Knowledge graph-based intelligent quantitative research platform system and method thereof

Abstract

The invention relates to the technical field of intelligent quantitative research platform systems, in particular to an intelligent quantitative research platform system based on a knowledge graph and a method thereof. The financial knowledge graph construction module is used for constructing a dynamic financial knowledge graph based on the multi-source data; the factor screening and optimizing module is used for extracting and optimizing quantitative factors from the financial knowledge graph and outputting an optimized factor set; the model analysis module is used for analyzing the contribution of the factor output by the factor screening and optimizing module to the income; the back-testing module is used for carrying out strategy back-testing based on the historical data and the factor set and verifying the yield and the risk index; and the result display module is used for displaying and outputting an analysis result of the back-testing module, and the dynamic weighting mechanism effectively improves the screening precision of factors, so that the model can keep higher income stability when facing different market conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent quantitative research platform systems, and specifically to an intelligent quantitative research platform system and method based on a knowledge graph. Background Art

[0002] With the rapid development of financial markets, a large amount of data continues to emerge, including news, fundamentals, technical indicators, public opinion and transaction data. The heterogeneity and complexity of these data have brought great challenges to investors in exploring potential market rules and optimizing investment strategies. In order to effectively utilize these data, intelligent quantitative research methods have emerged, among which data-driven quantitative factor mining and optimization have become the key. However, existing technologies still face many problems in dealing with this complex task.

[0003] At present, common quantitative factor research methods usually rely on a single data source or a fixed set of factors, and verify the effectiveness of the strategy through backtesting optimization. This method may perform well under specific market conditions, but when the market environment changes dynamically, the factor significance is often unstable, resulting in large fluctuations in returns. In addition, the redundancy and collinearity problems between factors are prevalent, which makes the model complexity and computational costs rise sharply, while weakening the interpretability and robustness of the model. More importantly, traditional methods usually lack dynamic updating and adaptability, making it difficult to capture changes in the market environment in real time, resulting in poor performance of strategies in complex and changing financial markets.

[0004] Especially in the process of factor screening and optimization, most existing technologies adopt a one-by-one screening method, which is inefficient and difficult to deal with large-scale factor sets. In addition, the existing models lack an effective balance mechanism between optimizing returns and controlling risks, resulting in the inability to balance the goals of maximizing returns and minimizing risks in the factor selection process. This contradiction between returns and risks has not been effectively resolved for a long time.

[0005] In terms of algorithm implementation, existing technologies often lack a mechanism for multi-algorithm collaboration. A single algorithm may have good effects in some aspects, but its limitations and contradictions are also very obvious. For example, although deep learning algorithms have significant advantages in factor feature extraction and pattern recognition, the interpretability of their results is poor; and although statistical methods are simple and intuitive, they are unable to handle complex nonlinear data relationships. These technical problems further limit the practical application effect of existing quantitative research systems.

[0006] In summary, the prior art mainly has the following technical problems:

[0007] 1. Factor significance and stability issues: Factors perform inconsistently in different market environments, resulting in large volatility in returns and poor stability.

[0008] 2. Factor redundancy and collinearity problems: The factors are highly correlated, which leads to complex models, high computational costs, and weak interpretability.

[0009] 3. Insufficient dynamic updating of factors and market adaptability: Existing models are difficult to dynamically adapt to market changes, and strategy performance is difficult to maintain consistency.

[0010] 4. Low factor screening efficiency: Existing methods lack efficient parallel processing mechanisms in the factor screening process and are difficult to cope with large-scale data.

[0011] 5. Conflict between profit and risk optimization: There is a lack of effective mechanism to balance the goals of profit maximization and risk control. Summary of the invention

[0012] The present invention proposes an intelligent quantitative research platform system and method based on knowledge graph, aiming to solve the above technical problems. The present invention comprehensively optimizes the factors such as factor significance, redundancy, dynamic update, screening efficiency and return-risk balance through multi-source data acquisition, financial knowledge graph construction and multi-level dynamic factor screening optimization algorithm.

