Investment strategy configuration method and device based on form semantic analysis and dynamic reasoning

Abstract

The application discloses an investment strategy configuration method and device based on form semantic analysis and dynamic reasoning, relates to the technical field of strategy configuration, and discloses an investment strategy configuration method based on form semantic analysis and dynamic reasoning, which comprises the following steps: obtaining a structured form field and an investment task description, and performing semantic analysis on the investment task description and the structured form field to obtain a structured task description; generating multi-source investment information according to the structured task description; generating a target portfolio scheme under a preset market scenario according to the structured task description, the multi-source investment information and historical market data; determining a disturbance feature vector according to the target portfolio scheme, and generating a re-optimization task description according to the disturbance feature vector; and optimizing a candidate asset weight configuration scheme according to the re-optimization task description, a preset difference degree penalty coefficient and a weight vector of a current holding, and determining an investment strategy configuration scheme. Therefore, adaptive construction of an investment portfolio under market disturbance is realized.

Description

Technical Field

[0001] This application relates to the field of strategy configuration technology, and in particular to an investment strategy configuration method and apparatus based on form semantic parsing and dynamic reasoning. Background Technology

[0002] With the specialization and scaling up of the asset management industry, portfolio management is gradually shifting from subjective decision-making that relies on human experience to a technology-driven model centered on data analysis and model optimization.

[0003] Given the aforementioned technical background, during portfolio operation, when market conditions or held assets change, the optimal portfolio is often recalculated through periodic or temporary overall rebalancing, with manual analysis used for intervention and adjustment when necessary. However, in practice, this approach heavily relies on professionals' understanding of model parameters and repeated debugging, and lacks prior planning and systematic response mechanisms for unforeseen events, easily leading to delayed decision-making. Summary of the Invention

[0004] The main purpose of this application is to provide an investment strategy configuration method and apparatus based on form semantic parsing and dynamic reasoning, which aims to solve the technical problem that it is difficult to efficiently and accurately understand and implement investment decision intentions under complex and ever-changing market disturbances during portfolio management, resulting in a lag in portfolio adjustment response.

[0005] To achieve the above objectives, this application proposes an investment strategy allocation method based on form semantic parsing and dynamic reasoning, the method comprising:

[0006] Obtain the structured form fields and investment task description, and perform semantic parsing on the investment task description and structured form fields to obtain the structured task description;

[0007] Generate multi-source investment information based on the structured task description;

[0008] Based on the structured task description, multi-source investment information and historical market data, a target portfolio scheme under a preset market scenario is generated. The target portfolio scheme includes a basic portfolio scheme and a backup portfolio scheme.

[0009] The perturbation feature vector is determined based on the target combination scheme, and the re-optimization task description is generated based on the perturbation feature vector;

[0010] Based on the task description of re-optimization, the preset difference penalty coefficient, and the weight vector of the current holdings, the candidate asset weight allocation scheme is optimized to determine the investment strategy allocation scheme.

[0011] Furthermore, to achieve the above objectives, this application also proposes an investment strategy configuration device based on form semantic parsing and dynamic reasoning. The device includes: a semantic parsing module for acquiring structured form fields and investment task descriptions, and performing semantic parsing on the investment task descriptions and structured form fields to obtain a structured task description; an opinion generation module for generating multi-source investment information based on the structured task description; a scheme combination module for generating a target portfolio scheme under a preset market scenario based on the structured task description, multi-source investment information, and historical market data, wherein the target portfolio scheme includes a basic portfolio scheme and a backup portfolio scheme; a task re-optimization module for determining a perturbation feature vector based on the target portfolio scheme and generating a re-optimized task description based on the perturbation feature vector; and a strategy configuration module for optimizing the candidate asset weight configuration scheme based on the re-optimized task description, a preset difference penalty coefficient, and the weight vector of the current holdings to determine the investment strategy configuration scheme.

[0012] In addition, to achieve the above objectives, this application also proposes an investment strategy configuration device based on form semantic parsing and dynamic reasoning. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the investment strategy configuration method based on form semantic parsing and dynamic reasoning as described above.

[0013] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the investment strategy configuration method based on form semantic parsing and dynamic reasoning as described above.

[0014] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the investment strategy configuration method based on form semantic parsing and dynamic reasoning as described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects:

[0016] By combining structured form fields with investment task descriptions and semantic parsing, input information is transformed into accurate and reusable structured task descriptions. Multi-source investment information is generated based on these descriptions, enriching the information dimensions for investment decisions. A target portfolio plan is generated based on the structured task descriptions, multi-source investment information, and historical market data. This target portfolio plan includes a basic portfolio plan and a backup portfolio plan, providing a foundation for multi-scenario portfolio configuration. Perturbation feature vectors are determined based on the target portfolio plan, and a re-optimization task description is generated, enabling targeted responses to perturbations during portfolio operation. The re-optimization process is constrained by a preset difference penalty coefficient, controlling the portfolio adjustment range. Compared to existing technologies, this approach solves the problems of inaccurate input, fragmented information, blind perturbation responses, and high rebalancing costs in traditional investment strategy configuration. It achieves precision in investment strategy configuration, multi-scenario adaptability, targeted perturbation responses, and low execution costs, effectively improving the controllability and operability of investment strategy configuration. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application.

[0020] Figure 2 This is a schematic diagram of the first multi-scenario forward-looking portfolio pre-configuration process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application.

[0021] Figure 3 This is a schematic diagram of the second multi-scenario forward-looking portfolio pre-configuration process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application.

[0022] Figure 4 This is a schematic diagram of the disturbance monitoring and event classification process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application.

[0023] Figure 5 This is a schematic diagram of the runtime disturbance and re-optimization process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application;

[0024] Figure 6 This is a schematic diagram of the re-optimization strategy routing and local repair process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application.

[0025] Figure 7 This is a schematic diagram of the result output and closed-loop interactive interface operation process provided in Embodiment 1 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application.

[0026] Figure 8 This is a flowchart illustrating Embodiment 2 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application;

[0027] Figure 9 This is a schematic diagram of the difference-controlled re-optimization process provided in Embodiment 2 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application;

[0028] Figure 10 A simplified flowchart illustrating the investment strategy configuration method based on form semantic parsing and dynamic reasoning provided in Embodiment 2 of this application;

[0029] Figure 11 This is a schematic diagram of the module structure of the investment strategy configuration device based on form semantic parsing and dynamic reasoning according to an embodiment of this application;

[0030] Figure 12 This is a schematic diagram of the hardware operating environment involved in the investment strategy configuration method based on form semantic parsing and dynamic reasoning in the embodiments of this application.

[0031] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0032] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0033] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0034] The main solution of this application embodiment is as follows: Obtain structured form fields and investment task descriptions, and perform semantic parsing on the investment task descriptions and structured form fields to obtain structured task descriptions; generate multi-source investment information based on the structured task descriptions; generate a target portfolio scheme under a preset market scenario based on the structured task descriptions, multi-source investment information, and historical market data, wherein the target portfolio scheme includes a basic portfolio scheme and a backup portfolio scheme; determine a perturbation feature vector based on the target portfolio scheme, and generate a re-optimization task description based on the perturbation feature vectors; optimize the candidate asset weight allocation scheme based on the re-optimization task description, a preset difference penalty coefficient, and the weight vector of the current holdings to determine the investment strategy allocation scheme.

[0035] In this embodiment, for ease of description, the following description will focus on an investment strategy configuration device based on form semantic parsing and dynamic reasoning.

[0036] Given that existing portfolio management technologies struggle to efficiently and accurately understand and implement investment decisions under complex and volatile market conditions, leading to delayed portfolio adjustments, this application provides a solution. This solution combines structured form fields with investment task descriptions and semantic parsing to transform input information into precise and reusable structured task descriptions. Based on these descriptions, multi-source investment information is generated, enriching the information dimensions for investment decisions. A target portfolio scheme, including basic and backup schemes, is generated by integrating the structured task descriptions, multi-source investment information, and historical market data, providing a foundation for multi-scenario portfolio configuration. Based on the target portfolio scheme, a disturbance feature vector is determined, and a re-optimization task description is generated, enabling targeted responses to disturbances during portfolio operation. Finally, a pre-set difference penalty coefficient constrains the re-optimization process, controlling the magnitude of portfolio adjustments. Compared to existing technologies, this approach solves the problems of inaccurate input, fragmented information, limited schemes, blind disturbance responses, and high rebalancing costs in traditional investment strategy configuration. It achieves precision in investment strategy configuration, multi-scenario adaptability, targeted disturbance responses, and low execution costs, effectively improving the controllability and operability of investment strategy configuration.

[0037] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone; or an electronic device capable of performing the above functions, or an investment strategy configuration device based on form semantic parsing and dynamic reasoning. The following description uses an investment strategy configuration device based on form semantic parsing and dynamic reasoning as an example to illustrate this embodiment and the subsequent embodiments.

[0038] Based on this, embodiments of this application provide an investment strategy configuration method based on form semantic parsing and dynamic reasoning, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application.

[0039] In this embodiment, the investment strategy configuration method based on form semantic parsing and dynamic reasoning includes steps S10 to S50:

[0040] Step S10: Obtain the structured form fields and investment task description, and perform semantic parsing on the investment task description and structured form fields to obtain the structured task description;

[0041] It should be noted that the structured form fields are a set of input fields designed in a modular and hierarchical manner, covering investment objective constraints, risk preferences of conditional asset pools, investment horizons, and disturbance handling strategies. They support dynamic dependencies at the field level, meaning that selecting a field automatically displays or locks related fields. These fields are grouped according to basic configurations, core objectives, hard constraints, and advanced strategy configurations, with dependency rules stored in JSON format for easy retrieval.

