Strategy generation method and device, equipment, storage medium and product

Abstract

The embodiment of the invention discloses a strategy generation method and device, equipment, a storage medium and a product. The method comprises the steps of obtaining a current market state; determining whether the current market state triggers a strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; under the condition that the strategy mapping rule is not triggered, determining a target strategy suggestion corresponding to the current market state based on an enhanced generation mechanism; and according to a preset structured information output format, converting the target strategy suggestion to obtain a target strategy corresponding to the market state. According to the invention, the accuracy of strategy generation is improved.

Description

Technical Field

[0001] This application relates to the field of financial technology, and in particular to a strategy generation method, apparatus, device, storage medium, and product. Background Technology

[0002] In options trading and robo-advisory scenarios, it is necessary to determine operational suggestions for options based on market conditions, that is, to generate operational strategies for options, so that operators can execute corresponding operations on the options according to the operational strategies.

[0003] In related technologies, a strategy generation system can be used to generate trading strategies for options. This system establishes a correspondence between market conditions and strategies. When the system obtains market conditions information for a particular option, it can determine the corresponding trading strategy based on the correspondence between the market conditions and strategies, as well as the market conditions information for that particular option. However, since the market conditions in this strategy generation system are set based on preset static rules or decision tree structures, they cannot adapt to the dynamic changes in market portfolio states. Especially in scenarios where multiple conditions interact, the strategy mapping logic is prone to failure or coverage blind spots, thereby reducing the accuracy of strategy generation. Summary of the Invention

[0004] To address the aforementioned technical problems, embodiments of this application aim to provide a strategy generation method, apparatus, device, storage medium, and product that can improve the accuracy of strategy generation.

[0005] The technical solution of this application is implemented as follows: This application provides a strategy generation method, the strategy generation method including: Obtain the current market status; Determine whether the current market state triggers the strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; Without triggering the aforementioned strategy mapping rules, the target strategy recommendation corresponding to the current market state is determined based on the enhanced generation mechanism; The target strategy suggestion is transformed according to the preset structured information output format to obtain the target strategy corresponding to the market state.

[0006] This application provides a strategy generation apparatus, the apparatus comprising: The acquisition unit is used to acquire the current market status; A determining unit is used to determine whether the current market state triggers a strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; if the strategy mapping rule is not triggered, a target strategy suggestion corresponding to the current market state is determined based on an enhanced generation mechanism. The conversion unit is used to convert the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format.

[0007] This application provides an electronic device, the electronic device comprising: The system includes a memory, a processor, and a communication bus. The memory communicates with the processor via the communication bus. The memory stores a strategy generation program that the processor can execute. When the strategy generation program is executed, the processor performs the strategy generation method described above.

[0008] This application provides a storage medium storing a computer program applied to a strategy generation device, characterized in that the computer program, when executed by a processor, implements the strategy generation method described above.

[0009] This application also provides a computer program product, including a computer program that can be executed by a processor to complete the steps described in the aforementioned strategy generation method.

[0010] This application provides a strategy generation method, apparatus, device, storage medium, and product. The strategy generation method includes: acquiring the current market state; determining whether the current market state triggers a strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; if the strategy mapping rule is not triggered, determining a target strategy suggestion corresponding to the current market state based on an enhanced generation mechanism; and converting the target strategy suggestion into a target strategy corresponding to the market state according to a preset structured information output format. Using the above method, when the strategy generation apparatus determines that the current market state has not triggered a strategy mapping rule, it uses an enhanced generation mechanism to determine the target strategy suggestion corresponding to the current market state. Then, it converts the target strategy suggestion into a target strategy corresponding to the market state according to a preset structured information output format. This allows the target strategy to be determined based on the enhanced generation mechanism even when multiple conditions interact and cause the strategy mapping logic to fail or have a coverage blind spot, thereby improving the accuracy of strategy generation. Attached Figure Description

[0011] Figure 1 A flowchart of a strategy generation method provided in an embodiment of this application; Figure 2 A schematic diagram of an exemplary strategy generation architecture provided for embodiments of this application; Figure 3 A schematic diagram of the composition structure of a strategy generation device provided in this application embodiment. Figure 1 ; Figure 4A schematic diagram of the composition structure of a strategy generation device provided in this application embodiment. Figure 2 .

[0012] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0014] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0015] 1. Option: A financial derivative that gives the holder the right, but not the obligation, to buy or sell an underlying asset at a predetermined price within a specified period. Options are divided into call options and put options, which respectively give the buyer the right to buy or sell the underlying asset at the strike price before a future date.

[0016] 2. Underlying Asset: The underlying asset is the underlying asset referenced in derivative contracts such as options and futures. For option contracts, the underlying asset includes financial products such as stocks, stock indices, bonds, and commodities.

[0017] 3. Information Extraction: This refers to the process of identifying and extracting specific information elements (such as time, indicators, actions, and strategy suggestions) from natural language text. It can include entity recognition, relation extraction, and event detection, and is a key technology for transforming unstructured language into structured knowledge.

[0018] 4. Semantic Understanding: This refers to the model's ability to deeply understand the meaning of text, including the identification and reasoning of contextual logic, technical terms, grammatical structure, and implicit intentions. It is the foundational ability of large language models to construct conditional reasoning and policy interpretation paths.

[0019] 5. Knowledge Graph Construction: This refers to constructing a graph structure composed of nodes (concepts) and edges (relationships) by extracting entities and their semantic relationships, reflecting the logical network of domain knowledge. In this invention, it is used to express the rule system of "market state combination → strategy suggestion".

[0020] 6. Structured Data: Structured data refers to data that can be stored in tables, databases, or other formats with a fixed structure. Structured data can be quantitative data, such as stock prices, trading volumes, and financial statements. It has clear field definitions and a fixed format, making it easy to process and analyze.

[0021] 7. Unstructured Data: Unstructured data refers to data that lacks a fixed format, structure, or standard, such as text, images, audio, and video. Typical examples of unstructured data include news articles, social media content, emails, reports, and analytical articles. Unstructured data requires data cleaning and processing to extract useful information.

[0022] 8. Conditional Reasoning: This refers to the process of automatically drawing conclusions based on rules or models under certain logical conditions. In this system, it is represented by the logical mapping process of "If the current market state meets condition X, then strategy Y is recommended".

[0023] 9. Mapping Rule: This refers to the set of rules that establish a correspondence between a specific combination of market states (such as an index rising + implied volatility falling + positive term structure) and corresponding strategy recommendations (such as selling put options). It is the core logical unit for the system to achieve automated strategy triggering.

[0024] 10. Retrieval-Augmented Generation (RAG): This is an artificial intelligence technique that combines information retrieval and generative models. In RAG, the model first finds relevant information from the database through a retrieval phase, and then generates the final output based on this. This technique effectively combines known information with generative capabilities, improving the accuracy and diversity of answers, and performs particularly well in complex tasks.

[0025] 11. Large Language Model (LLM): This refers to a deep learning-based Natural Language Processing (NLP) model that learns language rules and knowledge by processing large-scale text data, possessing powerful text generation and understanding capabilities. Common LLM models include OpenAI's GPT series models, which can perform various tasks such as text generation, question answering, and text translation. Through extensive training data and parameters, LLM models can effectively understand and generate the complex structure and semantics of natural language.

[0026] This application provides a strategy generation method, which is applied to a strategy generation device. Figure 1 A flowchart of a strategy generation method provided in this application embodiment is shown below. Figure 1 As shown, the strategy generation method may include: S101. Obtain the current market status.

[0027] The strategy generation method provided in this application is applicable to scenarios where the target strategy corresponding to the current market state is determined.

[0028] In the embodiments of this application, the strategy generation device can be implemented in various forms. For example, the strategy generation device described in this application may include devices such as mobile phones, cameras, tablet computers, laptops, handheld computers, personal digital assistants (PDAs), portable media players (PMPs), navigation devices, wearable devices, smart bracelets, pedometers, etc., as well as devices such as digital TVs, desktop computers, servers, etc.

[0029] In this embodiment, the current market status can be character information, numerical information, or other forms of information. The specific current market status can be determined according to the actual situation, and this embodiment does not limit it.

[0030] For example, the current market state is represented by a vector.

[0031] In this embodiment, the current market state refers to a multi-dimensional state vector constructed based on real-time market data and relevant financial indicators, used to describe the overall environment of the current market. Specifically, the current market state vector can contain factors of multiple dimensions, such as a structured combination of market information composed of multi-dimensional market indicators including the underlying index trend (i.e., the development trend of the underlying index), implied volatility trend (i.e., volatility trend), volatility term structure (i.e., the slope of the implied volatility term structure), and PCR (i.e., market sentiment indicator).

[0032] It should be noted that each indicator factor in the current market state (i.e., any one of the index trend, volatility trend, implied volatility term structure slope, and market sentiment indicator) is discretized and encoded as an enumeration value, thus forming a unified state representation. For example, the underlying index trend may be divided into three categories: UP, DOWN, and NEUTRAL; the implied volatility trend may be divided into three categories: UP, DOWN, and FLAT. In this way, complex market information can be transformed into a computer-recognizable form, facilitating subsequent logical judgments and rule matching.