[0013] The present invention proposes an intelligent quantitative research platform system based on knowledge graph, including:

[0014] Multi-source data acquisition module, used to acquire multi-source heterogeneous data in the financial market, including news data, fundamental data, technical indicator data, public opinion data and transaction data;

[0015] A financial knowledge graph construction module, used to construct a dynamic financial knowledge graph based on the multi-source data;

[0016] A factor screening and optimization module, used to extract and optimize quantitative factors from the financial knowledge graph and output an optimized factor set;

[0017] A model analysis module, used to analyze the contribution of the factors output by the factor screening and optimization module to the returns;

[0018] A backtesting module, used to perform strategy backtesting based on historical data and the factor set to verify yield and risk indicators;

[0019] The result display module is used to display and output the analysis results of the backtest module.

[0020] Preferably, the factor screening and optimization module comprises:

[0021] Dynamic weighted significance screening unit, used to assign dynamic weights to each factor in the initial factor set based on factor significance and market volatility, and screen out a factor set with higher significance;

[0022] A multi-objective redundancy removal optimization unit, used to perform redundancy optimization on the factor set output by the dynamic weighted significance screening unit to remove highly redundant factors;

[0023] The time-series adaptive factor dynamic update unit is used to dynamically adjust the factor weights and values ​​to adapt to changes in the market environment;

[0024] Parallel fast factor screening unit, used to parallelly calculate the optimal factor combination in the screening factor set;

[0025] The return contribution optimization unit is used to optimize the final factor set based on the return and risk trade-off.

[0026] Preferably, the dynamic weighted significance screening unit calculates the factor weights based on the following formula:

[0027]

[0028] Among them, W i,t is the dynamic weight of factor i at time t; σ i,t is the market volatility of factor i at time t; S i,t is the significance score of factor i at time t; n is the total number of factors in the factor set.

[0029] Preferably, the multi-objective redundancy reduction optimization unit calculates the factor redundancy by the following formula:

[0030] R(i,j)=I(i,j)+λ·(1-ρ(i,j))

[0031] Among them, R(i,j) is the redundancy score between factor i and factor j; I(i,j) is the mutual information between factor i and factor j; ρ(i,j) is the correlation coefficient between factor i and factor j; λ is the redundancy control parameter.

[0032] Preferably, the timing adaptive factor dynamic updating unit uses the following formula to dynamically update the factor value:

[0033] F i,t+1 =Bi-LSTM(F i,t ,h t )

[0034] Among them, F i,t+1 is the updated value of factor i at time t+1; F i,t is the current value of factor i at time t, h t is the hidden state at time t; Bi-LSTM is a bidirectional long short-term memory network.

[0035] Preferably, the parallel fast factor screening unit screens factors based on the following formula:

[0036]

[0037] Among them, F opt is the optimized factor; Fitness(F) is the fitness score of factor F; Time(F) is the calculation time of factor F.

[0038] Preferably, the revenue contribution optimization unit optimizes the final factor set based on the following formula:

[0039]

[0040] Among them, R i is the profit contribution of factor i, W i is the final weight of factor i, Risk(P) is the risk of portfolio P; β is the risk-return trade-off coefficient.

[0041] Preferably, the backtesting module further performs backtesting analysis based on the output factor set of the yield contribution optimization unit, and obtains indicator results of historical yield, maximum drawdown rate and volatility.

[0042] Preferably, the result display module comprises:

[0043] A data storage unit, used to store the analysis data of the backtest module;

[0044] Data analysis unit, used to further analyze the backtest results;

[0045] A data display unit, used to display the analysis results in the form of charts;

[0046] The report generation unit is used to output the optimization report.

[0047] The method of the system comprises:

[0048] Obtain data from multiple sources;

[0049] Constructing a financial knowledge graph based on the financial knowledge graph construction module;

[0050] Using the factor screening and optimization module to gradually screen and optimize factors;

[0051] Analyzing factor contributions using the model analysis module;

[0052] Performing backtest verification based on the backtest module;

[0053] Finally, the backtesting and optimization results are output through the result display module.

[0054] The beneficial technical effects brought by the technical solution of the present invention are:

[0055] Based on traditional quantitative research methods, this paper proposes an innovative technical framework of multi-algorithm collaboration, fully utilizes knowledge graphs and deep learning technologies, and realizes full-process improvements in factor screening, optimization, and strategy verification.

[0056] First, the present invention uses a dynamic weighted significance screening algorithm to dynamically adjust the factor weights in combination with market volatility and factor significance to ensure the stability of the factors under different market environments. This dynamic weighting mechanism effectively improves the screening accuracy of the factors, enabling the model to maintain a high level of return stability when facing different market conditions.