[0042] In addition, the investment task description is a collection of information entered by the user through structured form fields and natural language supplementary descriptions. It includes core content such as investment objectives and constraints, and the natural language supplementary descriptions can further refine the user's intent.

[0043] Furthermore, semantic parsing is a process of analyzing input information based on a natural language processing model. It can identify financial entities, relationships, quantities, time sequences, and logical constraints, and map them into structured semantic slots. During the parsing process, regular expression templates and entity dictionaries are combined to improve the recognition accuracy of structured expressions, and multiple candidate parsing schemes are generated for ambiguous sentences, along with confidence scores.

[0044] Furthermore, the structured task description is a standardized task specification formed after parsing, completion, and validation. It follows a predefined JSON schema and includes fields such as objectives, constraints, asset pools, and perturbation strategies. This description serves as the foundation for unified processing in subsequent stages to ensure data structure consistency.

[0045] Understandably, the process begins by acquiring the structured form fields and investment task description input by the user. The input is then subjected to integrity checks, logical conflict detection, type checks, and format standardization. Subsequently, a Transformer-based Chinese financial corpus fine-tuning model is used for joint encoding. A financial knowledge base is invoked to complete incomplete descriptions and perform consistency checks. The results are then filled into semantic slots to form a structured task description.

[0046] In one feasible implementation, step S10 may include steps S11 to S15:

[0047] Step S11: Obtain structured form fields, task description, and disturbance handling preferences, wherein the structured form fields include investment objectives, constraints, asset pool, risk preference, and disturbance handling strategy fields;

[0048] It should be noted that disturbance handling preferences are the user's pre-defined attitudes and strategic tendencies in response to different market disturbance events, clarifying the user's willingness to handle scenarios such as single asset suspension and extreme market volatility. These preferences will serve as an important reference for subsequent strategy routing decisions.

[0049] Furthermore, investment objectives are the return-related goals that users expect to achieve within the investment period, including specific details such as the target rate of return range and the pace of return realization. These objectives are the core guiding principle for portfolio development. Different investment objectives will directly affect the direction and proportion of asset allocation.

[0050] Additionally, constraints are limiting conditions set by users during the investment process, including industry concentration, single asset weight limits, and liquidity requirements, used to regulate the boundaries of portfolio allocation. These conditions will be followed as hard constraints during portfolio optimization.

[0051] Additionally, risk preference refers to a user's level of acceptance of investment risk, categorized into different levels such as conservative, moderate, high, and aggressive, used to match portfolio options that align with the user's risk tolerance. Risk preference influences the portfolio's risk exposure level and asset selection tendencies. Furthermore, the investment task description is a collection of information input by the user through structured form fields and supplementary natural language.

[0052] Additionally, the disturbance handling strategy field is a dedicated field in the structured form for recording user disturbance handling preferences. It belongs to the advanced strategy configuration level and forms a hierarchical structure with other basic fields. This field provides a standardized input entry point for users to explicitly express their disturbance response strategies.

[0053] Understandably, the modular, hierarchical form interface is used to obtain structured form fields input by users, while also receiving task descriptions and perturbation processing preferences input by users through forms and natural language, so that the collected information can comprehensively cover the core elements of investment.

[0054] Step S12: Generate an investment task description based on the task description and disturbance handling preferences;

[0055] Understandably, integrating the user-inputted task description and perturbation processing preferences, eliminating duplicate information and supplementing logical connections, forms a structurally complete and content-consistent investment task description, so that the user's investment intentions can be fully and accurately presented.

[0056] Step S13: Perform integrity checks, type conversions, and logical conflict detection on the structured form fields and investment task descriptions to obtain standard financial information;

[0057] It should be noted that completeness verification is a process of checking whether the required fields in the structured form and the investment task description are complete, ensuring that no core information is missing. For example, it checks whether key fields such as funding amount and investment period are filled in, to avoid affecting the generation of subsequent portfolio solutions due to incomplete information.

[0058] Additionally, type conversion is the process of uniformly converting user-input data of different formats into standardized data types, while logical conflict detection is the process of detecting contradictions between structured form fields and parameters in investment task descriptions through a built-in logic engine.

[0059] Furthermore, standard financial information is a standardized collection of financial data that has undergone integrity verification, type conversion, and logical conflict detection. The data format is unified and logically consistent, meeting the requirements for subsequent semantic parsing and combinatorial optimization. This information is the standardized processing result of the user's original input.

[0060] Understandably, the process involves first validating the structured form fields and investment task descriptions to ensure no required fields are missing; then performing type conversion to standardize the data format; and finally executing logic conflict detection to identify inconsistencies between parameters and obtain standard financial information.

[0061] Step S14: Input standard financial information into the language processing model for joint semantic encoding and parsing, identify the financial entities, quantitative constraints and logical relationships of the standard financial information, and map the financial entities, quantitative constraints and logical relationships into structured semantic slots;

[0062] It should be noted that the language processing model is based on the Transformer architecture and jointly fine-tuned on Chinese financial corpora and historical form data, possessing the ability to semantically understand and parse text in the financial field. This model can accurately identify key information and logical relationships in financial texts, providing core algorithmic support for semantic parsing. In this embodiment, the language processing model may include a large-scale language model.

[0063] Additionally, joint semantic encoding and parsing involves jointly encoding structured form field labels and standard financial information to form context-aware semantic vectors and then parsing key information. This process combines the structured features of form fields with the semantic features of textual information, improving the accuracy of parsing.

[0064] Furthermore, a financial entity is a specific financial object involved in standard financial information, including a specific name or identifier; quantitative constraints are numerical restrictive conditions contained in standard financial information; and logical relationships are the relationships between different elements in standard financial information, including causal conditions and progressive relationships.

[0065] Furthermore, structured semantic slots are predefined, standardized data structures used to store key parsed information. They contain multiple fields, such as target constraints, asset pools, and perturbation strategies, with each field corresponding to a specific type of parsing result. Semantic slots provide a unified storage format for the parsing results, facilitating subsequent processing and retrieval.

[0066] Understandably, standard financial information is input into a language processing model, and semantic vectors are formed through joint semantic encoding. The quantitative constraints and logical relationships of financial entities are identified, and these parsing results are then mapped to the corresponding structured semantic slots to complete the transformation from text information to structured data.

[0067] Step S15: Perform vector retrieval in the preset financial knowledge base to determine the regulatory rules, asset attributes, and historical pattern information of the structured semantic slots. Then, complete the knowledge of the structured semantic slots based on the regulatory rules, asset attributes, and historical pattern information to obtain a structured task description.

[0068] It should be noted that the pre-defined financial knowledge base is a database storing financial-related knowledge, including regulatory provisions, index compilation instructions, asset attributes, industry classifications, historical event patterns, and more. This knowledge base provides rich domain knowledge support for knowledge completion and consistency verification, ensuring that the parsing results comply with industry standards and regulatory requirements.

[0069] Additionally, vector retrieval is a method that quickly matches relevant knowledge by calculating the similarity between structured semantic slot vectors and text vectors in a pre-defined financial knowledge base. This method can efficiently locate knowledge content related to semantic slots, providing accurate candidate information for knowledge completion.

[0070] Furthermore, regulatory rules are financial regulatory provisions and requirements stored in a pre-defined financial knowledge base, including concentration limits and single-asset holding ratio limits, used to ensure that the investment portfolio meets compliance requirements. The integration of regulatory rules is an important guarantee for the compliance of the portfolio strategy.

[0071] Furthermore, asset attributes are the inherent characteristics of various assets stored in the pre-defined financial knowledge base, including information such as the asset's liquidity, risk level, return characteristics, and industry affiliation. These attributes provide detailed asset-based data for asset allocation and portfolio optimization.

[0072] Additionally, historical pattern information comprises data stored in a pre-defined financial knowledge base, including case studies of asset performance events and responses under historical market scenarios. This includes historical responses to asset allocation pattern disturbances under different market scenarios. Historical pattern information provides a reference for generating the current portfolio strategy.

[0073] Furthermore, knowledge completion is a process of supplementing and improving incomplete descriptions in structured semantic slots based on regulatory rules, asset attributes, and historical pattern information obtained through vector retrieval. In this embodiment, high dividend payouts can be completed by providing a specific list of constituent stocks of the CSI Dividend Index to ensure the completeness of semantic slot information.

[0074] Understandably, the structured semantic slots are retrieved by vector in the pre-set financial knowledge base, and the corresponding regulatory rules, asset attributes and historical pattern information are matched. This information is then used to supplement the structured semantic slots, fill in information gaps and correct inconsistencies, and finally form a structured task description.

[0075] In one feasible implementation, step S15 may include steps S151 to S154:

[0076] Step S151: Convert the structured semantic slots into query vectors, and perform vector retrieval in the preset financial knowledge base based on the query vectors to obtain candidate knowledge entries.

[0077] It should be noted that the query vector is vector data used for similarity matching, transformed from structured semantic slots, and can accurately represent the core semantic information of the structured semantic slots. This vector is generated through a specific encoding algorithm, providing a data foundation for quickly locating relevant knowledge in a pre-defined financial knowledge base.

[0078] Furthermore, candidate knowledge entries are sets of knowledge related to the query vector matched from a pre-defined financial knowledge base through vector retrieval. These entries cover various types of information that may be used to complete structured semantic slots, providing ample options for the subsequent selection of target knowledge entries.

