Method for optimizing multi-language complex network behavior prediction based on evolutionary calculation algorithm
By applying the evolutionary calculation algorithm in complex network behavior prediction, uniformly processing multi -language data and optimizing the selection of features, the dynamic, non -linearity and uncertainty of complex network behavior prediction in the multi -language environment are solved, and higher predictions are achieved Accuracy and model adaptability.
Patent Information
- Application Number
- CN202411361125.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-01-28
AI Technical Summary
In a multilingual big data environment, complex network behavior prediction faces dynamic, nonlinear and highly uncertain problems. Traditional methods are difficult to effectively capture cross-language behavior characteristics, and lack effective feature selection and redundant data removal mechanisms, which affects the prediction. accuracy.
Using a method based on the evolutionary computing algorithm, the unified processing of symbols, semantics, and formats of multi -language complex network behavior data, through the evolutionary computing algorithm selection and optimization data characteristics, adapting to adjust the parameters of neural network architecture and superstaria, realize the dynamic model of the model to realize the model of the model optimization.
Effectively capture the data characteristics of different languages, improve the accuracy of forecasting in the cross -language environment, remove redundant information and noise data, improve data quality and calculation efficiency, and enhance the adaptability and generalization of models.
Smart Images
Figure CN119377754A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of network behavior prediction, and in particular, relates to a method for optimizing multi-language complex network behavior prediction based on evolutionary computing algorithms. Background Art
[0002] With the advancement of globalization and the widespread use of multilingual social platforms, the demand for the application of behavioral analysis of complex networks in multilingual big data environments is growing. Complex networks, such as social networks, economic networks, and communication networks, are usually composed of nodes (individuals) and edges (interaction relationships), and their behaviors are dynamic, nonlinear, and highly uncertain. In a multilingual context, data sources are not only diverse but also involve different languages, which makes the prediction of network behavior more complex and challenging. Multilingual data contains a variety of semantic and cultural differences. Traditional single language or specific data set analysis methods are difficult to meet the needs of multilingual big data, cannot fully capture cross-language behavioral characteristics, and cannot maintain consistency and robustness of predictions in different language environments.
[0003] In addition, as the scale of multilingual data continues to expand, the data contains a large amount of redundant information and noise, and existing complex network analysis methods often lack effective feature selection and redundant data removal mechanisms. This redundant and noisy data not only affects the computational efficiency of the model, but may also reduce the accuracy of the prediction results. At the same time, the nodes and structures of complex networks often change dynamically, and traditional static analysis methods are difficult to adapt to such changes, resulting in the inability of the prediction model to timely reflect the real behavior changes of the network. In addition, complex network behavior exhibits a high degree of nonlinearity and randomness. Existing prediction models are difficult to achieve the best results through fixed parameters or manual parameter adjustment. A more adaptive method is needed to optimize model parameters to improve prediction accuracy. There are deficiencies in the existing technology. Summary of the invention
[0004] The purpose of this application is to provide a method for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithms, aiming to solve the technical problems caused by the high uncertainty of existing complex networks and the impact of multilingualism, redundant information and noise on network behavior prediction.
[0005] On the one hand, the present application provides a method for optimizing multilingual complex network behavior prediction based on an evolutionary computing algorithm, the method comprising the following steps:
[0006] s1. Unify the symbols, semantics and formats of network behavior data based on different languages to make it suitable for behavior prediction models of complex networks;
[0007] s2. Use evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data;
[0008] s3. Adaptively adjust the neural network architecture parameters of the behavior prediction model based on the dynamic changes of nodes and edges in the complex network through evolutionary computing algorithms to cope with the changes of the complex network;
[0009] s4. Aiming at the nonlinearity and uncertainty of behavior in complex networks, the configuration hyperparameters of the behavior prediction model are iteratively optimized through evolutionary computing algorithms to achieve automatic tuning of the model;
[0010] The complex network includes social networks, economic networks and communication networks; the nodes represent individuals in the complex network; the edges represent the interaction relationships between the individuals; the neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.