[0057] Secondly, in order to solve the problem of factor redundancy, the present invention introduces a multi-objective redundancy elimination optimization algorithm, which removes highly redundant factors by jointly calculating mutual information and correlation coefficients. This method not only reduces the complexity of the model, but also enhances the interpretability and robustness of the model.

[0058] In order to further adapt to the dynamic changes of the market, the present invention adopts a time-series adaptive factor dynamic update algorithm. Through the bidirectional long short-term memory network (Bi-LSTM), the dynamic update of the factor value is realized, so that the factor can reflect the changes in the market environment in real time. This dynamic update mechanism significantly improves the adaptability of the factor and helps the strategy to maintain competitiveness in a complex and changing market.

[0059] In terms of factor screening efficiency, the present invention proposes a parallel fast factor screening algorithm, which uses parallel computing technology to improve the screening speed and is particularly suitable for processing large-scale factor sets. This efficient screening method significantly shortens the model training and optimization time and improves the overall efficiency of the system.

[0060] Finally, in terms of the balance optimization between returns and risks, this paper proposes a dynamic trade-off mechanism by comprehensively considering the return contribution and risk control of factors through a return contribution optimization algorithm. This algorithm can achieve the best balance between maximizing returns and minimizing risks, thereby improving the overall performance of the strategy.

[0061] The present invention fully embodies the dynamics, synergy and robustness in algorithm design, and realizes the complementary, superposition and synergistic effects of factor screening and optimization process through multi-level optimization algorithm. The above beneficial effects show that the present invention significantly improves the performance and practical application value of the quantitative research system, and provides an efficient, intelligent and dynamic quantitative research platform for financial investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0063] Figure 2 This is a logic block diagram of the dynamic weighted significance screening of the present invention.

[0064] Figure 3 This is a logic block diagram of the multi-objective redundancy optimization of the present invention.

[0065] Figure 4 The present invention is a logic block diagram for dynamically updating the timing adaptive factor.

[0066] Figure 5 It is a logic block diagram of the parallel fast factor screening of the present invention.

[0067] Figure 6 Optimize logic block diagram for benefit contribution of the present invention. DETAILED DESCRIPTION

[0068] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0069] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0070] See also Figure 1-6 The present invention proposes an intelligent quantitative research platform system based on knowledge graph, including a multi-source data acquisition module 1, a financial knowledge graph construction module 2, a factor screening and optimization module 3, a model analysis module 4, a backtesting module 5 and a result display module 6. The modules work together to realize the full-process quantitative research from data acquisition, knowledge construction, factor screening, backtesting verification to result display.

[0071] Among them, the multi-source data acquisition module 1 is used to obtain multi-source heterogeneous data including news data, fundamental data, technical indicator data, public opinion data and transaction data from multiple data sources. The selection of these data sources is intended to cover various important information in the financial market. Preferably, news data can be sourced from authoritative websites in the financial industry, fundamental data can be obtained from professional databases such as Bloomberg, and technical indicator data and transaction data can be extracted from public data of exchanges.

[0072] In this embodiment, the financial knowledge graph construction module 2 integrates multi-source data and uses natural language processing technology and entity alignment algorithm to construct a financial knowledge graph. The knowledge graph can not only be updated dynamically, but also can mine potential market relationships from multi-dimensional related information, providing a basis for subsequent factor screening.

[0073] The factor screening and optimization module 3 screens and optimizes the quantitative factors in the knowledge graph, and improves the stability and effectiveness of the factors by integrating multiple algorithms. The model analysis module 4 interprets the optimized factors and analyzes their contribution to investment returns. The backtesting module 5 backtests the optimized factor set with historical data to evaluate its performance under real market conditions. Finally, the result display module 6 is used to display the research results in the form of graphics, reports, etc.

[0074] Preferably, the factor screening and optimization module 3 includes a dynamic weighted significance screening unit 31, a multi-objective redundancy elimination optimization unit 32, a time-series adaptive factor dynamic update unit 33, a parallel fast factor screening unit 34, and a benefit contribution optimization unit 35. These units work together to construct a hierarchical factor screening and optimization process.

[0075] The function of the dynamic weighted significance screening unit 31 is to calculate the dynamic weights according to the significance scores and market volatility of the factors in different periods, and screen out factors with high significance and stability. The multi-objective redundancy removal optimization unit 32 further optimizes these factors, removes highly correlated factors by calculating redundancy, and ensures that the final factor set has strong independence and interpretability.