[0033] In this application embodiment, the current market status can be obtained from other devices, from user input information, from a database, or through other means. The specific method of obtaining the current market status can be determined according to the actual situation, and this application embodiment does not limit it. For example, in practical implementation, the process of constructing the current market state can rely on real-time market data interfaces and quantitative analysis models. The latest market data can be obtained from exchanges, third-party data service providers, etc., and key indicators can be standardized using preset processing methods. For instance, the determination of implied volatility trends might be based on a data window of the past N days, using moving averages, slope analysis, etc., to determine whether the implied volatility trend is rising, falling, or stable. This method allows for dynamic capture of market changes, providing accurate basic input for subsequent strategy generation.

[0034] In this embodiment, before acquiring the current market state, the strategy generation device also establishes strategy mapping rules based on historical financial investment information. Specifically, the process of establishing strategy mapping rules based on historical financial investment information includes: acquiring the historical financial investment information; extracting historical market states and corresponding historical strategies from the historical financial investment information; establishing a knowledge graph based on the correspondence between the historical market states and the historical strategies; and using the knowledge graph as the strategy mapping rules.

[0035] It should be noted that historical financial investment information includes investment advisory reports, expert opinions, and other information. Investment advisory reports are research reports provided by professional investment advisors in the past.

[0036] In this embodiment, historical financial investment information can be obtained from user input, other devices, or other means. The specific method of obtaining historical financial investment information can be determined according to the actual situation, and this embodiment does not limit it.

[0037] In this embodiment of the application, the historical financial investment information includes historical market conditions and historical strategies corresponding to those historical market conditions.

[0038] In this embodiment, historical financial investment information refers to a collection of records of an investor's or a financial market's investment behavior and results over a past period. Historical financial investment information includes, but is not limited to, transaction time, underlying asset type (such as options or futures), transaction direction (buy / sell), strike price, expiration date, trading volume, open interest, profit and loss, etc. By acquiring and organizing historical financial investment information, it is possible to understand the strategies adopted under similar market conditions and the actual effects of these strategies. For example, if an investor chooses to buy a straddle and profits when market volatility rises, the relevant historical records contained in the historical financial investment information can serve as an important reference in the future strategy generation process.

[0039] Understandably, by acquiring historical financial investment information, it is possible to build a foundation for strategy recommendations supported by real market experience, avoiding complete reliance on theoretical models or human experience, thereby improving the practicality and verifiability of strategy recommendations.

[0040] In this embodiment, historical market state refers to the characteristics of the market environment at a given time, identified from historical financial investment information. These typically include index trends, implied volatility trends, volatility term structure, PCR change rate, and market sentiment indicators. State variables in historical market state describe the market situation under specific circumstances. Historical strategy refers to investment strategies adopted in response to historical market states, such as buying call options, selling put options, or constructing butterfly spreads. Through natural language processing, dependency parsing, and template matching, the strategy expressions in unstructured text are parsed into standardized forms and bound to the corresponding market states.

[0041] Understandably, by extracting the correspondence between historical market states and historical strategies, we can establish a connection between strategy logic and market background, providing clear data support for subsequent knowledge modeling and rule reasoning.

[0042] In this embodiment, the knowledge graph is a graph-structured database that represents entities and their relationships using nodes and edges. Each node represents a market state combination or strategy recommendation, and edges indicate a causal or applicability relationship between them. For example, an index rising and implied volatility increasing is a market state node, connected to the strategy node corresponding to the strategy recommendation of buying call options. The establishment of the knowledge graph not only enables the visual expression of strategy logic but also facilitates subsequent dynamic updates, version control, and path retrieval. By introducing the inductive power of Large Language Models (LLM), synonymous expressions can be normalized, reducing rule redundancy.

[0043] Understandably, by constructing a knowledge graph, the structured expression and efficient management of strategy logic are achieved, the interpretability and maintainability of strategy recommendations are improved, and the construction of the knowledge graph provides a solid foundation for subsequent intelligent matching.

[0044] In this embodiment, the strategy mapping rule is the core basis for determining whether a certain strategy should be triggered in the current market state. In this application, the strategy mapping rule no longer relies on static presets, but is dynamically constructed from nodes and edges in the knowledge graph. After the real-time market state vector is constructed, it can be matched with the rule nodes in the knowledge graph, a matching score can be calculated, and the strategy suggestion with the highest confidence level can be output. Furthermore, the rule weights can be dynamically adjusted based on factors such as historical backtesting win rate and hit frequency to ensure reasonable recommendation priority.

[0045] In this embodiment, by introducing a knowledge graph and applying it to the policy mapping rules, the flexibility and adaptability of the policy generation device are improved. In a complex and ever-changing market environment, this policy generation device can quickly respond to changes in demand and achieve policy recommendation functions with high coverage and high accuracy.

[0046] In this embodiment, by acquiring historical financial investment information and extracting the correspondence between historical market states and historical strategies, a knowledge graph is constructed. The knowledge graph is then used as a strategy mapping rule, which can effectively integrate historical market experience with the current market situation. This can improve the accuracy and interpretability of strategy recommendations, thereby enabling intelligent and adaptive option strategy generation and optimization.

[0047] In this embodiment of the application, the process by which the strategy generation device extracts historical market states and corresponding historical strategies from the historical financial investment information includes: inputting the historical financial investment information into an information extraction model to obtain initial historical market states and initial historical strategies; and parsing the correlation between the initial historical market states and the initial historical strategies to obtain the historical market states and the historical strategies.

[0048] In this embodiment, the information extraction model can be a model configured in the strategy generation device, a model transmitted to the strategy generation device from other devices, or a model obtained by the strategy generation device through other means. The specific way in which the strategy generation device obtains the information extraction model can be determined according to the actual situation, and this embodiment does not limit it.

[0049] In this embodiment, the information extraction model can be a BERT-CRF named entity recognition (NER) model, or a large model. The information extraction model can also be other models that can first identify the initial historical market state and initial historical strategy from historical financial investment information. The specific information extraction model can be determined according to the actual situation, and this embodiment does not limit it.

[0050] In this embodiment, the information extraction model refers to a machine learning or deep learning model used to identify and extract key market states and strategy expressions from unstructured text (such as investment advisory reports and strategy recommendations). The information extraction model can be based on natural language processing techniques (such as BERT-CRF, dependency parsing, etc.) and can automatically identify market condition descriptions (such as index oscillations, implied volatility declines) and corresponding strategy recommendations (such as selling straddles). Through semantic parsing, entity recognition, and logical structuring, the information extraction model can output preliminary market state variables and strategy expressions, providing basic data for subsequent processing.

[0051] Understandably, the use of information extraction models improves the automation and accuracy of information extraction, avoiding the inefficiency and subjective biases caused by manual intervention. Furthermore, due to the adoption of domain-adaptive fine-tuning mechanisms (such as NER for financial terminology and instruction fine-tuning of the large language model), the information extraction model possesses a strong understanding of the professional expressions used in the options market, ensuring the accuracy and operability of the extracted information.

[0052] In this embodiment, resolving the correlation refers to the process of establishing a logical mapping between the initially extracted market state and strategy recommendations. The process of resolving the correlation involves semantic reasoning, conditional judgment structure identification, and rule induction, with the aim of clarifying the causal logic between the market state as a precondition and the strategy recommendations. For example, from the original text "If the index fluctuates and implied volatility declines, it is recommended to sell straddles," two market state variables (index trend and implied volatility trend) and the combination logic of these two market state variables are identified, and this combination logic is mapped to the corresponding strategy recommendations.

[0053] Understandably, by analyzing the correlation between the initial historical market state and the initial historical strategy, suitable strategy recommendations can be quickly matched based on the current market state. Furthermore, through a logical verification mechanism, the intelligent matching engine can detect and correct inconsistent or contradictory mapping relationships, thereby improving the accuracy and reliability of strategy recommendations.

[0054] In this embodiment, historical market states and historical strategies are integrated into a knowledge graph for use by the subsequent intelligent matching engine. This not only enhances the interpretability of the strategy generation device but also supports version management, failure tracking, and dynamic updates of the strategy logic, improving the long-term adaptability and maintenance efficiency of the strategy generation device.

[0055] In this embodiment, by inputting historical financial investment information into the information extraction model and further analyzing the relationship between the initial market state and the strategy, structured strategy rules are automatically constructed from unstructured text. This can improve the reusability of strategy knowledge and the level of intelligent management, thereby enhancing the responsiveness of the strategy generation device to complex market situations and enabling high-coverage and high-accuracy strategy recommendation services.

[0056] In this embodiment of the application, the initial historical market state and the initial historical strategy can be parsed into a structural relationship of "if the initial market state, then the initial strategy", thereby obtaining the historical market state and the historical strategy.

[0057] For example, based on financial terminology NER, dependency parsing and instruction fine-tuning large language model, it can automatically identify market state descriptions (i.e. initial historical market state) and corresponding strategy recommendations (i.e. initial historical strategies) from research report text (i.e. investment advisory research reports), and parse the logical structure (such as "if X then Y") to form executable rule statements.

[0058] In this embodiment of the application, the process of the strategy generation device establishing a knowledge graph based on the correspondence between the historical market state and the historical strategy includes: adjusting the historical strategy using a preset strategy template format to obtain an adjusted historical strategy; and establishing a knowledge graph based on the correspondence between the historical market state and the adjusted historical strategy.

[0059] In this embodiment, the preset policy template format can be a policy template format transmitted by other devices, a policy template format configured in the policy generation device, or a template format obtained by the policy generation device through other means. The specific way in which the policy generation device obtains the preset policy template format can be determined according to the actual situation, and this embodiment does not limit it.

