To improve the ability to predict the evolution trend of public opinion, a public opinion evolution trend prediction model based on an improved wolf pack algorithm and optimized long-short term memory neural network is proposed. Use Halton Sequence to initialization to improve population diversity. Design step factor to perform Gauss-Sine perturbation transformation to improve wolf group exploration and development capabilities. Combine with the spiral in the whale optimization algorithm to improve the siege mechanism to enhance the local search ability of wolves. The bidirectional memory population is used to increase the cooperative ability of the wolf pack. The improved wolf pack algorithm (IWPA) is applied to the hyperparameter prediction of the LSTM neural network. Using keywords such as “COVID-19” and “Food Safety”, the experiment proves that the IWPA-LSTM neural network public opinion evolution prediction model has good accuracy and generality. The model is suitable for the prediction of various public opinion evolution trends.

Mission of this research is to better capture the complexity of large-scale financial systems and overcome the insufficiency of equilibrium based neoclassical finance and behavioral finance in revealing the mechanism of financial crisis and the emergence of order. This study started from the dilemma of existing theories and the motivation of introducting complexity science. Then advances of multi-agent simulation and complex network analysis are summarized and discussed as two fundimental instruments of complexity science. Research trends of complexity science in financial research, are proposed as the result of the discussions. This study provides methodology reference and analytical tools for the theoretical research and practical applications of complex financial systems in the new era.

In order to better understand the network, based on the similarity, let the nodes select multiple similar nodes to form similar node pairs. Through the Monte Carlo simulation results, a pairing algorithm based on the maximum node similarity and degree is proposed to discover the overlapping community structure of the network. Using multi-level most similarity to continue to optimize the community structure, find out the deep overlapping structure and sub-community structure of the network community. The proposed algorithm discovers the overlapping structure of the network based on the reason why the real network forms a community, and further optimizes the community structure, discovering the deep overlapping community structure of the network and its sub-community structure.

According to the complex correlation characteristics of the actual interdependent network and the important evaluation indicators in the network, five different coupling methods are proposed, and nine interdependent network models are established. Considering the information transmission, redistribution and cascading failures in the network, a cascading failure model based on betweenness artificial flow models is established. We based on the game theory, attack-defense game problems of critical infrastructure are analyzed from the perspective of complex network, and the robustness of various interdependent networks is analyzed. We discovered the preferences of game participants in the interdependent networks, providing decision support for the protection of infrastructure networks.

Scientific and reasonable cold chain logistics mode is the key to ensure the price and freshness competitiveness of agricultural products e-commerce. For the supply chain involving two agricultural products e-commerce, e-commerce self-run cold chain logistics and third party cold chain logistics provider, the Stackelberg game model that takes into account the competition between two agricultural products e-commerce, and the dominance of third-party cold chain logistics provider, as well as the evolutionary game model for the choice of two agricultural e-commerce cold chain logistics modes is constructed. The influence of cross price elasticity, freshness demand elasticity, and preservation input cost coefficient of self-run and third-party on the evolution path is revealed through the simulation. The results show that: if the cross price elasticity, freshness demand elasticity and the third-party preservation input cost coefficient increase respectively, self-run cold chain logistics is the stable evolutionary strategy, e-commerce tends to self-run cold chain logistics. If the self-run preservation input cost coefficient increases, the third-party cold chain logistics is the stable strategy, the e-commerce tends to cooperate with the third-party cold chain logistics provider.

In order to overcome the shortages of the particle swarm optimization algorithm, such as low accuracy, slow convergence and falling into local optima, a particle swarm optimization algorithm based on labor division and fuzzy control is proposed, which improves the algorithm by using the division of labor, parameter adaptive adjustment and simulated annealing with distance factors. Particles are divided into scout and rearguard ones, the former searches randomly and the latter learns from the best individual solutions as well as the best global solution to ensure the diversity of population and to accelerate the search. A sigmoid function is used to adjust the inertial weight and fuzzy logic is applied to balance exploration and exploitation capability of the algorithm. The best global particle is updated according to simulated annealing with distance factors taken into account, which improves the ability of the algorithms to jump out of the local optima. Simulation experiments on 25 standard test functions show that the improved algorithm has better performance in terms of convergence accuracy, speed and stability.

In order to better represent the rare entities with a small number of neighbors, this paper proposes a knowledge graph embedding model based on the nearest neighbors (NNKGE), which uses the K-Nearest Neighbor algorithm to obtain the nearest neighbors of the target entity as extended information. Based on this, the relational nearest neighbors-based knowledge graph embedding model (RNNKGE) is proposed. To generate an enhanced entity representation, the nearest neighbors of the target entity in relation are obtained by the improved K-Nearest Neighbor algorithm and encoded by the graph memory network. Through the analysis of the experimental results on the public datasets, the above two models outperform the benchmark model (CoNE) in the case of using only the nearest neighbor nodes, alleviating the data sparsity problem and improving the knowledge representation performance.

In order to identify important nodes in temporal networks, a node importance evaluation method is proposed in based on inter-layer neighborhood information entropy. Inspired by the directed flows model of temporal networks, the method introduces the parameter ω to fuse the inter-layer neighborhood topology information of node at adjacent snapshots, uses information entropy to describe the complexity of network structure, and also takes into account the global topological information. The effectiveness and applicability of the method is proved by using the SIR propagation model, Kendall correlation coefficient, Top-k metrics, and the proposed method is compared with six evaluation methods on six real datasets. The experimental results demonstrate that the method can more effectively identify the important nodes in the temporal network. Meanwhile, the identification of the nodes of with high importance is more accurate. In addition, the parameter ω can be adjusted to improve the evaluation effect of this method according to the topology of the temporal network. Last but not least, the time complexity of this method is O(mn), which is suitable for large-scale temporal networks.

In order to study the influence of higher-order structures on the evolution of cooperative behavior on scale-free networks, a network game model based on the Prisoner's Dilemma game is constructed. A second-order higher-order structure is introduced on the scale-free network, a triangular face game containing pairwise games is defined, and the higher-order structure parameters are used to link the pairwise game payoffs with the face game payoffs, and the influence of the higher-order structure on the evolution of cooperative behavior is analyzed through simulation experiments. The results show that when individuals with high connectivity prioritize cooperation and obtain high payoffs, other individuals with high connectivity will be prompted to choose cooperation, and once a stable "all-cooperative" triangular strategy structure is formed among individuals, the payoffs of each cooperator can be significantly increased, which in turn promotes the emergence of cooperative behaviors.