[0011] On the other hand, the present application also provides a system for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithms, using any of the above methods; the system comprises:
[0012] The data preprocessing module unifies the symbols, semantics and formats of network behavior data based on different languages to make it suitable for the behavior prediction model of complex networks;
[0013] The evolutionary computing feature selection module uses evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data;
[0014] The dynamic network analysis module uses evolutionary computing algorithms to adaptively adjust the neural network architecture parameters of the behavior prediction model according to the dynamic changes of nodes and edges in the complex network to cope with the changes of the complex network;
[0015] The adaptive optimization module, targeting the nonlinearity and uncertainty of behaviors in complex networks, iteratively optimizes the configuration hyperparameters of the behavior prediction model through evolutionary computing algorithms to achieve automatic tuning of the model;
[0016] The complex network includes social networks, economic networks and communication networks; the nodes represent individuals in the complex network; the edges represent the interaction relationships between the individuals; the neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.
[0017] This application aims to address the dynamics, nonlinearity and high uncertainty in multilingual complex networks. By uniformly processing and extracting features from multilingual data and optimizing feature selection using evolutionary computing algorithms, this ensures that data features in different language contexts can be effectively captured, thereby improving prediction accuracy in cross-language environments. Secondly, through the global search and optimization capabilities of evolutionary computing algorithms, redundant information and noise data can be efficiently screened, data quality can be improved, computing costs can be reduced, and the overall prediction accuracy of the behavior prediction model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 This is a main implementation flow chart of the method for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithm provided in Example 1 of the present application;
[0019] Figure 2 It is the predicted effect diagram obtained by the experiment;
[0020] Figure 3 This is the main architecture block diagram of the system for optimizing multi-language complex network behavior prediction based on evolutionary computing algorithms provided in Example 2 of the present application. DETAILED DESCRIPTION
[0021] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0022] The specific implementation of this application is described in detail below in conjunction with specific embodiments:
[0023] Embodiment 1:
[0024] Figure 1 The implementation process of the method for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithm provided in the first embodiment of the present application is shown. For the sake of convenience, only the part related to the embodiment of the present application is shown, which is described in detail as follows:
[0025] On the one hand, the present application provides a method for optimizing multilingual complex network behavior prediction based on an evolutionary computation algorithm, the method comprising the following steps:
[0026] s1. Unify the symbols, semantics and formats of network behavior data based on different languages to make it suitable for behavior prediction models of complex networks;
[0027] s2. Use evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data; this can effectively avoid the problem of directly using multilingual big data that contains a large amount of redundant information and noise data, which will reduce the prediction accuracy of the behavior prediction model.
[0028] s3. Adaptively adjust the neural network architecture parameters of the behavior prediction model based on the dynamic changes of nodes and edges in the complex network through evolutionary computing algorithms to cope with the changes of the complex network;
[0029] s4. Aiming at the nonlinearity and uncertainty of behavior in complex networks, the configuration hyperparameters of the behavior prediction model are iteratively optimized through evolutionary computing algorithms to achieve automatic tuning of the model;
[0030] The complex network includes social networks, economic networks and communication networks; the nodes represent individuals in the complex network; the edges represent the interaction relationships between the individuals; the neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.
[0031] Through the above steps, we can effectively deal with the challenges brought by the diversity and complexity of data to behavior prediction in a multilingual big data environment.
[0032] Furthermore, the step s1 comprises:
[0033] s11. Through natural language processing technology, stop words, punctuation marks, special symbols and other noise information in the data are removed to clean the text in the network behavior data.
[0034] Among them, stop words include: and.
[0035] s12. Use the pre-trained multilingual word embedding model to map network behavior data in different languages but with the same semantics into the same semantic space, ensure the consistency of multilingual network behavior data in the semantic space, and achieve multilingual semantic normalization;
[0036] Among them, the multilingual word embedding model preferably uses the multilingual version of the BERT model.
[0037] s13. Vectorize text data so that it can be used as input for complex network analysis models and achieve data formatting.
[0038] The above steps process multilingual data by cleaning, formatting, and normalizing, ensuring that data in different languages can be used in the same model. This also helps improve the cross-language data processing capabilities of the behavior prediction model.
[0039] Further, the step s2 comprises:
[0040] s21. Randomly select network behavior data in different languages for feature extraction as an initial feature set, wherein the feature set includes several feature combinations;
[0041] s22. For the features contained in each feature combination, use the behavior prediction model to predict the node behavior in the complex network, use cross-validation to evaluate the prediction accuracy of the feature combination, calculate the fitness value, and complete the fitness function evaluation;
[0042] s23. Select the population according to the fitness value, generate new feature combinations through crossover and mutation operations, and complete the genetic operation;
[0043] s24. Repeat steps s23 and s22 until the optimal feature combination that satisfies the prediction accuracy is found or the preset number of iterations is reached.