[0060] It should be noted that the preset strategy template format refers to a standardized, structured expression designed to unify historical strategies from different sources and with different expressions. This template format typically includes key parameters such as strategy type (e.g., buy call options, sell straddles), strike price, expiration date, and execution direction, and transforms unstructured natural language strategy descriptions into computable and comparable strategy structures.

[0061] It should be noted that the adjusted historical strategy refers to a structured strategy representation based on a preset strategy template format, after semantic normalization, grammatical correction, and logical completion of the original historical strategy content. Adjusting historical strategies eliminates synonym differences, fills in missing conditions, and ensures logical consistency, thereby improving the accuracy of subsequent knowledge graph construction and reasoning. Adjusting historical strategies using a preset strategy template format effectively solves the problems of inconsistent strategy expressions and fragmented information, giving strategies a unified structure and semantic foundation. This facilitates efficient matching, reasoning, and updating by the strategy generation device, thereby improving the accuracy and stability of the knowledge graph.

[0062] It's important to note that a knowledge graph is a graph-structured data model that represents entities and their relationships in the form of nodes and edges. Knowledge graphs can be used to store and manage the mapping between market states and strategy recommendations. Specifically, the market state is one node, and the strategy recommendation is another. The knowledge graph uses edges to represent the causal or relational relationships between the market state and the strategy recommendation. For example, a market state node (index rising and IV falling) is connected to a strategy node (sell straddle) via an edge. The edge's attributes might include metadata such as hit count, average win rate, and applicable boundaries.

[0063] The process of building a knowledge graph includes the following aspects: First, bind each adjusted historical strategy to the corresponding market state vector; Secondly, the binding relationship between each adjusted historical strategy and the corresponding market state vector is written into a graph database (such as Neo4j, JanusGraph, etc.) and a strategy rule triple (market state - adjusted historical strategy - confidence value).

[0064] Finally, the knowledge graph is displayed through visualization tools, allowing users to view path analysis, version evolution records, and anomaly warnings.

[0065] Understandably, by constructing a knowledge graph from adjusted historical strategies and market states, the strategy logic can be visualized, traceable, and dynamically maintained, enhancing the interpretability and flexibility of the strategy generation device. Simultaneously, it provides reliable data support for subsequent intelligent matching and reasoning. Standardizing historical strategies ensures a consistent semantic structure, thus providing a clear data foundation for knowledge graph construction. The knowledge graph creation process further clarifies and structures the relationship between strategies and market states, facilitating multi-dimensional analysis, reasoning, and optimization within the system.

[0066] In this embodiment, the preset strategy template format can be a clearly structured, standardized, and easily identifiable and reusable strategy template format to support graph modeling, strategy comparison, reasoning, and maintenance. For example, a unified strategy expression format (strategy name + type + composition). The preset strategy template format can also be other formats; the specific preset strategy template format can be determined according to actual circumstances, and this embodiment does not limit this.

[0067] In this embodiment, after obtaining the adjusted historical strategy, a knowledge graph can be established based on the historical market state, the adjusted historical strategy, and the correspondence between the historical market state and the adjusted historical strategy. Nodes in the knowledge graph represent the historical market state and the adjusted historical strategy, and the correspondence between the historical market state and the adjusted historical strategy is represented as the connections between nodes (edges). Version marking can be implemented by setting version numbers or timestamps on nodes and edges, and historical versions are retained in the graph database for traceability and comparison, thereby achieving strategy logic visualization, version control, and maintainability.

[0068] In this embodiment, the strategy generation device includes a financial semantic recognition unit. The financial semantic recognition unit is used to recognize financial terms and strategy expressions related to market conditions in text (i.e., historical financial investment information). The specific implementation method includes: using a Named Entity Recognition (NER) model based on BERT-CRF to label the input text at the word level and identify entities such as the underlying index, implied volatility (IV), volatility term structure, and market sentiment indicators (such as PCR). Alternatively, the recognition results are classified into standard variable names and value labels, and uniformly expressed as structured tuples, as shown in formula (1): (1) In this embodiment, the strategy generation device further includes a dependency parsing unit. The dependency parsing unit is used to analyze the implicit logical structure in the strategy text, including the semantic mapping relationship between condition judgments and strategy suggestions. Its technical implementation includes: performing dependency parsing on the input statement, identifying subject-predicate structure, conjunctions and modifiers, and constructing a grammatical dependency graph; based on the strategy logic template matching method, parsing the sentence structure into "condition clause (IF)" and "result clause (THEN)" to form a logical mapping chain; the condition logic combination is shown in formula (2); the strategy suggestion is shown in formula (3); (2) (3) Thus, the structured rule triples are constructed as shown in formula (4): (4) In this embodiment, the strategy generation device further includes a strategy semantic normalization and template mapping unit. To eliminate rule redundancy and conflicts caused by different expression methods, the strategy semantic normalization and template mapping unit introduces a unified strategy semantic model, including: establishing a lexical normalization mapping table to standardize synonymous expressions, such as classifying "IV increase", "volatility increase", and "implicit volatility surge" as IV_TREND = UP; and introducing a strategy behavior classification system to classify strategy suggestions into standard operation templates, such as:

[0069] In this embodiment, the strategy generation device further includes a rule modeling and knowledge graph construction unit. The rule modeling and knowledge graph construction unit is used to structurally represent the extracted rules and construct a strategy knowledge graph to facilitate subsequent reasoning and version management. The implementation includes: identifying each rule as a triplet as described in formula (5): (5) It should be noted that the above consists of one or more state variables, supporting logical expression, as shown in formula (6): (6) It should also be noted that the rule set is written into the graph database to construct a knowledge graph based on "duration status nodes", "policy nodes", and "logical connection edges" as basic units, which is used for rule visualization, dynamic activation, version tracking and failure management.

[0070] In this embodiment, the strategy generation device further includes a large-scale model semantic induction and rule generation unit. When rule extraction cannot be completed through template parsing, a finely tuned large language model in the financial domain (such as FinGPT or ChatGLM-Finance) can be invoked for semantic induction generation. Specific methods include: using "market state description + historical strategy corpus" as input, designing a few-sample prompt for suggestive learning; outputting structured JSON-formatted mapping rule fragments from the large model and inputting them into the knowledge graph; and verifying the generated content through a semantic consistency check module to ensure that the generated strategy has correct logical relationships and domain compliance.

[0071] In this embodiment, an automated modeling process for extracting "market state combinations to strategy recommendations" from unstructured text is realized through a financial semantic recognition unit, a dependency parsing unit, a strategy semantic normalization and template mapping unit, a rule modeling and knowledge graph construction unit, and a large model semantic induction and rule generation unit. The constructed rules can be used in the mapping inference engine and strategy generation module, which significantly improves the structured expression capability, coverage breadth, and maintainability of the strategy generation device of option strategy logic.

[0072] S102. Determine whether the current market state triggers the strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information.

[0073] In this embodiment of the application, after the strategy generation device obtains the current market state, it can determine whether the current market state triggers the strategy mapping rule.

[0074] In this embodiment, a strategy mapping rule is a predefined logical relationship between a combination of market conditions (i.e., historical market states) and historical strategies. These strategy mapping rules are derived from semantic parsing and structured modeling of a large amount of historical investment advisory reports, strategy documents, and transaction records. Strategy mapping rules can be expressed in any form, such as (market state combination, strategy suggestion), and support logical combination expressions, such as if X (first market state) and Y (second market state), then Z (strategy suggestion). These strategy mapping rules can be stored in a knowledge graph, where nodes represent market state combinations and strategy suggestions, and edges represent the causal relationship between them. Upon receiving the current market state, a matching engine is invoked to compare the state vector of the current market state with the strategy mapping rules in the knowledge graph to determine if a matching strategy mapping rule exists.

[0075] In this embodiment, the matching process can employ a combination of fuzzy logic judgment and weighted scoring mechanisms to ensure that even if the current market state does not perfectly match the precise conditions of a strategy mapping rule, a reasonable recommendation can still be given based on similarity. For example, if a strategy mapping rule requires both an increase in the index and an increase in implied volatility, and the current market state shows a slight increase in the index and a slight increase in implied volatility, the strategy mapping rule may still be considered to have a certain degree of matching and will be included in the recommendation candidate list. Furthermore, strategy mapping rules can be weighted and ranked based on factors such as historical backtesting win rate and hit frequency to improve recommendation quality.

[0076] In this embodiment of the application, the process by which the strategy generation device determines whether the current market state triggers a strategy mapping rule includes: establishing a multi-dimensional state vector based on the current market state; searching for multiple rule nodes in the strategy mapping rule that match the multi-dimensional state vector; determining multiple operation strategies corresponding to the multiple rule nodes, and multiple revenue information of the multiple operation strategies; determining multiple confidence levels corresponding to the multiple operation strategies based on the multiple revenue information; and determining whether the current market state triggers a strategy mapping rule based on the multiple confidence levels.

[0077] In this embodiment, the multidimensional state vector is a structured representation describing the current market state. The multidimensional state vector consists of multiple key market factors, such as the underlying index trend (upward / downward / oscillating), the direction of implied volatility (IV) change (upward / downward / stable), the slope of the implied volatility term structure (normal / inverted), and market sentiment indicators (such as the rate of change in the PCR ratio). Each key factor in the market is discretized and encoded as an enumerated value, and combined into a unified state vector to characterize the overall features of the market state under the current circumstances.