[0076] For example, in a certain market cycle, a factor with high volatility but low significance score may be assigned a lower weight and thus be screened out. Preferably, when the significance score \(S_{i,t}\) is less than a certain threshold such as 0.05, the factor is removed.

[0077] In one embodiment of the present invention, the present invention uses the following formula in the dynamic weighted significance screening unit 31 to calculate the factor weight:

[0078]

[0079] Among them, W i,t is the dynamic weight of factor i at time t, which is used to quantify the importance of the factor at that time point; σ i,t is the market volatility of factor i at time t, reflecting the risk level of the factor; S i,t is the significance score of factor i at time t, which is used to measure the statistical importance of the factor; n is the total number of factors in the current factor set, which is used for normalization.

[0080] Preferably, in one embodiment of the present invention, when the weight of a factor is less than 0.01, the factor will be marked as a non-significant factor and removed from the screening process. This setting ensures that the impact of the factor on the final result is effectively controlled in the case of low significance or high volatility.

[0081] In one embodiment of the present invention, the multi-objective redundancy reduction optimization unit 32 calculates the redundancy between factors using the following formula:

[0082] R(i,j)=I(i,j)+λ·(1-ρ(i,j))

[0083] Among them, R(i,j) is the redundancy score between factor i and factor j, and the larger the value, the more serious the redundancy. I(i,j) is the mutual information between factor i and factor j, which is used to measure the degree of information crossover between the two factors; ρ(i,j) is the correlation coefficient between factor i and factor j, which is used to measure the linear correlation between the two; λ is a redundancy control parameter. In a preferred embodiment of the present invention, the value of λ is 0.5 to balance the weights of mutual information and correlation coefficient.

[0084] For example, when the value of R(i,j) is greater than a certain threshold such as 0.8, factor j will be judged as a redundant factor and removed. This algorithm can effectively reduce the redundancy between factors and improve the generalization ability of the model.

[0085] In one embodiment of the present invention, the function of the timing adaptive factor dynamic updating unit 33 is described in detail. The unit uses a bidirectional long short-term memory network (Bi-LSTM) to dynamically update the factor value, and its calculation formula is as follows:

[0086] F i,t+1 =Bi-LSTM(F i,t ,h t )

[0087] Among them, F i,t+1 is the predicted value of factor i at time t+1; F i,t is the current value of factor i at time t, reflecting the performance of the factor in historical data; h t is the hidden state at time t, which is used to capture the temporal characteristics of the factor; Bi-LSTM is a bidirectional long short-term memory network, which is used to comprehensively consider the changes of the factor in the forward and reverse time series.

[0088] Preferably, in one embodiment, for some factors with high short-term volatility, the update frequency can be set to once a day, while for long-term stable factors, the update frequency can be reduced to once a week. This flexible update mechanism ensures that the factors can dynamically adapt to the rapid changes in the market, while avoiding unnecessary fluctuations caused by frequent adjustments of factors.

[0089] In one embodiment of the present invention, the parallel fast factor screening unit 34 uses parallel computing to screen out the optimal factor combination from the dynamically updated factor set. This unit implements the factor screening optimization through the following formula:

[0090]

[0091] Among them, F opt is the optimal factor set after optimization and screening; It is the factor set output by the timing adaptive factor dynamic update unit 33; Fitness(F) is the fitness score of factor F, which reflects the contribution of the factor to the return; Time(F) is the calculation time of factor F, which measures the complexity and efficiency of factor calculation. In a preferred embodiment, the calculation of the fitness score Fitness(F) comprehensively considers the factor's return contribution, risk reduction effect, and its stability in different market environments. Especially for high-frequency trading strategies, the optimization of the time parameter Time(F) is crucial, and it usually needs to be controlled within the millisecond range to ensure the real-time performance of the model.

[0092] The parallel fast factor screening unit 34 realizes fast screening of large-scale factor sets through a parallel computing framework (such as MapReduce or Spark), which significantly improves computing efficiency. In a specific market environment (such as intraday high-frequency trading), by optimizing the fitness and computing time of each factor, an optimal factor combination that takes into account both benefits and computing efficiency can be obtained in a short time.