[0044] Steps s21 to s24 can effectively remove redundant features in multi-language network behavior data, improve the accuracy and computational efficiency of the behavior prediction model, and also help improve the efficiency of hardware resource utilization.
[0045] Further, the step s3 comprises:
[0046] s31. Real-time monitoring of the dynamic changes of complex network nodes and edges, and recording changes in network topology;
[0047] s32. When the structure of a complex network changes, re-evaluate the performance of the existing behavior prediction model and find new parameter configurations through evolutionary algorithms so that the behavior prediction model can dynamically adapt to changes in the network structure and complete parameter optimization;
[0048] s33. Based on the optimized parameters, the behavior prediction model is updated in real time so that it can make behavior predictions based on the latest structural changes of the complex network, ensure the accuracy and timeliness of the predictions, and complete the dynamic update of the behavior prediction model.
[0049] Further, the step s4 comprises:
[0050] s41. Through the global search mechanism of evolutionary computing algorithm, explore various parameter configurations and feature combinations in the parameter space, ensure that the search process covers different areas of the parameter space, and complete the global search;
[0051] s42. Deeply optimize local areas of the parameter space to ensure that the final parameter settings can adapt to the behavior patterns of complex networks in a multilingual environment
[0052] s43. After each behavior prediction model optimization, the feedback mechanism is used to evaluate the prediction effect of the behavior prediction model; if the effect is not ideal, the system will continue to optimize the parameters in step s42 until the prediction effect reaches the optimal state.
[0053] The above steps s41 to s43 help to prevent the prediction model from falling into a local optimal solution.
[0054] Furthermore, the evolutionary computing algorithm adopts a distributed computing strategy to perform parallel evolutionary solution calculation; specifically:
[0055] y1. Represent the problem’s variables and constraints in a form that the evolutionary computation algorithm can handle, and divide the scope of its global search solution space; encode the problem to be solved.
[0056] y2. Use evolutionary computing algorithms to perform a global search of the entire solution space, and continuously optimize the solutions in the feature set through selection, crossover, and mutation operations;
[0057] y3. Use evolutionary computing algorithms to select multiple solutions with better performance; these solutions represent potential high-quality solutions in the global solution space.
[0058] y4. Calculate the position differences between the high-quality solutions in the solution space, and select high-quality points that are far away from each other as the starting points of the gradient descent method to increase the coverage of the global optimal point.
[0059] y5. Adopt the strategy of multiple starts, and use the selected high-quality points as the starting point of an independent gradient descent process to perform local search optimization;
[0060] y6. After multiple rounds of iterations, the solution with the highest prediction accuracy and the best solution is selected from these points as the final solution to the problem.
[0061] When the optimal point found by the above evolutionary computing algorithm in multiple search processes can cover the entire solution space, it can be considered that the algorithm has reached the global optimum.
[0062] Specifically, evolutionary computing algorithms are usually accompanied by a large amount of computing resource consumption, especially when dealing with high-dimensional and complex solution spaces, the amount of calculation will increase dramatically. In order to solve this problem, this application adopts a distributed computing strategy, aiming to significantly save experimental time and computing resources through parallel processing. In a distributed system, evolutionary computing can make full use of the capabilities of multiple computing nodes and evaluate each individual in the population in parallel. This parallel processing method is particularly suitable for scenarios where the cost of evaluating the function is high. Through distributed computing, the evaluation time of each generation of population can be greatly shortened, making the evolutionary algorithm perform better in terms of efficiency and scalability. This not only improves the overall computing efficiency, but also shows significant advantages in large-scale optimization problems.
[0063] Furthermore, the evolutionary computation algorithm uses a dynamic mutation rate to achieve rapid convergence; wherein the total number of iterations is N, the current number of iterations is N0, and the dynamic mutation rate G is expressed as:
[0064]
[0065] This formula ensures that the mutation rate is between 0.05 and 0.1. When the fitness is high, the mutation rate decreases and the optimal speed increases.
[0066] Specifically, the control of mutation rate in evolutionary computing algorithms is also a difficult point. When the fitness is high, the superposition of mutation results in the worse direction has a much greater impact on the result than the superposition of mutation results in the best direction. If the mutation rate decreases, it is possible to fall into the local optimum; if the mutation rate increases, it is difficult to increase the fitness, but it is possible to jump out of the local optimum. Therefore, the present application designs the above-mentioned dynamic mutation rate.