[0078] In this embodiment of the application, parameters such as the trend of the underlying index, volatility trend, and implied volatility term structure slope can be obtained from the current market state. Based on the underlying index trend, volatility trend, and implied volatility term structure slope, etc., a multi-dimensional state vector can be established.

[0079] In this embodiment, the multidimensional state vector can not only capture changes in a single market variable, but also reflect the coupling relationship between multiple market factors, thus providing a more comprehensive characterization of the complex current market situation. By converting nonlinear, dynamic market signals into standardized inputs, rapid identification and adaptation of strategy logic under different market scenarios can be achieved.

[0080] In this embodiment, by constructing a multi-dimensional state vector, complex market information can be abstracted into computable and comparable structured data. This structured data format provides a precise basis for subsequent rule matching and strategy generation. Ultimately, this data processing method helps improve the coverage and accuracy of strategy recommendations.

[0081] In this embodiment of the application, the strategy mapping rule can be a knowledge graph, which includes multiple nodes and edges connecting the multiple nodes. The knowledge graph includes historical market states, historical strategies corresponding to the historical market states, and the correspondence between the historical market states and the historical strategies. Multiple rule nodes that match the multidimensional state vector can be found in the knowledge graph, and the historical strategies corresponding to the multiple rule nodes are multiple operation strategies.

[0082] In this embodiment, the strategy mapping rules are logical relationships between market condition combinations and strategy recommendations stored in the form of a knowledge graph. Each rule node corresponds to a specific market state combination and the associated strategy recommendation. Based on the current market state vector, multiple rule nodes matching the current market state vector can be retrieved from the knowledge graph. These rule nodes may contain exact matches or fuzzy matches of condition combinations.

[0083] It should be noted that the fuzzy matching mechanism allows for matching of some market factors even when they are at boundary values ​​or in a state of uncertainty, thereby improving fault tolerance and adaptability. Furthermore, each rule node includes additional attributes such as historical hit frequency, win rate, and execution effect, which are used to evaluate the credibility and applicability of the rule node.

[0084] In this embodiment of the application, by searching for rule nodes in the knowledge graph that match the current market state vector, it is possible to quickly locate strategy suggestions applicable to the current market situation, and optimize the matching results by combining historical performance data, thereby improving the reliability and intelligence of strategy recommendations.

[0085] In this embodiment, multiple return information for multiple operational strategies can be obtained from historical financial investment information. Alternatively, historical data can be retrieved to obtain the multiple return information for multiple operational strategies. Other methods can also be used to determine the multiple return information for multiple operational strategies; the specific method for determining the multiple return information for multiple operational strategies can be determined according to the actual situation, and this embodiment does not limit this method.

[0086] For example, for each successfully matched rule node, the corresponding strategy suggestion can be extracted and converted into an executable operational strategy, such as buying call options, selling straddles, or constructing bull spreads. Simultaneously, a quantitative engine can be invoked to simulate the performance of the operational strategy under different scenarios based on current market data, outputting risk-return indicators such as the expected return distribution, maximum drawdown, and Sharpe ratio for each operational strategy.

[0087] In this embodiment of the application, the synchronous generation of operational strategies and revenue information not only provides strategy suggestions, but also provides the potential returns and risks of the operational strategies in actual execution, making it easier for users to make more reasonable decisions based on these returns and risks.

[0088] In this embodiment, by extracting the strategies corresponding to the matching rule nodes and simulating the benefit performance of the strategies corresponding to the matching rule nodes, the linkage between strategy suggestions and actual execution effects can be realized, thereby improving the practicality and guidance value of strategy recommendations.

[0089] In this embodiment, confidence level is a comprehensive score that measures the rationality and priority of recommending a certain operational strategy under the current market conditions. It can be determined by weighting factors such as the historical win rate, risk-reward ratio, and matching strength of the operational strategy. Each operational strategy can be scored and sorted according to its confidence level to form a final recommendation list.

[0090] Among them, a high-confidence trading strategy means that the strategy has performed well in similar market environments in the past, with a high success rate and stability; while a low-confidence trading strategy may indicate that the strategy has not been fully validated, or that the current market conditions deviate significantly from the original design conditions of the strategy.

[0091] In the embodiments of this application, the matching degree between multiple operation strategies and multiple rule nodes can be determined to obtain multiple matching degrees; the weights of multiple operation strategies can be determined to obtain multiple weights; and multiple confidence levels can be determined based on multiple weights and multiple matching degrees.

[0092] It should be noted that the historical backtesting win rate, the hit frequency in the last 30 days, and the time decay factor of the latest trigger can be determined for each of the multiple operational strategies. The weight of each operational strategy is determined based on the historical backtesting win rate, the hit frequency in the last 30 days, and the time decay factor of the latest trigger, thus obtaining multiple weights.

[0093] In this embodiment of the application, by introducing a confidence score mechanism, the recommendation value of different operation strategies can be effectively distinguished, which can guide users to give priority to operation strategies with high confidence, thereby improving the scientific nature and decision-making efficiency of users in the strategy selection process.

[0094] In this embodiment, a state matching engine (combining fuzzy logic judgment and condition tree structure) can be used to retrieve rule nodes that match the current market state in the knowledge graph and output strategy suggestions and hit paths. Rule weights are dynamically adjusted based on factors such as historical strategy performance and matching frequency to ensure reasonable recommendation priorities.

[0095] In this embodiment, if the confidence level of at least one operational strategy exceeds a preset threshold, it can be determined that the current market state has triggered a strategy mapping rule, and a corresponding strategy suggestion can be output. If the confidence levels of all operational strategies are below the threshold, it indicates that the current market state has not found a suitable strategy mapping rule. In this case, the RAG enhancement generation module will be activated to call historical similar scenario data to complete the strategy suggestion. This mechanism ensures that the strategy recommendation process has both strict screening criteria and flexible supplementation capabilities, thereby preventing the situation where the current market state does not match the existing rules, and thus avoiding the problem of not being able to output a strategy. By determining whether to trigger a strategy mapping rule through a confidence level judgment mechanism, it is possible to balance responsiveness and coverage while ensuring strategy quality, thereby achieving intelligent and adaptive strategy generation and recommendation.

[0096] Understandably, this application establishes a multi-dimensional state vector, matches rule nodes, generates operational strategies and corresponding profit information for these strategies, calculates confidence levels, and determines whether to trigger strategy mapping rules based on the calculated confidence levels. In this way, accurate modeling and strategy adaptation of the current market state can be achieved, thereby providing interpretable and practical strategy recommendations, and ultimately enhancing the professionalism and intelligence of options investment advisory services.

[0097] In this embodiment of the application, the process of the strategy generation device establishing a multi-dimensional state vector based on the current market state includes: determining the index development trend, volatility trend, implied volatility term structure slope, and market sentiment index of the target based on the current market state; and establishing the multi-dimensional dynamic vector based on the index development trend, volatility trend, implied volatility term structure slope, and market sentiment index.

[0098] It should be noted that the index trend is a judgment on the future price movement of the underlying asset (such as a stock index) in the current market, which can be predicted through technical analysis, historical data backtesting, or machine learning models. The index trend can be divided into three basic states: upward, downward, or sideways. For example, if the five-day moving average is higher than the twenty-day moving average, then the situation is considered an upward trend.

[0099] It's important to note that volatility trends reflect market expectations of future uncertainty, and these trends can be measured by changes in implied volatility (IV) of options. Volatility trends can be upward, downward, or flat. Rising volatility indicates increased market uncertainty, leading investors to favor protective option strategies; conversely, declining volatility may suggest a more rational market, making it a good time for investors to sell volatility-based strategies.

[0100] It's important to note that the implied volatility term structure slope refers to the relationship between implied volatility at different expiration dates. For example, the implied volatility of short-term options is lower than that of long-term options, and this difference results in a positive slope for the implied volatility term structure. However, in extreme market conditions, an inverted slope may occur, where the implied volatility of short-term options is higher than that of long-term options. Changes in the implied volatility term structure slope can be used to analyze market sentiment and assess investors' risk appetite.

[0101] It should be noted that market sentiment indicators are composite variables built upon multiple factors (such as the PCR ratio, trading volume, open interest, and news sentiment) to measure the emotional tendencies of market participants. Market sentiment indicators can serve as an auxiliary judgment tool, helping to identify excessive optimism or pessimism in the market.

[0102] In this embodiment, the four dimensions of index development trend, volatility trend, implied volatility term structure slope, and market sentiment indicators together constitute a multi-dimensional characterization of the current market state, which helps to understand the market environment more comprehensively. The multi-dimensional characterization of the market state can improve the accuracy and adaptability of strategy generation.

[0103] In this embodiment, the multidimensional dynamic vector encodes four market state indicators—index trend, volatility trend, implied volatility term structure slope, and market sentiment indicator—into a unified mathematical representation, facilitating logical judgment and rule triggering by the subsequent strategy matching engine. The state values ​​of each dimension are discretized into a finite set (e.g., UP / DOWN / NEUTRAL), and these state values ​​are mapped to enumerated values ​​or integer codes. These enumerated values ​​or integer codes are then combined into an n-dimensional vector, which represents the overall state of the current market.

[0104] In the embodiments of this application, the multidimensional dynamic vector has high flexibility and scalability, and can dynamically adjust the number of dimensions as market factors increase or decrease. For example, when a new market indicator (such as a capital flow indicator) is introduced, the multidimensional dynamic vector can be seamlessly upgraded simply by incorporating the new market indicator into the vector system.