[0079] Understandably, the structured semantic slots are first converted into query vectors using an encoding algorithm, and then the query vectors are used to perform similarity matching in a preset financial knowledge base to filter out knowledge content related to the structured semantic slots and form candidate knowledge items.

[0080] Step S152: Determine the target knowledge item based on the semantic similarity between the candidate knowledge item and the query vector and the timeliness weight of the candidate knowledge item.

[0081] It should be noted that the timeliness weight is a weight value determined based on the timestamp of the candidate knowledge item, used to measure the freshness and applicability of the candidate knowledge item. Generally, the more recent the timestamp of the candidate knowledge item, the higher the timeliness weight, and the better it reflects the latest state of the current market and regulation.

[0082] Furthermore, the target knowledge entry is the optimal knowledge entry selected after weighted calculation, which combines the semantic similarity between candidate knowledge entries and the query vector with the timeliness weight of the candidate knowledge entries. This entry can best meet the needs of filling structured semantic slots, so as to complete the relevance and timeliness of the information.

[0083] Understandably, the semantic similarity score between each candidate knowledge item and the query vector is calculated, and a weighted calculation is performed by combining the timeliness weight of each candidate knowledge item. The candidate knowledge item with the highest score is selected as the target knowledge item based on the weighted score ranking.

[0084] Step S153: Determine regulatory rules, asset attributes and historical pattern information based on the target knowledge items, and complete the structured semantic slots with knowledge based on the regulatory rules, asset attributes and historical pattern information to obtain the completed semantic slots.

[0085] It should be noted that semantic slot completion utilizes the regulatory rules, asset attributes, and historical pattern information from the target knowledge entry to supplement and improve incomplete descriptions in the structured semantic slots, resulting in semantic slots with comprehensive and accurate information that meets the preset structured requirements.

[0086] Understandably, the process involves extracting regulatory rule asset attributes and historical pattern information from target knowledge entries, associating and matching this information with structured semantic slots, filling information gaps in the structured semantic slots, correcting inconsistencies, and thus obtaining complete semantic slots.

[0087] Step S154: Perform structured verification on the completed semantic slots to obtain a structured task description.

[0088] It should be noted that structured validation is a process of checking whether the completed semantic slots conform to preset pattern constraints and type constraints. This includes checks on field mandatory requirements, type validation, enumeration value validity, and cross-field consistency. This validation process ensures the structural regularity and logical consistency of the completed semantic slots.

[0089] Understandably, a comprehensive check is performed on the completed semantic slots using preset validation rules. If issues such as missing fields, incorrect types, illegal enumeration values, or cross-field logical conflicts are found, a traceable error code and location information are generated. If the validation passes, a structured task description is obtained directly.

[0090] Step S20: Generate multi-source investment information based on the structured task description;

[0091] It should be noted that multi-source investment information is standardized information formed by integrating perspectives from multiple sources, including opinions generated by user subjective opinion language models and structured data opinions obtained from statistical econometric models or factor models. These opinions are uniformly encoded to form opinion vectors and confidence matrices usable by combined optimization models.

[0092] Understandably, based on the structured task description, one or more language models are invoked to independently generate opinion samples multiple times by adjusting the sampling temperature or using random seeds, recording information such as the confidence interval of expected return parameters. Simultaneously, user subjective opinions and opinions output by statistical models are acquired, the statistical characteristics of each sample are calculated, the confidence scores of assets with significant disagreements are reduced, and all opinions are merged using a weighted average or hierarchical voting mechanism to generate a unified opinion vector and confidence matrix, i.e., multi-source investment information.

[0093] In one feasible implementation, step S20 may include steps S21 to S24:

[0094] Step S21: Input the structured task description into the language processing model for risk prediction to obtain expected return parameters and risk information;

[0095] It should be noted that risk prediction is a process in which a language processing model, based on a structured task description and historical market data, predicts the risk level of an asset within a given prediction window. The prediction process considers factors such as the asset's historical volatility characteristics, industry trends, and market environment to ensure the reasonableness of the prediction results.

[0096] Furthermore, the expected return parameter is the level of return that the asset may achieve within a future prediction window, as predicted by the language processing model. It is usually presented as a return range or a specific numerical value. This data is one of the core references for subsequent asset allocation and portfolio optimization.

[0097] Additionally, risk information comprises various data related to asset risk output by the language processing model, including indicators such as the asset's volatility, maximum drawdown, and downside risk. This information comprehensively reflects the asset's risk characteristics, providing data support for portfolio risk control.

[0098] Understandably, when a structured task description is input into a language processing model, the model combines its learned financial knowledge and historical data to analyze the future performance of assets and output corresponding expected return parameters and risk information, providing basic data for subsequent calculations.

[0099] Step S22: Calculate the asset's expected return parameters and the variance of view uncertainty based on the expected return parameters and risk information;

[0100] It should be noted that the expected return parameter of an asset is a level of expected return calculated for a single asset, and is the result of a refined breakdown of the expected return parameter output by the language processing model. The expected return parameter of each asset is related to the asset's attributes, market prospects, and constraints, providing a basis for the weight allocation of individual assets.

[0101] Additionally, opinion uncertainty variance is a statistical indicator used to measure the degree of fluctuation in opinion samples generated by a language processing model, reflecting the level of uncertainty in the prediction results. This variance is obtained by calculating the dispersion of multiple independently generated opinion samples, providing a quantitative basis for subsequent confidence level adjustments.

[0102] Understandably, based on expected return parameters and risk information, the analysis is performed by asset dimension to determine the expected return parameters for each asset. At the same time, the variance of opinion uncertainty is obtained by calculating the statistical characteristics of multiple generated opinion samples.

[0103] Step S23: Perform weighted fusion based on the asset expected return parameter, opinion uncertainty variance and preset confidence weight to obtain opinion vector and confidence matrix;

[0104] It should be noted that the pre-set reliability weights are pre-defined values ​​used to weigh the importance of different source viewpoints, including the weight of the language processing model's viewpoint and the weights of other potential viewpoint sources. These weights can be obtained through training based on historical prediction errors or configured directly according to business needs to facilitate the appropriate integration of different viewpoints.

[0105] Additionally, weighted fusion is a process that combines the expected return parameters of assets and the variance of opinion uncertainty according to pre-set confidence weights. During the fusion process, assets with significant disagreements are subject to confidence reduction through Bayesian contraction estimation or penalty coefficients to ensure the reasonableness of the fusion result.

[0106] Furthermore, the viewpoint vector is a standardized return prediction vector formed after weighted fusion, containing the final expected return parameters for each asset and corresponding to the asset pool index. This vector is one of the core inputs to the portfolio optimization model and directly affects the asset allocation ratio.

[0107] Furthermore, the confidence matrix is ​​a diagonally dominant matrix data structure. The matrix elements reflect the confidence level of the expected return parameters of the corresponding assets, and the diagonal elements are typically derived from the variance of opinion uncertainty. This matrix can quantify the reliability of opinions, providing a basis for risk constraints in portfolio optimization.

[0108] Understandably, based on pre-set confidence weights, the expected return parameters of each asset are weighted and calculated. At the same time, the confidence level is adjusted in combination with the variance of opinion uncertainty, and the confidence level of assets with disagreements is reduced, ultimately forming an opinion vector and a confidence matrix.

[0109] Step S24: Generate multi-source investment information based on opinion vectors and confidence matrices.

[0110] It is understandable that the opinion vector and confidence matrix are integrated so that their data structures are aligned with the asset pool index, forming a unified format of multi-source investment information for subsequent multi-scenario combination pre-configuration.

[0111] Step S30: Based on the structured task description, multi-source investment information and historical market data, generate a target portfolio scheme under a preset market scenario, wherein the target portfolio scheme includes a basic portfolio scheme and a backup portfolio scheme.

[0112] It should be noted that historical market data refers to past market-related data used for backtesting and scenario simulation, including historical return, volatility, correlation, and liquidity data, and is an important basis for evaluating the performance of portfolio strategies. This data will be used to label market scenarios and verify the feasibility of the strategy.

[0113] Additionally, preset market scenarios are predefined categories of market states, including scenarios such as bull markets, bear markets, sideways markets, and volatile markets. These can be automatically identified through historical data clustering or manually labeled by users. Different scenarios correspond to different market operating characteristics and risk-return performance.

[0114] Furthermore, the target portfolio scheme is a set of portfolio configuration schemes generated to adapt to different preset market scenarios. It includes a basic portfolio scheme and a backup portfolio scheme, and each scheme includes core elements such as asset weight allocation and risk buffer. This set of schemes needs to be evaluated through historical backtesting and scenario simulation.

[0115] Furthermore, the basic portfolio solution is a core configuration scheme generated based on the current market scenario and structured task description. It serves as the primary basis for portfolio operation and has clear profit objectives and risk constraints. This scheme must meet the core needs of users and be adapted to the current market environment.

[0116] Additionally, the backup portfolio options are alternative configurations generated for different types of pre-set disturbances or changes in market scenarios. These options differ from the basic portfolio options by less than a pre-set upper limit and include contingency parameters such as risk buffer ratios and priority trading orders. These options are used for rapid adjustments in response to market disturbances.

[0117] Understandably, by combining structured task descriptions with multi-source investment information and historical market data, the time scale is first divided into three levels: strategic, tactical, and execution. Basic combination schemes and multiple backup combination schemes are then generated for different preset market scenarios. Historical backtesting and scenario simulations are performed on each scheme to calculate feasibility, robustness, and execution complexity indicators. The schemes and evaluation results are then stored in a scheme library.