[0093] In one embodiment of the present invention, the profit contribution optimization unit 35 optimizes the final factor set by the following formula to achieve a dynamic balance between profit and risk:

[0094]

[0095] Among them, R i is the return contribution of factor i, reflecting the impact of the factor on the portfolio return; W i is the final weight of factor i, which is dynamically adjusted by the above modules; Risk(P) is the risk measurement function of the investment portfolio P, and preferably VaR (value at risk) or CVaR (conditional value at risk) can be used as the risk indicator; β is the risk-return trade-off coefficient, which is used to balance return and risk, and its preferred value is between 0.1 and 0.5. The specific value is dynamically adjusted according to the risk preference of investors; n is the number of factors in the final factor set.

[0096] For example, a factor shows an average annualized return of 10% in historical data backtesting, but its corresponding VaR indicator is 5%. When the risk-return trade-off coefficient β is set to 0.3, the factor's contribution to the final return of the portfolio is:

[0097] R final =10·W i -0.3·5=8.5

[0098] Through the above optimization formula, the negative impact of high-risk factors on portfolio returns can be effectively reduced while ensuring the stability of overall returns.

[0099] The return contribution optimization unit 35 continuously optimizes the return and risk trade-off parameters by further combining genetic algorithms and reinforcement learning techniques, realizes dynamic adaptive adjustment of factor sets, and improves the adaptability of investment strategies in different market environments.

[0100] In one embodiment of the present invention, the backtesting module 5 verifies the effectiveness of the factors and the actual performance of the investment strategy by backtesting the historical data of the final optimized factor set. The module uses the simulation data and historical market data generated by deep learning to calculate the key indicators of the investment strategy, including annualized rate of return, maximum drawdown rate and Sharpe ratio.

[0101] In a preferred embodiment of the present invention, the backtest module 5 adopts the following backtest process:

[0102] 1. Data preparation: Select the optimized factor set and the corresponding historical market data;

[0103] 2. Strategy generation: Generate specific investment strategies based on the optimized factor set;

[0104] 3. Profit calculation: Calculate the daily profit of the strategy under historical market data;

[0105] 4. Risk assessment: Calculate the maximum drawdown and volatility of the strategy;

[0106] 5. Result output: Generate a backtest report based on the above calculation results.

[0107] Taking actual data as an example, the backtesting results of a certain investment strategy from 2015 to 2020 showed that the annualized return was 15%, the maximum drawdown rate was 10%, and the Sharpe ratio was 1.5, indicating that the strategy can achieve higher returns while controlling risks.

[0108] In one embodiment of the present invention, the result display module 6 includes a data storage unit 61, a data analysis unit 62, a data display unit 63 and a report generation unit 64. These units work together to achieve intuitive display and analysis of backtest results.

[0109] Among them, the data storage unit 61 is used to store a large amount of historical return and risk data generated during the backtesting process; the data analysis unit 62 extracts key indicators through multi-dimensional analysis of these data; the data display unit 63 displays the analysis results in the form of charts, such as return curves, drawdown curves and risk distribution diagrams; the report generation unit 64 organizes the analysis results into a structured report for user reference.

[0110] In a preferred embodiment, the data display unit 63 supports dynamic interaction, and the user can choose to view the changes in returns and risks according to different time windows. For example, by sliding the time axis, the user can quickly view the strategy performance within a certain period of time, thereby helping the user to better understand the performance of the investment strategy in different market environments.

[0111] The operation method of the present invention comprises the following steps:

[0112] First, heterogeneous data in the financial market are obtained from the multi-source data acquisition module 1, including news, fundamentals, technical indicators, and transaction data. These data are standardized and structured through the financial knowledge graph construction module 2 to form a dynamic financial knowledge graph.

[0113] Then, the factor screening and optimization module 3 extracts and optimizes quantitative factors from the knowledge graph, and gradually optimizes the factor set through five hierarchical steps: dynamic weighted significance screening, multi-objective de-redundancy optimization, time-series adaptive factor update, parallel fast factor screening, and benefit contribution optimization.

[0114] Next, the model analysis module 4 conducts an in-depth analysis of the optimized factor set to evaluate its contribution to returns. Subsequently, the backtesting module 5 backtests and verifies the investment strategy based on historical data, and obtains a reliable investment backtesting report through multiple experiments.

[0115] Finally, the result display module 6 presents the backtesting results to users in the form of intuitive charts and reports to help users better understand and optimize investment strategies.