[0105] In this embodiment of the application, by abstracting complex market states into standardized vectors, it is possible to quickly match preset strategy rules under large-scale market data and support fuzzy judgment mechanism, so that strategy recommendation has higher coverage and real-time response capability.

[0106] In this embodiment, by constructing a multi-dimensional dynamic vector based on index trends, volatility trends, implied volatility term structure slopes, and market sentiment indicators, key characteristics of market conditions can be comprehensively captured. This improves the accuracy and generalization ability of strategy matching, thereby enhancing adaptability to complex market environments and ultimately providing users with more targeted and feasible option strategy recommendations.

[0107] In this embodiment of the application, the process by which the strategy generation device determines whether the current market state triggers a strategy mapping rule based on the plurality of confidence levels includes: determining that the current market state has not triggered a strategy mapping rule when each of the plurality of confidence levels is less than a first threshold; and determining that the current market state triggers a strategy mapping rule when each of the plurality of confidence levels is greater than or equal to the first threshold.

[0108] In this embodiment, the first threshold may be a threshold configured in the strategy generation device, a threshold transmitted to the strategy generation device from other devices, or a threshold obtained by the strategy generation device through other means. The specific way in which the strategy generation device obtains the first threshold can be determined according to the actual situation, and this embodiment does not limit it.

[0109] It should be noted that the value of the first threshold can be determined according to the actual situation, and this application embodiment does not limit it.

[0110] In this application embodiment, confidence level refers to a quantitative assessment of the degree of matching of a certain condition or state, which can be expressed in numerical form. In this application, confidence level is used to measure the degree of matching between the current market state and a certain preset strategy mapping rule. The lower the confidence level, the worse the adaptability of a certain preset strategy mapping rule to the current market state, and therefore it cannot be used as a valid basis for strategy suggestions. The first threshold is a preset numerical boundary. When all confidence levels are lower than the first threshold, it indicates that no strategy mapping rule can effectively match the current market state, and therefore the strategy mapping rule is not triggered.

[0111] In this embodiment of the application, by setting a first threshold and comparing confidence levels, it can be ensured that the strategy mapping rule is triggered only when there is a sufficient match between the current market state and multiple rule nodes in the strategy mapping rule, thereby avoiding the generation of low-quality or misjudged strategies and improving the accuracy and reliability in the strategy recommendation process.

[0112] In this embodiment, if multiple confidence levels are greater than or equal to a first threshold, it indicates that the current market state highly matches at least one strategy mapping rule, possessing high credibility to support the corresponding strategy suggestion output. Thus, the current market state is considered to meet the strategy mapping conditions, and the corresponding strategy execution process is triggered based on this judgment. The mechanism based on confidence threshold judgment helps to quickly identify highly matched market situations and respond promptly, thereby improving the real-time performance and intelligence level of the strategy generation device. In this embodiment of the application, by setting clear confidence threshold judgment conditions, suitable strategy mapping rules can be accurately identified in a variety of market state combinations, which can improve the coverage and response speed of strategy recommendations, and enhance its practical value and decision-making efficiency.

[0113] Understandably, by introducing a confidence level and a first threshold judgment mechanism, effective filtering and precise triggering of strategy mapping rules can be achieved when facing complex and ever-changing market environments. First, the confidence level of each strategy mapping rule is calculated based on the current market state. Then, the first threshold is used to determine whether the strategy mapping conditions are met, ultimately deciding whether to execute the corresponding strategy. The entire process achieves closed-loop control from data input to strategy output, improving intelligent decision-making capabilities and response efficiency.

[0114] In this embodiment, the steps in determining whether the current market state triggers a strategy mapping rule are logically closely related during actual implementation: First, current market state information needs to be obtained; next, the matching degree (i.e., confidence level) between each strategy mapping rule and the current market state is calculated; then, the comparison result between the confidence level and a first threshold is used to determine whether the strategy mapping rule is triggered; finally, if the triggering condition is met, the strategy execution process begins. This series of operations constitutes a complete strategy judgment and execution chain, ensuring that reasonable and efficient strategy responses can be made in different market environments.

[0115] In this embodiment of the application, after the strategy generation device establishes a multi-dimensional state vector based on the current market state, if no rule node matching the multi-dimensional state vector is found in the strategy mapping rules, it determines that the current market state has not triggered the strategy mapping rules.

[0116] In this embodiment, if no policy mapping rule is matched, it is directly determined that the policy has not been triggered, avoiding entering redundant processing or outputting invalid suggestions. This improves response efficiency and judgment accuracy, thereby reducing unnecessary computational resource consumption and ultimately enhancing the real-time performance and stability of the overall policy recommendation.

[0117] In this embodiment, after constructing the multidimensional state vector of the current market state, an attempt is made to match it within the existing strategy mapping rule base. If no rule node matching the state combination of the multidimensional state vector is found, it indicates that the current market state is not covered by existing rules, and no clear strategy suggestion can be generated. In the absence of a matching rule node, other complex judgment logic is not executed further; instead, it is directly determined as not triggered, i.e., no strategy suggestion is generated.

[0118] In this embodiment, in practical applications, if the current market state exhibits characteristics such as index oscillation, decreased implied volatility, and normal term structure, but the combination of these market state characteristics does not have a corresponding strategy mapping rule in the knowledge graph, then it is determined that the strategy mapping rule has not been triggered. This allows for robust operation even when the rule base is incomplete or in a cold start phase, preventing erroneous outputs due to missing rules. Furthermore, combined with a dynamic weight adjustment mechanism, once a similar market state appears in the future and is manually annotated with new rules, the strategy base can be automatically learned and updated, further optimizing future strategy recommendation capabilities.

[0119] In this embodiment, the strategy generation device includes a market state recognition and intelligent mapping layer. This layer constructs a multi-dimensional market state vector based on real-time market data and intelligently matches it with a pre-built strategy mapping rule base to automatically output the optimal strategy suggestion for the current market situation. This solves the problems of traditional strategy systems relying on static rules, difficulty in covering complex combined states, and lack of reasoning ability.

[0120] Specifically, the market state recognition and intelligent mapping layer includes a multi-factor market state vector construction module. This module transforms key market indicators into standardized state variables and constructs a unified multi-dimensional state representation. The market state is defined as a vector composed of n state variables, as shown in formula (7): (7) Each dimension Representing the discrete state of a market factor, for example: s1(t): Exponential trend (UP / DOWN / NEUTRAL) s2(t): Implied volatility trend (UP / DOWN / FLAT) s3(t): Volatility term structure slope (NORMAL / INVERTED) s4(t): PCR change rate (HIGH / LOW / STABLE) In this embodiment of the application, the state variable is calculated from market data, as shown in formula (8): (8) In this embodiment of the application, all state variables are discretized and uniformly encoded into enumerated values ​​to form the state code at the current moment.

[0121] In this embodiment, the market state recognition and intelligent mapping layer further includes a strategy rule matching and mapping reasoning module. The strategy rule matching and mapping reasoning module is used to perform matching judgments based on the current state vector and the strategy mapping rules in the knowledge graph, and output corresponding strategy suggestions. Specifically, all rules are modeled in the form of formula (9): (9) Where Ci is a combinational logic expression and Si is a strategy suggestion.

[0122] For example:

[0123] In this embodiment of the application, a matching score is calculated for each rule, as shown in formula (10): (10) in: The number of variables in the state vector that match this rule; The total number of conditions included in this rule; If the score meets the threshold (e.g.) ( ), which means it is considered a hit.

[0124] In this embodiment of the application, a fuzzy judgment mechanism can also be supported: for uncertain factors such as IV boundary values, fuzzy logic is introduced as shown in formula (11): (11) This makes strategy matching more fault-tolerant and flexible.

[0125] In this embodiment, the market state identification and intelligent mapping layer further includes a dynamic weight adjustment and priority ranking module. To improve the accuracy of strategy recommendations and the decision interpretation capability of strategy outputs, the system introduces a rule scoring and weighting mechanism into the dynamic weight adjustment and priority ranking module.

[0126] Each rule Related dynamic weights The factors in formula (12) determine the outcome: (12) in: Historical backtesting win rate; Hit frequency in the last 30 days; The latest time decay factor; It is an adjustable hyperparameter.

[0127] In this embodiment of the application, the matched strategy is based on Sort the suggestions and output the strategy recommendations with the highest confidence level.

[0128] In this embodiment, the market state recognition and intelligent mapping layer may further include an output explanation path generation module. The output explanation path generation module is used to ensure that the matching result can be traced back to a specific rule node. The system automatically generates a matching path description text, such as: "The current market state matches strategy rule R17: the index is rising and the implied volatility is rising and the term structure is inverted. The historical hit frequency is 11 times, the average win rate is 63.2%, and the recommended strategy is 'buy call options'." Through the mechanism of market state recognition and intelligent mapping layer, the system can accurately match the strategy logic that best fits the current state in real time in a dynamic market environment, which significantly improves the adaptability, real-time performance and interpretability of the strategy recommendation system, and provides a solid logical foundation for the subsequent historical completion and strategy generation modules.

[0129] S103. Without triggering the strategy mapping rule, determine the target strategy suggestion corresponding to the current market state based on the enhanced generation mechanism.

[0130] In this embodiment of the application, after the strategy generation device determines whether the current market state triggers the strategy mapping rule, if it determines that the strategy mapping rule has not been triggered, it determines the target strategy suggestion corresponding to the current market state based on the enhanced generation mechanism.