[0118] In one feasible implementation, step S30 may include steps S31 to S34:

[0119] Step S31: Use the investment objectives and constraints in the structured task description as optimization conditions, and generate a basic portfolio scheme based on the optimization conditions, multi-source investment information and historical market data.

[0120] It should be noted that the optimization conditions are the basis for combination optimization, which is composed of the investment objectives and constraints in the structured task description. The investment objectives clearly define the return-related demands, such as the target rate of return range, while the constraints define the investment boundaries, such as the upper limit of industry concentration. Together, they determine the direction and limitations of the combination optimization.

[0121] Understandably, the investment objectives and constraints in the structured task description are determined as optimization conditions. Combined with opinion vectors and confidence matrices from multi-source investment information, as well as historical market data, an external optimization solver is invoked to generate basic combination schemes that meet the optimization conditions.

[0122] Step S32: Generate a backup combination scheme based on the preset disturbance type and the basic combination scheme, wherein the backup combination scheme includes the disturbance type and the preset switching conditions;

[0123] It should be noted that the preset disturbance types are predefined types of market events that may affect the operation of the portfolio, including single asset suspension, abnormal industry fluctuations, large-scale redemptions, etc., covering various disturbance scenarios at the asset level, industry level, market level, and capital level.

[0124] Additionally, the backup combination plan is an alternative configuration plan generated for the preset disturbance type. The difference between it and the basic combination plan is less than the preset upper limit. It includes emergency parameters such as risk buffer ratio, set of assets that can be quickly liquidated, and priority transaction order, which are used for rapid switching when a disturbance occurs.

[0125] Furthermore, the preset switching conditions are the specific conditions for triggering the activation of the backup combination scheme, including the judgment criteria such as disturbance type, disturbance level, and impact scope. For example, when a single asset is suspended from trading and the weight ratio of that asset exceeds 5%, the corresponding backup scheme is triggered.

[0126] Understandably, for various preset disturbance types, and provided that the constraints are met, backup combination schemes are generated based on the basic combination scheme. The disturbance type and preset switching conditions corresponding to each backup combination scheme are clearly defined so that the difference between the backup scheme and the basic scheme is controllable.

[0127] Step S33: Under the preset market scenario, historical backtesting and scenario simulation are performed on the basic combination scheme and the backup combination scheme respectively to obtain the target evaluation index;

[0128] It should be noted that the preset market scenarios are predefined categories of market states, including bull markets, bear markets, sideways markets, and highly volatile markets. These can be automatically identified through historical data clustering or manually labeled by users, and each scenario corresponds to unique market operation characteristics.

[0129] In addition, historical backtesting is the process of simulating the operation of a portfolio strategy on historical market data. It simulates changes in holdings and accumulation of returns at a preset frequency, such as monthly rebalancing, in order to verify the performance of the strategy in the past market.

[0130] Furthermore, scenario simulation assesses the resilience of portfolio solutions under stress scenarios by simulating extreme market environments such as historical extreme declines and liquidity crises, thus supplementing and reinforcing historical backtesting.

[0131] In addition, target evaluation metrics are quantitative indicators that measure the performance of portfolio strategies, including feasibility metrics such as turnover rate and market impact cost, robustness metrics such as maximum drawdown and value at risk, and execution complexity metrics such as the number of assets involved.

[0132] Understandably, under various preset market scenarios, historical backtesting and scenario simulations are performed on the basic combination scheme and the backup combination scheme respectively, and the target evaluation indicators corresponding to the two types of schemes are calculated to comprehensively quantify the overall performance of the schemes.

[0133] In this embodiment, the scheme will be backtested based on historical data from the past three years, and simulations of extreme historical declines will be overlaid to calculate target evaluation indicators such as annualized return, maximum drawdown, and average turnover rate.

[0134] Step S34: Generate the target combination scheme based on the basic combination scheme, the backup combination scheme, and the target evaluation index.

[0135] Understandably, the basic combination scheme, each backup combination scheme, and the corresponding target evaluation indicators are integrated to clarify the relationship between the schemes and the applicable scenarios, forming a complete target combination scheme, which is then stored in the scheme library for future use.

[0136] Reference Figure 2 , Figure 2This is a schematic diagram of the first multi-scenario forward-looking combination pre-configuration process of the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application. Figure 2 As shown, starting with a structured task description, the system identifies investment cycles and risk preferences, divides timescales into strategic, tactical, and execution layers, selects historical sample intervals, and labels market scenarios. It can utilize multi-source opinion vectors and confidence matrices to generate basic portfolio solutions using opinion vectors and scenario labels, while simultaneously generating alternative portfolio solutions for preset disturbance types. Next, it performs historical backtesting and stress scenario simulations to calculate feasibility robustness and execution complexity indices, generates scenario label trigger rule thresholds for solution metadata, and inputs these indices into the solution structure and metadata of the solution library. This allows for the generation of suitable asset allocation solutions for investment portfolios under different market scenarios, and the feasibility and effectiveness of these solutions can be evaluated through historical backtesting and scenario simulations. It can provide investors with customized investment portfolio solutions under different market conditions, while ensuring the feasibility and robustness of the solutions.

[0137] In one feasible implementation, multi-scenario forward-looking combined pre-configuration can also refer to Figure 3 , Figure 3 This is a schematic diagram of the second multi-scenario forward-looking portfolio pre-configuration process in the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application. Figure 3 As shown, fund managers log into the system, select a strategic asset allocation scenario, fill in form parameters, including investment period, benchmark index, return target, maximum drawdown, industry limits, and capital size, and set disturbance handling strategies, such as partial recovery during trading halts, defensive measures during market downturns, and prioritizing highly liquid assets for large-scale redemptions. The system generates a structured task description, target constraints, asset pool, and disturbance preferences through semantic parsing, knowledge-enhanced Natural Language Processing (NLP), and financial knowledge base completion. It also generates user viewpoint vectors, a large language model (LLM), and a statistical viewpoint model output. The right side describes the system's operation after activating the forward-looking pre-configuration module, generating a basic portfolio scheme and backup schemes for different disturbance scenarios. Historical backtesting is performed for stress testing and scenario simulation, and feasibility, robustness, and execution complexity indicators are calculated. Schemes are written to a scheme library and stored using a unified JSON structure. The front end displays the performance, returns, risks, turnover rates, and costs of each scheme. Users select a basic scheme and confirm a set of backup schemes, constructing a scenario-scheme mapping matrix, performing structural clustering to identify alternative scheme groups, and completing the multi-scenario forward-looking portfolio pre-configuration process.

[0138] Step S40: Determine the perturbation feature vector based on the target combination scheme, and generate a re-optimization task description based on the perturbation feature vector;

[0139] It should be noted that the disturbance feature vector is vector data used to describe key information about disturbance events, including disturbance type, disturbance level, impact range, and estimated duration. It can identify disturbances at different levels, such as the asset layer, industry layer, market layer, and funding layer. When multiple disturbances exist, the logical relationships and priorities between them are recorded.

[0140] Additionally, the re-optimization task description is a task specification document that guides the combined re-optimization. It contains core content such as re-optimization mode labels and target strategy combinations. The re-optimization modes include local repair mode and overall replanning mode.

[0141] Understandably, based on the target portfolio scheme, continuous monitoring of market data, transaction execution information, and external event streams is performed. Disturbances are identified using detection methods such as threshold sliding windows, generating disturbance feature vectors. Combining the evaluation results in the current position status scheme library with user preset preferences, a dynamic inference engine is used to select a re-optimization mode, determine the corresponding strategy from the expert strategy library, and then generate a re-optimization task description.

[0142] In one feasible implementation, step S40 may include steps S41 to S45:

[0143] Step S41: During the operation of the target combination scheme, collect market data, transaction execution information, and external event streams;

[0144] It's important to note that market data is a collection of data reflecting the real-time trading status of the market, including core data such as price and volume from Level-1 Market Data (L1MD) and Level-2 Market Data (L2MD). This data forms the foundation for monitoring market volatility and asset performance. Additionally, trade execution information records the process of implementing portfolio trades, including trade confirmations, trade prices, trade quantities, and execution times, reflecting the actual implementation of the trades. Furthermore, external event streams consist of data on various influencing factors from outside the market, including textual or data information such as news announcements and macroeconomic data releases. These events can directly impact asset prices and market trends.

[0145] Understandably, during the operation of the target portfolio plan, market data, transaction execution information, and external event streams are collected in real time through data interfaces in order to comprehensively capture market dynamics and various potential influencing factors.

[0146] Step S42: Identify and classify disturbance events in market data, transaction execution information, and external event streams to obtain disturbance event classification results;

[0147] It should be noted that disturbance event identification and classification is the process of filtering and categorizing events that affect the combined operation through specific detection methods. Employing detection methods based on thresholds, sliding windows, and pattern recognition, it can accurately capture various anomalies or sudden situations. The identification process combines data features and event attributes to ensure no key disturbances are missed.

[0148] In addition, the disturbance event classification results are the results of classifying the identified disturbance events by level and type, including asset level, industry level, market level and capital level, as well as specific types such as single asset suspension, industry fluctuations, and large-scale redemptions.

[0149] In this embodiment, the suspension of trading of constituent stocks will be classified as asset-level disturbances, and medium-sized fund redemptions will be classified as fund-level disturbances.

[0150] Understandably, the collected market data, transaction execution information, and external event streams are analyzed, disturbance events are identified through preset detection methods, and then classified into corresponding levels and types according to preset rules to obtain disturbance event classification results.

