-ferrous metal futures market has fluctuated greatly and the trading risk has been increasing. In this paper, the dynamic characteristics of the linkage relationship of non-ferrous metal futures prices are studied. Based on the analysis of the prices of copper, aluminum and zinc, the paper constructed two directed weighted networks of copper-aluminum-zinc futures price linkage relationship. The paper analyzed the moderate distribution and marginal power of price networks. Then the paper studied the network topology and its evolution characteristics such as distribution of median centrality, and proximity. The results show that the price linkage of copper, aluminum and zinc futures in 2008-2018 is relatively stable in a few key relationship models. Through the analysis of the marginal rights in the network, it is found that the copper-aluminum-zinc futures price linkage relation model tends to remain stable for a period of time. By analyzing the betweenness centrality and the closeness centrality of the network, itis found that the frequent occurrence of key media nodes has certain regularity, and it is directly related to the market price trend and the linkage characteristics of the three metals. Therefore, based on the above results, this paper proposes a method to identify the trend of price linkage effect, and puts forward effective suggestions for the application of this method.

The clustering algorithm of relational data is very important for communication research. Firstly, uses the iterative system to metaphorize the individual structure changing, expresses the distance between the output and the state, and also uses the node with the highest structural level sequence to represent the cluster; then introduces the Hausdorff distance into the DBSCAN algorithm. Thus, the summation operator that merges with the same structure and the union operator of the scale on the hierarchy become compressible. We use the data of complex network researchers to evaluate the effectiveness of the algorithm. The layer of cooperation network has different network structure; the keywords have high transmission efficiency in the level 2; the reciprocal relationship has the most effecton knowledge dissemination. These new findings prove that the algorithm can reflect the influence of the network structure on the propagation effect by introducing the compressible measure Hausdorff distance of the Hutchinson operator. The design idea of the algorithm is correct.

To improve the safety and reliability during flight training, the actual flight training network data was empirical investigated using the complex network theory. The relative size of maximum connected sub-graph size and network overall efficiency are adopted to investigate the network invulnerability under random and deliberate attack against node attack and edge attack separately. The simulation results indicate that FTN shows fundamental scale-free and small world network characteristics, and strong robustness to random attack and obvious invulnerability to the deliberate attack under node attack, while FTN shows partly robustness to both random attack and deliberate attack under edge attack. Thus, it can be seen that the invulnerability of FTN was maintained by a small number of key airports with high degree or betweenness, which will cause network paralysis with sharply decreasing efficiency and rapidly worsening connected reliability in case of its damage inactivation.

Social recommendation algorithm based on social network is a popular method in recommendation system at present. However, there are many relationships among users in real social networks, and each relationship has different effects on recommendation. Therefore, the introduction of a social relationship in recommendation will inevitably affect the accuracy of recommendation results. In this paper, based on the multi-subnet composited complex network model, a multi-relationship composited network is constructed on the user-item bipartite graph, and a mass diffusion recommendation algorithm based on multi-relationship composited network is proposed. The experimental results on the real datasets Epinions and FilmTrust show that the recommendation algorithm with two kinds of social relationships is better than the recommendation algorithm with one kind of social relationship and traditional mass diffusion algorithm in the accuracy of recommendation.

In this paper, the adaptive algorithm of optimally weighted stochastic pool network is studied. The mean square error is used as the output performance evaluation index of the stochastic pooling network. The recursive expressions of the least mean square (LMS) algorithm and the Kalman-LMS algorithm are derived. The related results show that, for the nonstationary case of varying variances of inputs, both adaptive algorithms can converge to the optimal solution of weight vectors. However, the Kalman-LMS algorithm not only has a fast convergence speed, but also the weight mean square deviation is optimal at each step. When the number of network nodes is small, Kalman-LMS can obtain a smaller mean square error. As the number of network nodes increases, the mean square error obtained by the two adaptive algorithms tends to be consistent.