[0131] It should be noted that, even if the strategy mapping rules are not triggered and the rules are not directly matched, the system will search for similar historical scenarios and their corresponding strategies based on the current market state, and supplement the strategy suggestions through the enhanced generation mechanism. In essence, it is filling the strategy gaps that are not covered by existing rules.

[0132] In this embodiment, the enhanced generation mechanism is a technique combining historical market scenario retrieval and large language model reasoning. It is used to supplement target strategy recommendations when the current market state does not match existing strategy mapping rules or the output confidence of the strategy mapping rules is insufficient. Specifically, firstly, several historical cases most similar to the current market state are retrieved from the historical market evolution database. Then, these cases, along with the current market state, are input as prompts into the fine-tuned large language model. The large language model generates reasonable strategy recommendations based on these prompts. The generated strategy recommendations undergo semantic consistency verification and logical rationality checks to ensure they conform to the basic principles and risk control requirements of options trading.

[0133] In practice, the enhanced generation mechanism relies on efficient semantic retrieval algorithms (such as HNSW vector indexing) and high-performance natural language generation models (such as ChatGLM-Finance). For example, a record from May 2023 can be retrieved from a historical database. The market conditions in this record are very similar to the current market conditions. This record recommends selling a straddle. This strategy can be used as a reference input into the large language model to guide it in generating similar strategy suggestions. Furthermore, the generated results can be optimized by incorporating characteristic parameters of the current market conditions (such as strike price range and expiration date selection) to make the generated results more operational and adaptable.

[0134] In this embodiment of the application, the process by which the strategy generation device determines the target strategy suggestion corresponding to the current market state based on the enhanced generation mechanism includes: retrieving multiple historical sample cases similar to the current market state from the structured semantic scene library; inputting the multiple historical sample cases and the current market state into a large language model to obtain the target strategy suggestion.

[0135] In this embodiment, the structured semantic scenario library refers to a database that stores information such as historical market conditions, relevant strategy recommendations, and execution results in a unified format. The structured semantic scenario library not only includes quantitative indicators (such as volatility and index trends) but also integrates unstructured text information (such as research report summaries and investment advisor opinions), forming a multi-dimensional, searchable data set. Based on the structured semantic scenario library, historical scenarios similar to the current market condition can be quickly retrieved, providing a reference for generating target strategy recommendations. The establishment of the structured semantic scenario library may include multiple stages such as data cleaning, feature encoding, vector representation, and index optimization to ensure that the data has high semantic matching degree and retrieval efficiency.

[0136] In this embodiment, the structured semantic scenario library provides rich historical case support. Especially when the current market state cannot directly match existing rules, effective strategy suggestions can be extracted and reasoned to complete the analysis by searching similar scenarios in the library. Utilizing the structured semantic scenario library greatly enhances the generalization and knowledge transfer capabilities of the strategy generation device, avoiding situations where strategy output is blank or invalid due to incomplete rule coverage. The structured semantic scenario library can also organize data in a time-series manner to model market evolution paths. For example, within a specific time period, the market may experience a combination of rising exponential volatility, declining implied volatility, and an inverted term structure. This complex market scenario is often difficult to accurately describe with a single rule. By using the structured semantic scenario library, historical backgrounds with similar characteristics can be identified, and suitable option strategy combinations can be recommended based on the library's functionality, thereby improving the accuracy and practicality of strategy generation.

[0137] In this embodiment, semantic similarity calculation and vector retrieval techniques are used to select historical sample cases that highly match the current market state from a structured semantic scene library. Specifically, the current market state can first be transformed into a high-dimensional semantic embedding vector, and then combined with structured features (such as implied volatility trends and index movements) and unstructured features (such as research report descriptions and market sentiment) for joint matching. Subsequently, efficient vector retrieval algorithms such as HNSW (Hierarchical Navigable SmallWorld) or FAISS (Facebook AI Similarity Search) are used to quickly find several of the most similar historical records as reference information.

[0138] In this embodiment, when retrieving multiple historical sample cases similar to the current market state from the structured semantic scene library, multiple matching dimensions are comprehensively considered, including but not limited to: semantic similarity score, time distance, and market factor matching degree. These factors jointly determine the weight of each historical sample, thereby affecting the quality of the final strategy recommendation. For example, if a historical sample is highly consistent with the current market state in terms of implied volatility curve shape, volatility surface structure, etc., the strategy recommendation of that historical sample case will have higher reference value; conversely, if the market backgrounds of the two are significantly different, the relevance of the strategy recommendation of that historical sample case will be low, which will reduce the priority of the strategy recommendation of that historical sample case.

[0139] Understandably, by retrieving multiple historical sample cases similar to the current market state from a structured semantic scenario library, one can quickly identify historical scenarios comparable to the current situation when facing unknown or complex market states, and generate reasonable strategy recommendations based on these historical scenarios. This not only compensates for the shortcomings in coverage of existing technologies but also makes the strategy generation process more logical and explainable, thereby enhancing users' trust in the generated target strategy recommendations.

[0140] In this embodiment, there is a close relationship between the structured semantic scene library and historical sample cases. The structured semantic scene library is the foundation for retrieving historical samples, while historical sample cases are the key input for generating strategy suggestions. Therefore, only when the structured semantic scene library is rich in content and well-organized can the quality of subsequent retrieval and strategy generation be ensured.

[0141] In the embodiments of this application, the large language model can be ChatGLM-Finance or Qwen-Finance, or other models. The specific large language model can be determined according to the actual situation, and the embodiments of this application do not limit it.

[0142] In this embodiment, the large language model performs three functions: semantic understanding, logical reasoning, and strategy generation. First, the large language model performs deep semantic analysis on the current market state and historical samples to identify the similarities and differences between them. Second, the large language model derives strategy logic suitable for the current context based on these similarities and differences. Finally, the large language model outputs the generated target strategy suggestions in natural language, along with corresponding logical explanations and risk warnings.

[0143] In this embodiment, the large language model can be trained using instruction fine-tuning, thereby giving it strong adaptability to the financial field. For example, for the options market, the large language model is infused with a large amount of expertise regarding market state descriptions, strategy types, and the applicable conditions for these strategy types, ensuring high compliance and professionalism when generating strategies. Furthermore, to prevent deviations or illusions in the generated content, a logical consistency verification module can be set up to perform secondary verification of the strategy suggestions output by the large language model, ensuring that these strategy suggestions are logically consistent with the market states input by the large language model.

[0144] Understandably, large models enable intelligent transfer from historical experience to current decisions, allowing for the generation of high-quality policy recommendations even in the absence of explicit rules. Furthermore, because the generation process of target policy recommendations is highly interpretable, users not only understand what to do but also why to do it, thus enhancing the credibility and practical value of the policy generation device.

[0145] In this embodiment, by inputting historical sample cases and the current market state into a large language model, intelligent generation and dynamic updating of target strategy recommendations can be achieved. This can improve the accuracy and coverage of strategy recommendations, further enhancing the professionalism and real-time responsiveness of option strategy services.

[0146] In this embodiment of the application, before the strategy generation device retrieves multiple historical sample cases similar to the current market state from the structured semantic scenario library, it also obtains historical market states, historical strategy suggestions, and historical execution results corresponding to the historical strategy suggestions; and establishes the structured semantic scenario library based on the historical market states, the historical strategy suggestions, and the historical execution results.

[0147] It should be noted that historical market state refers to the specific environment and conditions the market was in over a past period, including but not limited to descriptions of key market factors such as the underlying index trend, implied volatility (IV) trend, volatility term structure slope, and PCR change rate. Market state can be represented in a discretized manner, such as an index rising, implied volatility falling, and a normal term structure. After quantifying and standardizing market state, a unified multidimensional state vector can be constructed for subsequent rule matching and similar scenario retrieval.

[0148] It should be noted that historical strategy recommendations include option trading strategy suggestions given by investment advisors or other experts under historical market conditions. These recommendations include specific option operation types (such as buying a call option or selling a straddle), strike price, expiration date, and may also include judgments on market expectations. Historical strategy recommendations are an important data source for the strategy generation device to learn and optimize its strategy recommendation logic.

[0149] It's important to note that historical execution results refer to the actual gains or losses achieved after executing historical strategy recommendations under historical market conditions. These results reflect the performance of historical strategy recommendations in the actual market and are a crucial basis for evaluating strategy effectiveness and adjusting rule weights. By recording the execution results of each strategy recommendation, backtesting analysis can be performed to identify high-win-rate, low-risk strategy patterns, and future strategy recommendations can be optimized based on the identified patterns.

[0150] Understandably, by acquiring historical market conditions, historical strategy suggestions, and historical execution results, a high-quality data foundation can be provided for the construction of a structured semantic scenario library, enhancing the strategy generation device's ability to understand historical market contexts, more accurately matching the current market condition with historical success cases, and improving the coverage and recommendation quality of strategy generation.

[0151] In this embodiment, the structured semantic scenario library is a database that transforms unstructured historical market information into a queryable and reasonable structured database. The structured semantic scenario library organizes historical market states, historical strategy recommendations, and the corresponding execution results of these recommendations into unified structured entries, facilitating rapid retrieval and generation of corresponding strategy recommendations in real-time market conditions. The structured semantic scenario library defines that each entry may include the following fields: Market state vector: a multidimensional state code composed of multiple discretized market factors.

[0152] Strategy suggestion template: Standardized strategy expression format, such as BUY_CALL_SPREAD; Performance metrics include profit / loss, Sharpe ratio, and maximum drawdown. Semantic context description: Unstructured text summary used to supplement the explanation of the market background and strategic thinking at the time.