[0151] Reference Figure 4 , Figure 4 This is a schematic diagram of the disturbance monitoring and event classification process in the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application. Figure 4 As shown, the system accesses three types of raw information: real-time market data streams, fund flows and subscription / redemption records, and announcements and macroeconomic event text streams. It then runs five parallel processing paths: price and return monitoring, trading volume and liquidity monitoring, industry index and benchmark index change monitoring, fund event monitoring, and text event analysis. Each path extracts features from the corresponding data. Short-term statistical characteristics are calculated using sliding window and weighted moving average (EWMA) methods in the liquidity threshold monitoring, industry and market volatility clustering, extreme fund flow event identification, and rule change and institutional change identification stages. This transforms the raw data into quantifiable abnormal signals. Based on preset rules, these signals are categorized into four levels: asset-level disturbance identification, industry and market-level disturbance identification, fund-level disturbance identification, and institutional change disturbance identification. A disturbance feature vector D is generated, containing the disturbance type, level, impact range, and duration estimate. Further disturbance features are added to form a complete description. This vector is written to the disturbance event log and simultaneously sent to the re-optimization strategy routing module, providing a decision-making basis for subsequent partial repair or overall replanning.

[0152] Step S43: Generate a disturbance feature vector based on the disturbance event classification result, wherein the disturbance feature vector includes the disturbance type and disturbance level;

[0153] It should be noted that the disturbance type is a definition of the essential attributes of the disturbance event, including specific categories such as single asset suspension, price limits, abnormal industry fluctuations, large-scale subscriptions and redemptions, and changes in regulatory rules. Each type corresponds to a specific impact logic.

[0154] Furthermore, the disturbance level is a standard for measuring the degree of impact of a disturbance event, and is divided into levels such as severe, moderate, and minor, determined according to the potential scope and intensity of the disturbance's impact on portfolio performance. For example, the disturbance level is determined by at least one threshold, such as the magnitude of volatility jumps, the percentage decrease in liquidity, the weighting of suspended assets, and the redemption ratio.

[0155] Understandably, based on the classification results of disturbance events, key information such as disturbance type, disturbance level, impact range, and duration estimation are extracted, and this information is quantified to construct a disturbance feature vector.

[0156] In this embodiment, when a constituent stock is suspended from trading and a medium-sized fund redemption occurs simultaneously, the generated disturbance feature vector will include information on the type of severe disturbance at the asset level and medium disturbance at the fund level, as well as the corresponding impact range and duration estimates.

[0157] Step S44: Determine the target re-optimization mode based on the disturbance feature vector, the current position status, and the historical evaluation results, and determine the target strategy based on the target re-optimization mode;

[0158] It should be noted that the current portfolio status reflects the real-time asset allocation during the execution of the target portfolio strategy. This includes the latest weights, liquidity scores, industry tags, and correlation matrices of each asset, serving as a crucial benchmark for rebalancing decisions. This information is updated in real time to ensure consistency with the actual portfolio.

[0159] In addition, historical evaluation results are the evaluation data of the basic combination scheme and the backup combination scheme in the target combination scheme in the past, including feasibility indicators, robustness indicators, execution complexity indicators, etc., which are stored in the scheme library for query and retrieval.

[0160] Furthermore, the target re-optimization mode is a combination of adjustment modes determined based on the disturbance situation, including a local repair mode and a global replanning mode, which are applicable to disturbances of different intensities and ranges, respectively. The local repair mode is for local or minor disturbances, while the global replanning mode is for systemic or severe disturbances.

[0161] Furthermore, the target strategy is a strategy selected from a pre-defined expert strategy library that is suitable for the target re-optimization mode. The expert strategy library includes strategies such as minimum variance strategy, risk parity strategy, maximum Sharpe ratio strategy, and low transaction cost strategy. Different modes correspond to different optimal strategies.

[0162] Understandably, by combining the perturbation feature vector, the current position status, and the historical evaluation results retrieved from the solution library, the dynamic inference engine performs rule matching and scoring calculation to determine the target re-optimization mode, and then selects the target strategy that is suitable for the mode from the expert strategy library.

[0163] In one feasible implementation, step S44 may include steps S441 to S444:

[0164] Step S441: Match the perturbation feature vector with the conditional expression of the preset rule table to obtain candidate re-optimization modes. The candidate re-optimization modes include local repair mode and global replanning mode.

[0165] It should be noted that the preset rule table is a data table that stores routing rules in the form of "conditional expression – action instruction". The rules are associated with perturbation feature vectors, scene labels and user preferences, and clarify the corresponding re-optimization pattern tendency under different perturbation conditions. These rules are the core basis for dynamic reasoning, so as to ensure the standardization and consistency of pattern matching.

[0166] Additionally, the conditional expression is a logical expression used in the preset rule table to determine the matching relationship of disturbances, including the determination conditions such as disturbance type, level, and scope of influence.

[0167] Furthermore, the candidate re-optimization patterns are a set of alternative adjustment patterns obtained through conditional expression matching. This set includes only local repair patterns and global reprogramming patterns, forming the basis for subsequent scoring and selection. This set ensures focused pattern selection and avoids irrelevant patterns interfering with the decision.

[0168] Furthermore, the local repair mode is an adjustment mode targeting local or minor disturbances. It limits the assets to be adjusted to the local assets affected by the disturbance and their highly correlated assets, while imposing weight locks or change caps on unaffected assets. The core objective of this mode is to minimize adjustment costs and avoid global structural changes.

[0169] The other option, the overall restructuring mode, is an adjustment mode for systemic or severe disturbances. It allows for structural adjustments across the entire asset pool, prioritizing the use of expert strategies aimed at robustness or defense. This mode is suitable for scenarios where the market environment undergoes fundamental changes, requiring a restructuring of the core portfolio configuration.

[0170] Understandably, key attributes such as disturbance type, level, and scope of influence are extracted from the disturbance feature vector and matched one by one with the conditional expressions in the preset rule table to select re-optimization modes that meet the current disturbance situation, thus forming candidate re-optimization modes.

[0171] In this embodiment, when the disturbance feature vector is a severe disturbance at the asset layer and a moderate disturbance at the funding layer, the candidate re-optimization modes are obtained by matching the conditional expressions in the preset rule table as a local repair mode and an overall replanning mode.

[0172] Step S442: Based on the current position status and historical evaluation results, calculate the execution cost score and risk suitability score under the candidate re-optimization mode;

[0173] It should be noted that the execution cost score is a scoring indicator that quantifies the implementation cost of candidate re-optimization models. Its core criteria include the penalty score for the difference between the proposed solution and the current holdings, the expected transaction cost score, and the execution complexity score. A lower score indicates a lower execution cost, better meeting the requirements for low-cost adjustments.

[0174] Additionally, the risk fit score is a rating metric that measures the degree to which candidate re-optimization models fit the current perturbation risk. It is calculated based on the robustness score of the scheme under similar historical perturbation scenarios and the matching degree of risk exposure. The higher the score, the stronger the model's ability to cope with the current risk and the better its fit.

[0175] Understandably, the difference, expected transaction costs, and execution complexity of the candidate re-optimization models are calculated based on the current position status, and converted into an execution cost score. Data from scenarios similar to the current disturbance are retrieved from historical evaluation results to analyze the robustness and risk response effect of the candidate models in past scenarios, and a risk adaptation score is generated.

[0176] Step S443: Perform a weighted synthesis based on the execution cost score and risk adaptation score to obtain the target re-optimization model;

[0177] Understandably, the weighting coefficients for the pre-set execution cost score and risk adaptation score can be obtained based on training based on historical decision-making effects or configured according to business needs. The comprehensive score of each candidate re-optimization mode is calculated through a weighted summation formula, and the candidate mode with the highest comprehensive score is selected as the target re-optimization mode.

[0178] Step S444: Determine the target strategy based on the target re-optimization mode and the preset expert strategy library.

[0179] It should be noted that the preset expert strategy library is a database storing various professional investment strategies, including minimum variance strategies, risk parity strategies, maximum Sharpe ratio strategies, and low transaction cost strategies. Each strategy is labeled with its appropriate re-optimization mode and applicable scenarios. This knowledge base provides rich professional resources for strategy selection, ensuring both adaptability and professionalism.

[0180] Furthermore, the target strategy is the optimal strategy selected from a pre-defined expert strategy library, tailored to the target re-optimization mode. It accurately matches the current disturbance response needs and mode execution requirements. The choice of the target strategy directly determines the specific execution direction and effect of the re-optimization.

[0181] Understandably, based on the determined target re-optimization mode, the preset expert strategy library is queried, all expert strategies marked as suitable for the mode are selected, and the one that best fits the needs is chosen as the target strategy, taking into account the current disturbance characteristics and holding status.

[0182] In this embodiment, when the target re-optimization mode is the local repair mode, a low transaction cost expert strategy that is suitable for the mode is selected from the preset expert strategy library and determined as the target strategy in order to minimize the adjustment cost.