[0067] As attached Figure 2 The experimental prediction effect diagram shown. The red curve is the predicted data on the test set, the green curve is the actual data, the horizontal axis is time, and the vertical axis is the predicted data value (previously preprocessed vectorized data). It can be seen that the predicted data is closer to the actual data. It can be seen that this application proposes a method for optimizing the prediction of multilingual complex network behavior based on evolutionary computing algorithms. This method achieves accurate prediction and adaptive optimization of behavior in complex networks by uniformly preprocessing multilingual data, combining evolutionary computing algorithms for feature selection and optimization, and dynamically adjusting model parameters.
[0068] This application adopts multilingual data preprocessing and feature extraction methods to perform unified formatting and semantic feature extraction for data sources from different languages. In this stage, natural language processing (NLP) technology is used to process and normalize text data in different languages to ensure that multilingual data can be uniformly input into complex network analysis models. This application uses evolutionary computing algorithms (such as genetic algorithms, differential evolution algorithms, etc.) that are improved based on distributed computing strategies to select and optimize features. Evolutionary computing automatically finds the optimal feature combination by simulating natural selection and evolutionary processes, thereby removing redundant information and noise data and improving the computational efficiency and prediction accuracy of the model. The global search capability of the algorithm ensures that the behavior prediction model can effectively cope with large-scale, multi-dimensional multilingual data.
[0069] Embodiment 2:
[0070] As attached Figure 3 As shown, the present application also provides a system for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithms, using any of the above methods; the system includes:
[0071] The data preprocessing module unifies the symbols, semantics and formats of network behavior data based on different languages to make it suitable for the behavior prediction model of complex networks;
[0072] The evolutionary computing feature selection module uses evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data;
[0073] The dynamic network analysis module uses evolutionary computing algorithms to adaptively adjust the neural network architecture parameters of the behavior prediction model according to the dynamic changes of nodes and edges in the complex network to cope with the changes of the complex network;
[0074] The adaptive optimization module, targeting the nonlinearity and uncertainty of behaviors in complex networks, iteratively optimizes the configuration hyperparameters of the behavior prediction model through evolutionary computing algorithms to achieve automatic tuning of the model;
[0075] The complex network includes social networks, economic networks and communication networks; the nodes represent individuals in the complex network; the edges represent the interaction relationships between the individuals; the neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.
[0076] As attached Figure 1As shown, the technical solution of this application is implemented through a complete system architecture, which is divided into a data preprocessing module, an evolutionary computing feature selection module, a dynamic network analysis module, and an adaptive optimization module. Each module is closely integrated with the data flow and feedback mechanism to ensure the efficient operation and optimization of the system.
[0077] In summary, this application uniformly processes and extracts features from multilingual data, and optimizes feature selection using evolutionary computing algorithms to ensure that data features in different language contexts can be effectively captured, thereby improving prediction accuracy in cross-language environments. At the same time, through the global search and optimization capabilities of evolutionary computing algorithms, redundant information and noise data can be efficiently screened, data quality can be improved, computing costs can be reduced, and the overall prediction accuracy of the behavior prediction model can be improved.
[0078] In addition, this application enhances the adaptability to dynamic changes in complex networks. The model can automatically adjust according to real-time changes in the network structure to ensure that the prediction results are more accurate and timely.
[0079] Finally, this application utilizes the adaptive optimization mechanism of evolutionary computing algorithms to prevent the behavior prediction model from falling into a local optimal state, enhance the ability of the behavior prediction model to cope with nonlinear and uncertain changes in complex networks, and further improve the generalization ability and accuracy of the prediction.
[0080] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A method for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithms, characterized in that: The method comprises the following steps: s1. Unify the symbols, semantics and formats of network behavior data based on different languages to make it suitable for behavior prediction models of complex networks; s2. Use evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data; s3. Adaptively adjust the neural network architecture parameters of the behavior prediction model based on the dynamic changes of nodes and edges in the complex network through evolutionary computing algorithms to cope with the changes of the complex network; s4. Aiming at the nonlinearity and uncertainty of behavior in complex networks, the configuration hyperparameters of the behavior prediction model are iteratively optimized through evolutionary computing algorithms to achieve automatic tuning of the model; The complex network includes social network, economic network and communication network; The nodes represent individuals in a complex network; The edges represent the interaction relationships between the individuals; The neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.