[0153] In this embodiment, when constructing a structured semantic scene library, classification, clustering, and indexing can be performed based on fields such as market state vectors, strategy suggestion templates, execution result metrics, and semantic context descriptions to support efficient semantic retrieval and similarity comparison. For example, using HNSW or FAISS vector indexing technology, millisecond-level approximate nearest neighbor search can be achieved; simultaneously, by introducing a natural language processing model, the model can extract semantic features of strategy suggestions, thereby enhancing the relevance of the retrieval.

[0154] In this embodiment, by establishing a structured semantic scenario library based on historical market states, historical strategy suggestions, and historical execution results, deep modeling of historical market evolution scenarios can be achieved. By establishing this library, the strategy generation device's ability to generate strategies in unknown market conditions can be improved. This allows for the use of historical experience to fill in the gaps in the rule system, thereby significantly enhancing the intelligence and generalization capabilities of the strategy generation device.

[0155] In this embodiment, by acquiring historical market states, historical strategy recommendations, and historical execution results, and constructing a structured semantic scenario library, an intelligent decision support framework with memory capabilities can be formed. This intelligent decision support framework can not only accurately capture changes in market states, but also quickly generate reasonable target strategy recommendations based on successful experiences extracted from historical execution results, thereby significantly improving the adaptability and decision-making efficiency of the intelligent decision support framework in complex market environments.

[0156] In this embodiment, historical market states, strategy recommendations, and execution results can be organized in a time-series manner to form a searchable structured semantic scenario library. Market state vectors and context descriptions are used to jointly retrieve similar historical cases, employing the HNSW vector retrieval algorithm and a semantic matching scoring mechanism. The current market state plus historical similar strategies are used as input prompts to a large language model to generate strategy recommendation text, and a logical consistency judgment module ensures the accuracy and compliance of the content.

[0157] In this embodiment, the strategy generation device further includes a historical context retrieval and a RAG-enhanced generation layer. The historical context retrieval and RAG-enhanced generation layer are used to construct a semantic association channel with historical market evolution scenarios when the current market state does not match existing mapping rules or the strategy confidence is insufficient. Through semantic enhancement retrieval and generative model reasoning, it outputs logically sound and referable option strategy suggestions, thereby supplementing the blind spots in the rule system and realizing the generalization and knowledge transfer capabilities of intelligent strategy reasoning.

[0158] Specifically, the historical context retrieval and RAG enhancement generation layer includes a historical market evolution context database construction module. This module establishes a structured historical market data and strategy sample database to support the retrieval of similar scenarios in the current market state.

[0159] Each record in the sample data includes fields as shown in formula (13): (13) in: S i : Market state vectors at historical points in time (such as index trends, IV trends, term structures, etc.); M i Quantitative indicators (such as trading volume, open interest, PCR, etc.); C i Unstructured semantic context (such as research report summaries and investment advisor conclusions); Strategy i The result of the strategy recommended or implemented at that time (structured template expression).

[0160] It should be noted that the fields in the sample data come from sources including historical research reports, validated strategy samples, and system backtesting results. Through cleaning, vectorization, and label standardization, a unified retrieval format is constructed.

[0161] In this embodiment, the historical context retrieval and RAG enhancement generation layer includes a market state semantic vector construction module. To achieve semantic-level similarity scenario comparison, the market state semantic vector construction module transforms the current market state into a high-dimensional semantic embedding vector.

[0162] The structured coding process is shown in formula (14): (14) In this embodiment, the historical context retrieval and RAG-enhanced generation layer further includes a historical context semantic retrieval and similarity matching module. This module retrieves the k most similar historical samples to the current state from the historical scene database, providing a reference for the generation model.

[0163] The similarity measurement function (structured + semantic joint) is shown in formula (15): (15) in: cos: cosine similarity of semantic vectors; DTW: Dynamic Time Warping, used to compare the similarity of time series indicator curve shapes; Weighting parameter, adjustable.

[0164] It should be noted that the vector retrieval structure includes: using HNSW or FAISS vector indexes, supporting millisecond-level approximate nearest neighbor search. The output result is shown in formula (16): (16) In this embodiment, the historical context retrieval and RAG enhancement generation layer further includes a RAG enhancement strategy generation module. After acquiring similar historical scenarios, the RAG enhancement strategy generation module uses the Retrieval Enhancement Generation (RAG) structure to generate current strategy suggestions, thereby achieving reasoning completion.

[0165] The technical architecture of the RAG enhanced policy generation module includes input construction, generation model configuration, policy semantic parser, logical consistency verification module, and result confidence evaluation.

[0166] Specifically, the input construction refers to the Prompt content being composed of a summary of the current market state and a retrieved summary of historical strategies. Examples include: Current market characteristics: index fluctuating, implied volatility declining, and term structure normal.

[0167] Similar historical scenario (1): In July 2022, the market performance... The recommended strategy is to "sell straddles". Similar historical scenario (2): In May 2023, the market performance... The strategy was to "construct a put butterfly spread". Based on the above, please provide feasible strategy recommendations for the current market.

[0168] The generated model configuration uses a large language model (such as ChatGLM-Finance or Qwen-Finance) that is fine-tuned using instructions; the model output is a structured strategy suggestion plus a logical explanation.

[0169] The policy semantic parser transforms natural language output into structured policy structures (such as BUY_CALL_SPREAD).

[0170] The logic consistency verification module verifies the consistency between the generated results and the input market state logic (e.g., "IV decline" should not recommend "buy straddle") to prevent strategy illusion.

[0171] Result confidence assessment: Combining semantic similarity score and historical hit rate to assess output reliability, which is then used for system ranking and decision-making.

[0172] Historical context retrieval and RAG-enhanced generation layers enable the retrieval of similar experiences and the reconstruction of strategy logic in unknown market conditions, enhancing the system's responsiveness to complex and irregular scenarios. They are key supplementary modules for realizing intelligent and generalized strategy recommendation systems.

[0173] S104. The target strategy suggestion is transformed according to the preset structured information output format to obtain the target strategy corresponding to the market state.

[0174] In this embodiment, after the strategy generation device determines the target strategy suggestion corresponding to the current market state based on the enhanced generation mechanism, it transforms the target strategy suggestion into the target strategy corresponding to the market state according to the preset structured information output format.

[0175] In this embodiment, the preset structured information output format can be the format information configured in the strategy generation device, the format information transmitted to the strategy generation device by other devices, or the format information obtained by the strategy generation device through other means. The specific way in which the strategy generation device obtains the preset structured information output format can be determined according to the actual situation, and this embodiment does not limit it.

[0176] In this embodiment, the preset structured information output format can be a standardized strategy template used to transform strategy suggestions described in natural language into executable operational instructions. The preset structured information output format typically includes elements such as strategy type (e.g., buy call, sell put, straddle), execution price, expiration date, quantity, stop-loss and take-profit points, etc. The preset structured information output format not only facilitates user understanding but also provides fundamental support for subsequent risk assessment, simulated execution, and automated trading.

[0177] In practice, the structured conversion process of the pre-defined structured information output format is usually completed by a strategy structure parser. This parser uses techniques such as regular expressions and pattern matching algorithms to extract key parameters from the generated natural language text and map these parameters to standard fields. Furthermore, the strategy parameters can be adjusted based on factors such as market liquidity and transaction costs to ensure good executability and risk control capabilities.

[0178] The strategy generation method provided in this application constructs a multi-dimensional market state vector, matches it with structured strategy rules, and introduces an enhanced generation mechanism when a match is not found, ultimately transforming strategy suggestions into structured execution plans. This method not only improves the coverage and accuracy of strategy recommendations but also enhances the interpretability and operability of the strategy generation device. It achieves a complete closed loop from unstructured market information to executable strategies.

[0179] In this embodiment of the application, the process by which the strategy generation device transforms the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format includes: parsing the target strategy suggestion using the preset structured information output format to obtain parsed information; performing a profit and risk assessment based on the parsed information to obtain an assessment result; and generating the target strategy based on the assessment result.

[0180] In this embodiment, the output format of the preset structured information can be information configured in the strategy generation device, information transmitted to the strategy generation device from other devices, or information obtained by the strategy generation device through other means. The specific way in which the strategy generation device obtains the output format of the preset structured information can be determined according to the actual situation, and this embodiment does not limit it.

[0181] In this embodiment, the preset structured information output format is a template structure used to standardize the expression of option strategies. The preset structured information output format may include key parameters such as strategy type (e.g., buying a call, selling a straddle), strike price, expiration date, quantity, stop-loss / take-profit points, etc., and is stored in JSON or other machine-readable formats. By using the preset structured information output format, it is possible to ensure that target strategy suggestions from different sources or in different forms are processed and displayed uniformly, providing consistent basic data for subsequent risk assessment and strategy generation. For example, upon receiving a target strategy suggestion in natural language form, the suggestion is first converted into information conforming to the preset structured information output format for further processing. The structured expression not only improves compatibility and scalability but also facilitates data interaction between the structured expression and other modules (such as a simulation evaluation engine).

[0182] For example, the output format of the preset structured information can be a parameterized combination structure that includes elements such as option direction, strike price, expiration date, and margin requirements.