[0183] Reference Figure 5 , Figure 5 This is a schematic diagram of the runtime disturbance and re-optimization process of the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application. Figure 5As shown, the system continuously scans for real-time market data, capital announcements, and other information during operation. Upon detecting a disturbance event such as the suspension of trading in constituent stock A or a medium-sized redemption, event feature extraction is triggered. The system calculates the concentration of suspended asset weights, redemption ratios and concentrations, and current market status characteristics. The event classification module labels the event as severe disturbance at the asset layer and moderate disturbance at the capital layer. A composite disturbance feature vector D is then generated, integrating the additional time and trigger source of disturbances at the asset and capital layers. Simultaneously, the system reads the current holding status w0, liquidity correlation, and industry tags. It queries the solution library to obtain the evaluation results of basic and high-liquidity backup solutions, and reads the user's disturbance preference configuration. It prioritizes the local repair and liquidity enhancement mode, constructs a re-optimization model with w0 as a reference to set the upper limit for asset weight changes, introduces differential and transaction cost constraints, and calls the optimization algorithm. The solver performs fine-tuning when computational resources or time constraints meet a preset threshold; otherwise, it outputs a feasible solution. The overall weight difference of the re-optimized scheme w1 does not exceed a preset threshold. Simultaneously, it generates a set of comparison schemes, including the original holding unchanged scheme, a complete switch to a defensive backup scheme, and a locally repaired scheme with controlled differences. It displays a comparison of return curves, risk indicators, turnover rate, and estimated transaction costs, and generates natural language explanations to illustrate key changes such as concentration changes and improved liquidity. It selects a locally repaired scheme or makes minor adjustments to the scheme. Selecting a scheme can directly execute the instruction, converting the re-optimized scheme into buy / sell instructions. At the same time, it records decision and preference data, including the perturbation feature vector D, the candidate scheme scoring results, the final selection, and the fine-tuning range. When a locally repaired scheme is selected in multiple similar scenarios, the priority of the locally repaired mode is automatically increased as the default recommendation for this type of perturbation.

[0184] Step S45: Generate a re-optimization task description based on the target re-optimization mode and target strategy.

[0185] It is understandable that by integrating the determined target re-optimization mode and target strategy, clarifying the execution logic and related parameters of both, a complete re-optimization task description is formed, providing input for the combinatorial optimization and replanning module.

[0186] Reference Figure 6 , Figure 6 This is a schematic diagram of the re-optimization strategy routing and local repair process in the first embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning of this application. Figure 6As shown, starting with the disturbance feature vector D, the current holding state w0 is read. Simultaneously, basic and related backup schemes are loaded from the scheme library, and user disturbance preferences and historical decision records are read. The dynamic inference engine then performs rule matching and score calculation, selecting a re-optimization mode based on the results. If local repair is selected, the adjustment is limited to a subset of assets, and unaffected assets are locked. Low-transaction-cost experts are selected from the expert strategy library to generate a local repair re-optimization task description. If overall replanning is selected, structural adjustments are allowed across the entire asset pool. Robust or defensive experts are selected from the expert strategy library to generate an overall replanning re-optimization task description. Both task descriptions are submitted to the portfolio optimization and replanning module. The module returns candidate re-optimization schemes and difference assessment results, updating preference parameters and rule weights accordingly. The results are then sent to the result output and closed-loop interaction module. The entire structure, through score-driven mode selection, quickly maps disturbance events into executable re-optimization tasks, achieving differentiated portfolio adjustments and a continuous closed loop with user preferences. This ensures that the portfolio can dynamically adjust with minimal cost or maximum defensiveness after a disturbance occurs.

[0187] Step S50: Optimize the candidate asset weight allocation scheme based on the re-optimization task description, the preset difference penalty coefficient, and the weight vector of the current holdings, and determine the investment strategy allocation scheme.

[0188] It should be noted that the preset difference penalty coefficient is a pre-set coefficient used to constrain the range of portfolio fluctuations. Used in conjunction with the difference index, it can control transaction costs and market shock risks. This coefficient is determined by balancing the return objective and the cost of weight changes, and is an important component of the optimization objective function.

[0189] Additionally, the investment strategy allocation plan is the final allocation plan output after re-optimization, including updated asset weight allocation, fund usage ratios, and execution instruction suggestions. This plan must meet user and regulatory constraints, and the weight changes must be within a preset range. Furthermore, the current holdings weight vector is a vector data composed of the weights of various assets in the current holdings. Each element corresponds to the allocation ratio of a single asset in the current portfolio, comprehensively reflecting the current asset allocation structure.

[0190] Furthermore, the candidate asset weight allocation schemes are potential asset allocation schemes initially generated based on the re-optimization task description. These schemes include candidate allocation weights for various assets and serve as the basis for subsequent difference calculations and optimization solutions. These schemes must undergo difference verification and constraint satisfaction checks before they can become the final feasible schemes.

[0191] Understandably, taking the current holdings as a reference, a re-optimization model is constructed based on the re-optimization task description. Norm differences and other indicators are calculated based on the weight vectors of the current holdings to measure the magnitude of portfolio changes. A preset difference penalty coefficient is multiplied by the overall difference index as a penalty term, which is then incorporated into the optimization objective function. An external optimization solver is called to perform calculations, optimizing the candidate asset weight allocation scheme and outputting an investment strategy allocation scheme while satisfying various constraints.

[0192] Reference Figure 7 , Figure 7 This diagram illustrates the output and closed-loop interactive interface of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application. Figure 7 As shown, the system starts with the input result set, including three candidate schemes: original solution, partial repair, and overall replanning. These results are fed into the portfolio performance comparison view, which presents differences in terms of return, volatility, drawdown, and turnover rate. At the same time, the risk and cost view displays transaction costs, slippage, and liquidity usage. The causal chain explanation view reveals the origin of the decision along the link from input fields to constraints to optimization to output. The disturbance event timeline view marks key nodes in the response path according to the disturbance type, level, and response. All views are integrated into the user interface. The user operation area provides three actions: confirmation, rejection, and fine-tuning. If the user selects confirmation or fine-tuning, the system immediately updates the user's preferences and decision records, and feeds the latest preferences back to the dynamic inference module for rule and parameter updates. This forms a closed loop from visualization to human decision-making to preference learning, making subsequent re-optimization routes more aligned with the fund manager's personal style and institutional constraints.

[0193] This embodiment provides an investment strategy configuration method based on form semantic parsing and dynamic reasoning. Through form-based semantic parsing, multi-source viewpoint fusion, multi-scenario forward-looking pre-configuration, controlled-degree-of-difference re-optimization, and closed-loop preference learning, it solves the technical problems of traditional portfolio optimization relying on static input, one-size-fits-all perturbation response, uncontrollable rebalancing costs, and difficulty in quantifying user intent. It achieves the beneficial effects of rapid connection between multi-scenario pre-configuration before operation and perturbation response during operation, differentiated processing of multiple perturbations at different levels, significant reduction in transaction costs and impact risks, and continuous self-improvement through human-machine collaboration and strategy evolution.

[0194] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 8 Step S50 of the investment strategy configuration method based on form semantic parsing and dynamic reasoning includes steps S51 to S54:

[0195] Step S51: Calculate the first norm difference and the second norm difference between the candidate asset weight allocation scheme and the current holding weight vector;

[0196] It should be noted that the first norm difference is an indicator used to measure the difference in total turnover between the candidate asset weight allocation plan and the current holding weight vector. It is calculated as the sum of the absolute values ​​of the changes in all asset weights. This indicator directly reflects the total scale of portfolio adjustments and is a core quantitative basis for controlling transaction costs.

[0197] Furthermore, the second norm dissimilarity is an indicator used to measure the difference in portfolio structure between the candidate asset weight allocation scheme and the current holding weight vector. It is calculated as the square root of the sum of the squares of the changes in all asset weights. This indicator can reflect the degree of change in the overall portfolio structure, preventing drastic changes in the core allocation logic.

[0198] Understandably, the weight vectors of the current holdings and the weight vectors of the candidate asset weight allocation schemes are extracted, and the change value of the weight of each asset class is calculated one by one. The first norm difference is obtained by summing the absolute values ​​of the weight changes, and the second norm difference is obtained by calculating the square root of the sum of squares of the weight changes.

[0199] In one feasible implementation, step S51 may include steps S511 to S513:

[0200] Step S511: Determine the first norm weight coefficient and the second norm weight coefficient based on the preset difference penalty coefficient;

[0201] It should be noted that the first norm weighting coefficient is used to weigh the importance of the first norm difference in the target difference. Its value is usually between 0 and 1, and together with the second norm weighting coefficient, it forms a complete weighting system. The magnitude of this coefficient directly affects the contribution of the total turnover rate difference to the target difference, which aligns with the need to control transaction costs.

[0202] Additionally, the second norm weighting coefficient is used to weigh the importance of the second norm difference in the target difference. Its value also ranges from 0 to 1, and its sum with the first norm weighting coefficient is 1. This coefficient primarily adjusts the impact of portfolio structural changes on the target difference, ensuring the stability of the portfolio's core configuration logic.

[0203] Understandably, based on the magnitude of the preset difference penalty coefficient and the business's emphasis on transaction costs and structural stability, the first norm weight coefficient and the second norm weight coefficient are allocated so that the weight allocation is consistent with the overall optimization objective.

[0204] Step S512: The first norm difference degree and the second norm difference degree are weighted according to the first norm weight coefficient and the second norm weight coefficient respectively to obtain the first weighted difference degree and the second weighted difference degree.

[0205] It should be noted that the first weighted difference is the result of multiplying the first norm difference by the first norm weight coefficient, and is a weighted total turnover rate difference indicator. This indicator retains the core information of the total turnover rate while adjusting its influence in the target difference through weighting.

[0206] Additionally, the second-weighted difference is the result of multiplying the second-norm difference by the second-norm weight coefficient, and is a weighted index of the difference in portfolio structure changes. This index can accurately reflect the degree of change in the core configuration of the portfolio after weight adjustment.

[0207] It is understandable that multiplying the first norm difference by the first norm weight coefficient yields the first weighted difference; multiplying the second norm difference by the second norm weight coefficient yields the second weighted difference, thus completing the weighted transformation of the two types of difference.