[0116] Through the mutual cooperation of the above modules, the present invention realizes a complete intelligent quantitative research platform based on knowledge graph. The whole system takes knowledge graph as the core, and solves the problems of low factor screening efficiency, factor redundancy and unstable returns in the prior art through multi-level dynamic factor screening optimization algorithm, which greatly improves the intelligence and reliability of investment strategy. This specification describes in detail the complete workflow from data acquisition to result display, ensuring the comprehensiveness and clarity of the technical solution.

[0117] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. The intelligent quantitative research platform system based on knowledge graph is characterized by: include: Multi-source data acquisition module, used to acquire multi-source heterogeneous data in the financial market, including news data, fundamental data, technical indicator data, public opinion data and transaction data; A financial knowledge graph construction module, used to construct a dynamic financial knowledge graph based on the multi-source data; A factor screening and optimization module, used to extract and optimize quantitative factors from the financial knowledge graph and output an optimized factor set; A model analysis module, used to analyze the contribution of the factors output by the factor screening and optimization module to the returns; A backtesting module, used to perform strategy backtesting based on historical data and the factor set to verify yield and risk indicators; The result display module is used to display and output the analysis results of the backtest module.

2. The system according to claim 1, characterized in that The factor screening and optimization module includes: Dynamic weighted significance screening unit, used to assign dynamic weights to each factor in the initial factor set based on factor significance and market volatility, and screen out a factor set with higher significance; A multi-objective redundancy removal optimization unit, used to perform redundancy optimization on the factor set output by the dynamic weighted significance screening unit to remove highly redundant factors; The time-series adaptive factor dynamic update unit is used to dynamically adjust the factor weights and values ​​to adapt to changes in the market environment; Parallel fast factor screening unit, used to parallelly calculate the optimal factor combination in the screening factor set; The return contribution optimization unit is used to optimize the final factor set based on the return and risk trade-off.

3. The system according to claim 2, characterized in that The dynamic weighted significance screening unit calculates the factor weights based on the following formula: Among them, W i,t is the dynamic weight of factor i at time t; σ i,t is the market volatility of factor i at time t; S i,t is the significance score of factor i at time t; n is the total number of factors in the factor set.

4. The system according to claim 3, characterized in that The multi-objective redundancy reduction optimization unit calculates the factor redundancy by the following formula: R(i,j)=I(i,j)+λ·(1-ρ(i,j)) Among them, R(i,j) is the redundancy score between factor i and factor j; I(i,j) is the mutual information between factor i and factor j; ρ(i,j) is the correlation coefficient between factor i and factor j; λ is the redundancy control parameter.

5. The system according to claim 4, characterized in that The timing adaptive factor dynamic update unit uses the following formula to dynamically update the factor value: F i,t+1 =Bi-LSTM(F i,t ,h t ) Among them, F i,t+1 is the updated value of factor i at time t+1; F i,t is the current value of factor i at time t, h t is the hidden state at time t; Bi-LSTM is a bidirectional long short-term memory network.

6. The system according to claim 5, characterized in that The parallel fast factor screening unit screens factors based on the following formula: Among them, F opt is the optimized factor; Fitness(F) is the fitness score of factor F; Time(F) is the calculation time of factor F.

7. The system according to claim 6, characterized in that The revenue contribution optimization unit optimizes the final factor set based on the following formula: Among them, R i is the profit contribution of factor i, W i is the final weight of factor i, Risk(P) is the risk of portfolio P; β is the risk-return trade-off coefficient.

8. The system according to claim 7, characterized in that The backtest module further performs backtest analysis based on the output factor set of the yield contribution optimization unit, and obtains indicator results of historical yield, maximum drawdown rate and volatility.

9. The system according to claim 8, characterized in that The result display module includes: A data storage unit, used to store the analysis data of the backtest module; Data analysis unit, used to further analyze the backtest results; A data display unit, used to display the analysis results in the form of charts; The report generation unit is used to output the optimization report.

10. The method of the system according to claim 9, characterized in that include: Obtain data from multiple sources; Construct a financial knowledge graph based on the financial knowledge graph construction module; Using the factor screening and optimization module to gradually screen and optimize factors; Analyzing factor contributions using the model analysis module; Performing backtest verification based on the backtest module; Finally, the backtesting and optimization results are output through the result display module.

Images