2. The method according to claim 1, characterized in that The step s1 comprises: s11. Through natural language processing technology, remove stop words, punctuation marks, special symbols and other noise information in the data to clean the text in the network behavior data; s12. Use the pre-trained multilingual word embedding model to map network behavior data in different languages but with the same semantics into the same semantic space, ensure the consistency of multilingual network behavior data in the semantic space, and achieve multilingual semantic normalization; s13. Vectorize text data so that it can be used as input for complex network analysis models and achieve data formatting.
3. The method according to claim 1, characterized in that The step s2 comprises: s21. Randomly select network behavior data in different languages for feature extraction as an initial feature set, wherein the feature set includes several feature combinations; s22. For the features contained in each feature combination, use the behavior prediction model to predict the node behavior in the complex network, use cross-validation to evaluate the prediction accuracy of the feature combination, calculate the fitness value, and complete the fitness function evaluation; s23. Select each feature combination in the initial feature set according to the fitness value, and generate a new feature combination through crossover and mutation operations to complete the genetic operation; s24. Repeat steps s23 and s22 until the optimal feature combination that satisfies the prediction accuracy is found; or the preset number of iterations is reached.
4. The method according to claim 1, characterized in that The step s3 comprises: s31. Real-time monitoring of the dynamic changes of complex network nodes and edges to record changes in network topology; s32. When the structure of a complex network changes, re-evaluate the performance of the existing behavior prediction model and find new parameter configurations through evolutionary algorithms so that the behavior prediction model can dynamically adapt to changes in the network structure and complete parameter optimization; s33. Based on the optimized parameters, the behavior prediction model is updated in real time so that it can make behavior predictions based on the latest structural changes of the complex network, ensure the accuracy and timeliness of the predictions, and complete the dynamic update of the behavior prediction model.
5. The method according to claim 1, characterized in that The step s4 comprises: s41. Through the global search mechanism of evolutionary computing algorithm, explore various parameter configurations and feature combinations in the parameter space, ensure that the search process covers different areas of the parameter space, and complete the global search; s42. Deeply optimize local areas of the parameter space to ensure that the final parameter settings can adapt to the behavior patterns of complex networks in a multilingual environment; s43. After each behavior prediction model optimization, the feedback mechanism is used to evaluate the prediction effect of the behavior prediction model; if the effect is not ideal, the system will continue to optimize the parameters in step s42 until the prediction effect reaches the optimal state.
6. The method according to claim 1, characterized in that The evolutionary computing algorithm adopts a distributed computing strategy to perform parallel evolutionary solution calculation; specifically: y1. Express the variables and constraints of the problem in a form that the evolutionary computation algorithm can handle, and divide the scope of its global search solution space; y2. Use evolutionary computing algorithms to perform a global search of the entire solution space, and continuously optimize the solutions in the feature set through selection, crossover, and mutation operations; y3. Filter out multiple solutions with better performance through evolutionary computing algorithms; y4. Calculate the position differences between the high-quality solutions in the solution space, and select high-quality points that are far away from each other as the starting point of the gradient descent method; y5. Adopt the strategy of multiple starts, and use the selected high-quality points as the starting point of an independent gradient descent process to perform local search optimization; y6. After multiple rounds of iterations, the solution with the highest prediction accuracy and the best solution is selected from these points as the final solution to the problem.
7. The method according to claim 1, characterized in that The dynamic mutation rate used by the evolutionary computation algorithm is to achieve rapid convergence; where the total number of iterations is N, and the current number of iterations is N0, the dynamic mutation rate G is expressed as: This formula ensures that the mutation rate is between 0.05 and 0.
1.
8. A system for optimizing multilingual complex network behavior prediction based on evolutionary computing algorithms, characterized in that: Adopt the method as claimed in any one of claims 1 to 7; The system includes: The data preprocessing module unifies the symbols, semantics and formats of network behavior data based on different languages to make it suitable for the behavior prediction model of complex networks; The evolutionary computing feature selection module uses evolutionary computing algorithms to select and optimize the features of preprocessed network behavior data; The dynamic network analysis module uses evolutionary computing algorithms to adaptively adjust the neural network architecture parameters of the behavior prediction model according to the dynamic changes of nodes and edges in the complex network to cope with the changes of the complex network; The adaptive optimization module, targeting the nonlinearity and uncertainty of behaviors in complex networks, iteratively optimizes the configuration hyperparameters of the behavior prediction model through evolutionary computing algorithms to achieve automatic tuning of the model; The complex network includes social networks, economic networks and communication networks; the nodes represent individuals in the complex network; the edges represent the interaction relationships between the individuals; the neural network architecture parameters include: weight and bias; the hyperparameters include: number of iterations and learning rate.