[0208] Step S513: Determine the target difference based on the first weighted difference and the second weighted difference.

[0209] It is understandable that by using a linear combination approach, the first weighted difference degree and the second weighted difference degree are added together to obtain the target difference degree, so that the target difference degree can comprehensively reflect the difference in total turnover rate and the difference in portfolio structure changes after weighting.

[0210] Step S52: Determine the target difference based on the preset difference penalty coefficient, the first norm difference, and the second norm difference.

[0211] It should be noted that the target difference is a comprehensive difference index obtained by weighting the first norm difference and the second norm difference by a preset difference penalty coefficient, which can take into account both the total turnover rate and the structural change constraints.

[0212] It is understandable that by pre-setting the weights of the first-norm difference degree and the second-norm difference degree, and then linearly combining the two according to the weights to obtain the overall difference degree, the overall difference degree is multiplied by the pre-set difference degree penalty coefficient to finally obtain the target difference degree, which provides a unified difference constraint basis for the subsequent construction of the target optimization problem.

[0213] Step S53: Construct the objective optimization problem based on the combined optimization objective function, objective difference degree, and re-optimization task description;

[0214] It should be noted that the combinatorial optimization objective function is a mathematical expression used to define the core objective of combinatorial optimization. It typically includes three parts: a profit objective, a risk objective, and a penalty term, where the penalty term consists of the objective difference degree. This function is the core carrier for balancing profit, risk, and adjustment costs, and directly determines the direction of optimization.

[0215] In addition, the objective optimization problem is a mathematical optimization problem formed by integrating and combining the objective function, objective difference degree and constraints in the description of the re-optimization task. It includes core elements such as objective function, variable boundary constraints, regulatory constraints and liquidity constraints.

[0216] It is understandable that, with the combinatorial optimization objective function as the core, the objective difference is incorporated as a penalty term into the objective function. At the same time, the re-optimization mode constraints, strategy constraints, and user-preset investment constraints in the re-optimization task description are included to construct a complete objective optimization problem so that the optimization process not only conforms to the adjustment direction but also satisfies various constraints.

[0217] In this embodiment, if the re-optimization task is described as a local repair mode, the objective optimization problem will take the upper limit of the weight change of the undisturbed assets as a constraint, and at the same time, the objective difference degree will be included as a penalty term along with the profit objective and the risk objective in the objective function, forming an optimization problem under multiple constraints.

[0218] Step S54: Solve the objective optimization problem to obtain the investment strategy allocation scheme.

[0219] Understandably, by selecting an appropriate external optimization solver based on the type of the target optimization problem, the external optimization solver can be invoked to solve the target optimization problem. First, a feasible solution that satisfies all hard constraints can be quickly obtained. Then, if time permits, fine optimization can be performed to finally output an investment strategy configuration scheme that meets the requirements.

[0220] In this embodiment, for the target optimization problem under the local repair mode, an open-source solver is selected to quickly obtain a feasible solution so that the disturbed assets are adjusted in place and the weight changes of the undisturbed assets do not exceed the upper limit. Then, the weights are finely adjusted through fine optimization so that the target difference is controlled within a preset range.

[0221] Reference Figure 9 , Figure 9 This is a schematic diagram of the difference-controlled re-optimization process of the second embodiment of the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application. Figure 9As shown, the current holding weight vector is read as a benchmark, and the weight vectors of candidate solutions are compared with it. The difference value is obtained by calculating the L1 or L2 distance. If the difference exceeds a preset threshold, the solution is rejected and an adjustment strategy is triggered. The strategy includes reducing the adjustable asset set or tightening the upper limit of weight changes. Alternatively, the solution can be directly rejected and a new candidate solution can be generated. If the difference does not exceed the threshold, the next step is to write the difference penalty term into the optimization objective function. The penalty term, together with the return, risk and cost, constitutes the comprehensive evaluation objective function. The four indicators of return, risk, cost and difference are weighted and summed in the objective function. The solver finds the weight configuration that maximizes the comprehensive score under the constraints. After selecting the solution with the highest comprehensive score, the final re-optimized solution is output. This process allows the new solution to control the distance from the existing holdings while improving the return and risk characteristics, avoiding excessive trading and impact costs, and achieving a gradual rebalancing with controlled difference.

[0222] This embodiment provides an investment strategy configuration method based on form semantic parsing and dynamic reasoning. By introducing joint weighting of first norm difference degree and second norm difference degree, dynamic adjustment of difference degree penalty coefficient, multi-constraint modeling of objective optimization problem, and closed-loop control of difference degree threshold, it achieves the beneficial effects of accurately controlling portfolio change range, significantly reducing turnover rate and impact cost, realizing gradual portfolio adjustment with controlled difference degree, and improving the robustness of plan execution while improving return and risk.

[0223] For example, to help understand the implementation process of the investment strategy configuration method based on form semantic parsing and dynamic reasoning obtained by combining this embodiment with the above embodiment one, please refer to... Figure 10 , Figure 10 A simplified flowchart illustrating an investment strategy configuration method based on form semantic parsing and dynamic reasoning is provided, specifically:

[0224] Starting from the human-computer interaction and form input module, user input is received and sent to the semantic parsing and knowledge enhancement module. This module calls the retrieval enhancement function and reads the financial knowledge base and regulatory rule base to complete the information. Subsequently, the multi-source viewpoint generation and unified encoding module calls the language model and statistical model to output a unified viewpoint. The result is sent to the multi-scenario forward-looking combination pre-configuration module. This module uses historical and real-time market data to perform backtesting and simulation, generates benchmark and backup plans, and writes them into the plan library in a unified format. During system operation, the disturbance monitoring and event classification module subscribes to market data, records detected disturbance events in the disturbance event log, and updates the disturbance history. The re-optimization strategy routing module has a built-in dynamic inference engine that reads... The system retrieves the rule table and strategy configuration, queries the solution library to obtain candidate solutions, scores them based on user preferences and preference features in the decision record library, selects either a local repair or overall replanning mode, and issues the task to the combinatorial optimization and replanning module. This module solves for new weights under the constraints of difference and transaction cost. The results are sent to the result output and closed-loop interaction module for display. At the same time, user confirmation or fine-tuning behavior is recorded and written back to the user preference and decision record library for parameter updates in the next round of dynamic inference engine. This forms a complete closed loop from input parsing to solution pre-configuration to perturbation response and finally back to preference learning. All modules share data and model resources to ensure that configuration optimization response is compliant and continuously evolves.

[0225] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the investment strategy configuration method based on form semantic parsing and dynamic reasoning in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0226] This application also provides an investment strategy configuration device based on form semantic parsing and dynamic reasoning. Please refer to... Figure 11 The investment strategy configuration device based on form semantic parsing and dynamic reasoning includes:

[0227] The semantic parsing module 10 is used to obtain the structured form fields and investment task description, and to perform semantic parsing on the investment task description and structured form fields to obtain the structured task description.

[0228] The viewpoint generation module 20 is used to generate multi-source investment information based on the structured task description;

[0229] The scheme combination module 30 is used to generate target combination schemes under preset market scenarios based on structured task descriptions, multi-source investment information and historical market data. The target combination schemes include basic combination schemes and backup combination schemes.

[0230] The task re-optimization module 40 is used to determine the perturbation feature vector based on the target combination scheme and generate a re-optimization task description based on the perturbation feature vector.

[0231] The strategy configuration module 50 is used to optimize the candidate asset weight configuration scheme based on the re-optimization task description, the preset difference penalty coefficient and the weight vector of the current holdings, and to determine the investment strategy configuration scheme.

[0232] The investment strategy configuration device based on form semantic parsing and dynamic reasoning provided in this application, employing the investment strategy configuration method based on form semantic parsing and dynamic reasoning in the above embodiments, can solve the technical problem of difficulty in efficiently and accurately understanding and implementing investment decision intentions under complex and ever-changing market disturbances during portfolio management, thus leading to a lag in portfolio adjustment response. Compared with the prior art, the beneficial effects of the investment strategy configuration device based on form semantic parsing and dynamic reasoning provided in this application are the same as those of the investment strategy configuration method based on form semantic parsing and dynamic reasoning provided in the above embodiments, and other technical features in the investment strategy configuration device based on form semantic parsing and dynamic reasoning are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0233] This application provides an investment strategy configuration device based on form semantic parsing and dynamic reasoning. The device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enable the at least one processor to perform the investment strategy configuration method based on form semantic parsing and dynamic reasoning described in Embodiment 1. (Refer to the following...) Figure 12 It shows a structural schematic diagram of an investment strategy configuration device based on form semantic parsing and dynamic reasoning suitable for implementing embodiments of this application. Figure 12 The investment strategy configuration device based on form semantic parsing and dynamic reasoning shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application. Figure 12As shown, the investment strategy configuration device based on form semantic parsing and dynamic reasoning may include a processing unit 1001, which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the investment strategy configuration device based on form semantic parsing and dynamic reasoning. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the investment strategy configuration device based on form semantic parsing and dynamic reasoning to communicate wirelessly or wiredly with other devices to exchange data.

[0234] The investment strategy configuration device based on form semantic parsing and dynamic reasoning provided in this application, employing the investment strategy configuration method based on form semantic parsing and dynamic reasoning in the above embodiments, can solve the technical problem of difficulty in efficiently and accurately understanding and implementing investment decision intentions under complex and ever-changing market disturbances during portfolio management, thus leading to a lag in portfolio adjustment response. Compared with the prior art, the beneficial effects of the investment strategy configuration device based on form semantic parsing and dynamic reasoning provided in this application are the same as the beneficial effects of the investment strategy configuration method based on form semantic parsing and dynamic reasoning provided in the above embodiments, and other technical features in this investment strategy configuration device based on form semantic parsing and dynamic reasoning are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0235] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0236] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon. These computer-readable program instructions are used to execute the investment strategy configuration method based on form semantic parsing and dynamic reasoning in the above embodiments. The computer-readable storage medium carries one or more programs. When these programs are executed by an investment strategy configuration device based on form semantic parsing and dynamic reasoning, the device performs the following: acquires structured form fields and an investment task description; performs semantic parsing on the investment task description and the structured form fields to obtain a structured task description; generates multi-source investment information based on the structured task description; generates a target portfolio scheme under a preset market scenario based on the structured task description, the multi-source investment information, and historical market data, wherein the target portfolio scheme includes a basic portfolio scheme and a backup portfolio scheme; determines a perturbation feature vector based on the target portfolio scheme and generates a re-optimization task description based on the perturbation feature vector; and optimizes the candidate asset weight configuration scheme based on the re-optimization task description, a preset difference penalty coefficient, and the weight vector of the current holdings to determine an investment strategy configuration scheme. Computer program code for performing the operations of this application can be written in one or more programming languages, or a combination thereof, including object-oriented programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a LAN (Local Area Network) or a WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0237] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the specific unit. The readable storage medium provided in this application is a computer-readable storage medium storing computer-readable program instructions for executing the above-described investment strategy configuration method based on form semantic parsing and dynamic reasoning. This solves the technical problem of delayed portfolio adjustment response due to the difficulty in efficiently and accurately understanding and implementing investment decision intentions under complex and volatile market conditions during portfolio management. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the investment strategy configuration method based on form semantic parsing and dynamic reasoning provided in the above embodiments, and will not be repeated here. This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the investment strategy configuration method based on form semantic parsing and dynamic reasoning as described above. The computer program product provided in this application can solve the technical problem of delayed portfolio adjustment response due to the difficulty in efficiently and accurately understanding and implementing investment decision intentions under complex and volatile market conditions during portfolio management. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the investment strategy configuration method based on form semantic parsing and dynamic reasoning provided in the above embodiments, and will not be repeated here.

[0238] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for investment strategy configuration based on form semantic parsing and dynamic reasoning, characterized in that, The method comprises: obtaining a structured form field and an investment task description, and performing semantic analysis on the investment task description and the structured form field to obtain a structured task description; generating multi-source investment information according to the structured task description; generating a target portfolio scheme under a preset market scenario according to the structured task description, the multi-source investment information and historical market data, wherein the target portfolio scheme comprises a basic portfolio scheme and a backup portfolio scheme; determining a perturbation feature vector according to the target portfolio scheme, and generating a re-optimization task description according to the perturbation feature vector; optimizing a candidate asset weight configuration scheme according to the re-optimization task description, a preset difference penalty coefficient and a weight vector of a current holding to determine an investment strategy configuration scheme; the step of obtaining a structured form field and an investment task description, and performing semantic analysis on the investment task description and the structured form field to obtain a structured task description comprises: obtaining a structured form field, a task description and a perturbation processing preference, wherein the structured form field comprises an investment target, a constraint condition, an asset pool, a risk preference and a perturbation processing strategy field; generating an investment task description according to the task description and the perturbation processing preference; performing integrity check, type conversion and logical conflict detection on the structured form field and the investment task description to obtain standard financial information; inputting the standard financial information into a language processing model for joint semantic encoding and analysis, identifying financial entities, quantity constraints and logical relationships of the standard financial information, and mapping the financial entities, the quantity constraints and the logical relationships into structured semantic slots; performing vector retrieval in a preset financial knowledge base to determine regulatory rules, asset attributes and historical pattern information of the structured semantic slots, and performing knowledge completion on the structured semantic slots according to the regulatory rules, the asset attributes and the historical pattern information to obtain a structured task description; the step of performing vector retrieval in a preset financial knowledge base to determine regulatory rules, asset attributes and historical pattern information of the structured semantic slots, and performing knowledge completion on the structured semantic slots according to the regulatory rules, the asset attributes and the historical pattern information to obtain a structured task description comprises: converting the structured semantic slots into query vectors, and performing vector retrieval in a preset financial knowledge base according to the query vectors to obtain candidate knowledge items; determining a target knowledge item according to the semantic similarity of the candidate knowledge items and the query vectors and the timeliness weight of the candidate knowledge items; determining regulatory rules, asset attributes and historical pattern information according to the target knowledge item, and performing knowledge completion on the structured semantic slots according to the regulatory rules, the asset attributes and the historical pattern information to obtain completed semantic slots; performing structured check on the completed semantic slots to obtain a structured task description; The step of generating a target portfolio scheme under a preset market scenario according to the structured task description, the multi-source investment information, and historical market data comprises: taking the investment target and constraint in the structured task description as an optimization condition, and generating a basic portfolio scheme according to the optimization condition, the multi-source investment information, and historical market data; generating a backup portfolio scheme according to a preset disturbance type and the basic portfolio scheme, wherein the backup portfolio scheme comprises a disturbance type and a preset switching condition; performing historical backtesting and scenario simulation on the basic portfolio scheme and the backup portfolio scheme respectively under a preset market scenario to obtain a target evaluation index; generating a target portfolio scheme according to the basic portfolio scheme, the backup portfolio scheme, and the target evaluation index.

2. The method of claim 1, wherein, The step of generating multi-source investment information according to the structured task description comprises: inputting the structured task description into a language processing model to obtain expected return parameters and risk information through risk prediction; calculating asset expected return parameters and opinion uncertainty variance according to the expected return parameters and the risk information; performing weighted fusion according to the asset expected return parameters, the opinion uncertainty variance, and a preset confidence weight to obtain an opinion vector and a confidence matrix; generating multi-source investment information according to the opinion vector and the confidence matrix.

3. The method of claim 1, wherein, The step of determining a disturbance feature vector according to the target portfolio scheme, and generating a re-optimization task description according to the disturbance feature vector comprises: collecting market quotation data, transaction execution information, and external event streams in the running process of the target portfolio scheme; performing disturbance event identification and classification on the market quotation data, the transaction execution information, and the external event streams to obtain a disturbance event classification result; generating a disturbance feature vector according to the disturbance event classification result, wherein the disturbance feature vector comprises a disturbance type and a disturbance level; determining a target re-optimization mode according to the disturbance feature vector, a current holding state, and a historical evaluation result, and determining a target strategy according to the target re-optimization mode; generating a re-optimization task description according to the target re-optimization mode and the target strategy.

4. The method of claim 3, wherein, The step of determining a target re-optimization mode according to the disturbance feature vector, a current holding state, and a historical evaluation result, and determining a target strategy according to the target re-optimization mode comprises: matching the disturbance feature vector and a condition expression of a preset rule table to obtain a candidate re-optimization mode, wherein the candidate re-optimization mode comprises a local repair mode and an overall re-planning mode; calculating an execution cost score and a risk adaptation score in the candidate re-optimization mode according to a current holding state and a historical evaluation result; performing weighted synthesis according to the execution cost score and the risk adaptation score to obtain a target re-optimization mode; determining a target strategy according to the target re-optimization mode and a preset expert strategy library.

5. The method of claim 1, wherein, The step of optimizing a candidate asset weight configuration scheme according to the re-optimization task description, a preset difference degree penalty coefficient, and a weight vector of a current holding state to determine an investment strategy configuration scheme comprises: determine a first norm difference degree and a second norm difference degree between the candidate asset weight configuration scheme and the weight vector of the current position; determine a target difference degree according to a preset difference degree penalty coefficient, the first norm difference degree and the second norm difference degree; construct a target optimization problem according to a portfolio optimization objective function, the target difference degree and the re-optimization task description; solve the target optimization problem to obtain an investment strategy configuration scheme.

6. The method of claim 5, wherein, The step of determining the target difference degree according to the preset difference degree penalty coefficient, the first norm difference degree and the second norm difference degree comprises: determine a first norm weight coefficient and a second norm weight coefficient according to the preset difference degree penalty coefficient; weight process the first norm difference degree and the second norm difference degree respectively according to the first norm weight coefficient and the second norm weight coefficient to obtain a first weighted difference degree and a second weighted difference degree; determine the target difference degree according to the first weighted difference degree and the second weighted difference degree.

7. An investment strategy configuration device based on form semantic analysis and dynamic reasoning, characterized in that, The investment strategy configuration device based on form semantic analysis and dynamic reasoning executes the investment strategy configuration method based on form semantic analysis and dynamic reasoning in any one of claims 1 to 6, and the device comprises: a semantic analysis module configured to obtain a structured form field and an investment task description, and perform semantic analysis on the investment task description and the structured form field to obtain a structured task description; a viewpoint generation module configured to generate multi-source investment information according to the structured task description; a scheme combination module configured to generate a target combination scheme under a preset market scenario according to the structured task description, the multi-source investment information and historical market data, wherein the target combination scheme comprises a basic combination scheme and a backup combination scheme; a task re-optimization module configured to determine a perturbation feature vector according to the target combination scheme, and generate a re-optimization task description according to the perturbation feature vector; a strategy configuration module configured to optimize a candidate asset weight configuration scheme according to the re-optimization task description, a preset difference degree penalty coefficient and a weight vector of a current position, and determine an investment strategy configuration scheme.

Images