[0183] In this embodiment, the risk-return assessment is the result of a quantitative analysis of the profitability and risk exposure of a target strategy recommendation under potential market conditions. The risk-return assessment can be based on option pricing models (such as the Black-Scholes model) and historical market data backtesting methods to calculate indicators such as the expected return, maximum drawdown, Sharpe ratio, Delta value, Gamma value, and Vega value of the target strategy recommendation. These indicators together constitute a comprehensive evaluation system for the target strategy recommendation, helping to determine whether the target strategy recommendation has recommendation value.

[0184] Specifically, based on the analyzed target strategy recommendation parameters, Monte Carlo simulations or historical scenario backtesting can be used to predict the performance of the target strategy recommendation under different market conditions. For example, for a target strategy recommendation of a buy straddle, the profit and loss distribution of the buy straddle target strategy recommendation during periods of significant market volatility might be simulated, and the risk exposure of the buy straddle target strategy recommendation would be assessed accordingly. Finally, an evaluation result containing multiple dimensions will be output for reference in the next step of strategy generation.

[0185] For example, Monte Carlo simulations can be performed on the return distribution of a strategy under different market conditions using QuantLib / self-developed pricing engine, outputting profit and loss curves and risk exposure indicators to obtain evaluation results.

[0186] It should be noted that the target strategy is the final executable strategy determined by assessing the returns and risks of relevant financial assets and optimizing and adjusting it based on the assessment results.

[0187] The process of generating a target strategy typically includes the following aspects: 1. The strategy structure is confirmed by the investment manager. The investment manager determines the optimal option combination structure based on the evaluation results, such as whether to choose a single-leg or multi-leg strategy, and decides on key parameters such as strike price, expiration date, and position size.

[0188] 2. Execution parameter settings: Based on factors such as market liquidity and transaction costs, set reasonable stop-loss / take-profit points, holding periods, and other execution details to ensure that the target strategy is practically operable.

[0189] 3. Strategy Priority Ranking: When multiple feasible strategies exist, they are ranked according to indicators such as win rate, risk exposure, and historical performance, and the target strategy that best suits the current market conditions is recommended.

[0190] 4. Output structured results: Transform the target strategy suggestions into structured execution instructions for users to execute directly or for automated trading systems to call.

[0191] In this embodiment, by introducing a preset structured information output format to parse the target strategy recommendations, an efficient conversion from unstructured text to structured data can be achieved. This improves the consistency and accuracy of target strategy recommendation processing, thereby supporting more complex return and risk assessment logic, and ultimately generating more robust and operable option strategies corresponding to market conditions.

[0192] In this embodiment of the application, when there are multiple recommendation strategies, priority ranking and combination recommendations can be made based on dimensions such as historical win rate, risk exposure, and execution frequency.

[0193] In this embodiment, after the strategy generation device converts the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format, it also monitors the market performance information of the target strategy; and marks the target strategy based on the market performance information meeting preset warning conditions.

[0194] It should be noted that market performance information refers to market data, changes in returns, risk indicators, and other key parameters for evaluating the performance of the target strategy, recorded after the target strategy is actually executed or simulated. Market performance information may include, but is not limited to, the following: Actual profit or loss amount; Profit / Loss Ratio; Maximum Drawdown; Sharpe Ratio; Volatility changes during the strategy execution period; The Greek values ​​(such as Delta, Gamma, Vega, etc.) in the option portfolio undergo numerical changes; Analyze market data (such as the trend of the underlying index, implied volatility, trading volume, etc.) and compare it with the timing of strategy execution.

[0195] Market performance information can be obtained through real-time data collection, historical backtesting, or simulated trading, and is used to evaluate the stability, profitability, and adaptability of the target strategy under different market environments. Market performance information is not only the foundation for continuous optimization of the strategy generation device, but also a crucial basis for the risk management module to determine whether an early warning mechanism needs to be triggered.

[0196] Understandably, by monitoring the market performance of the target strategy, a comprehensive understanding of its effectiveness in the actual market can be achieved, thus providing data support for subsequent risk control, strategy adjustments, and model updates. This enables closed-loop feedback for the strategy generation device, improving its adaptability and long-term stability, thereby continuously optimizing the quality of strategy generation and increasing the user's return on investment.

[0197] It should be noted that preset warning conditions refer to a set of thresholds or logical judgment criteria set based on risk control rules, user preferences, or historical experience. When market performance information reaches or exceeds the preset warning conditions, the target strategy will be considered to have potential risks or abnormal behavior, and the target strategy will be flagged. Common warning conditions include: The current revenue is detected to be lower than the expected threshold.

[0198] Maximum drawdown exceeds the set limit The target strategy failed to generate positive returns over several consecutive periods. Volatility deviates from the normal range; A rapid change in the Delta value of an option portfolio indicates that investors may face significant price sensitivity risk. The target strategy performs unexpectedly after a specific market event occurs.

[0199] In this embodiment, marking a target strategy means adding it to a high-risk monitoring list, suspending its recommended use, reminding users to manually review it, or automatically initiating an adjustment process for the target strategy. For example, if a target strategy incurs losses for three consecutive weeks in a simulation test, and the maximum drawdown exceeds 5%, the target strategy will be automatically marked as high-risk, and investment advisors will be notified to review it.

[0200] In this embodiment, by dynamically monitoring market performance information and marking target strategies based on preset early warning conditions, potential problems can be identified in the early stages of strategy execution, preventing further losses. This dynamic monitoring of market performance information and marking target strategies based on preset early warning conditions enhances risk control capabilities, reduces the negative impact of strategy failure, and strengthens the robustness of the strategy generation device and user trust in it.

[0201] Understandably, by introducing a market performance monitoring and early warning mechanism after strategy generation, a closed-loop feedback mechanism for strategy execution effectiveness is achieved. This approach ensures the accuracy and security of strategy recommendations, thereby driving continuous optimization of the strategy generation device and ultimately enabling the construction of a more intelligent, efficient, and sustainable options strategy generation and optimization system.

[0202] In this embodiment of the application, after a loss warning is triggered, the weight parameters (such as win rate, hit frequency, etc.) can be dynamically adjusted through a feedback mechanism to achieve weight updates and priority adjustments based on the actual strategy performance.

[0203] In the embodiments of this application, the recommendation strategy can be output in the form of natural language + structured parameters, along with an explanation path (such as hit rules and historical reference cases).

[0204] It should be noted that the hit rule refers to the rule entries of "market conditions → strategy suggestions" matched by the system during the strategy recommendation process. These rules come from the built rule base and are presented along with the strategy output as part of the interpretation path, supporting risk feedback and system self-learning.

[0205] In this embodiment of the application, the market performance of the recommended strategy can be tracked in real time. If a loss warning is triggered, feedback information is recorded and the rule weights are updated.

[0206] In this embodiment of the application, after the strategy generation device converts the target strategy suggestion into the target strategy corresponding to the market state according to the preset structured information output format, it also receives the annotation information of the target strategy; and updates the structured semantic scene library based on the annotation information and the target strategy.

[0207] It should be noted that annotation information refers to additional explanations or evaluations added by humans or devices to the generated strategy recommendations. Annotation information can include the effectiveness score of the target strategy, risk level assessment, execution feedback results, etc. Through annotation information, the strategy generation device can obtain the latest data on the performance of the target strategy, which can then be used for subsequent model training and rule optimization. For example, investment advisors can score a target strategy based on actual trading results (e.g., high / medium / low) or add annotations (e.g., the target strategy performed poorly when volatility decreased). Annotation information can serve as incremental learning samples, helping to identify the applicability boundaries of the target strategy under specific market conditions and dynamically adjust the weights or triggering logic of relevant rules.

[0208] Understandably, by receiving annotation information of the target strategy, the strategy generation device can obtain actual execution feedback and professional evaluation opinions of the target strategy, providing data support for subsequent rule updates and model optimization, thereby achieving continuous adaptation and accuracy improvement.

[0209] It's important to note that the Structured Semantic Scenario Library (SSL) is a structured database used to store historical market scenarios, target strategy recommendations, and the execution results of those recommendations. The SSL organizes data in a time-series manner and includes multiple fields such as market state vectors, unstructured semantic descriptions, target strategy recommendation templates, and execution result metrics. By analyzing received annotation information, new market scenarios and target strategy recommendations can be combined and written into the SSL, and the labels or weights of existing records can be adjusted based on the annotation content. For example, if a target strategy is marked as having a high win rate after multiple uses, the confidence score of the corresponding scenario record will increase, and this target strategy may be preferentially matched to similar market states. Conversely, if a target strategy is marked as inefficient or too risky, its weight will decrease or it will be marked as not recommended. Furthermore, the SSL can also be used to support similar scenario retrieval in the RAG (Rapid Algorithm for Optimized Context) generation mechanism, improving the accuracy and logical consistency of target strategy completion.

[0210] In this embodiment, by updating the structured semantic scenario library based on annotation information and target strategy, dynamic maintenance and expansion of strategy knowledge can be achieved, enhancing the ability to cope with complex market situations. At the same time, it provides richer and more accurate historical references for future target strategy generation, forming a closed-loop optimization mechanism.

[0211] It is understandable that by receiving annotation information of the target strategy and updating the structured semantic scenario library based on the annotation information, the continuous iteration of strategy knowledge and automatic optimization of the rule base can be achieved, thereby improving its strategy adaptability in different market environments and significantly enhancing the intelligence level and practicality of the strategy generation device.

[0212] In this embodiment of the application, investment advisors can be supported in scoring and annotating strategy recommendations to form high-quality labeled samples for incremental optimization of the generation model.

[0213] In this embodiment, the strategy generation device further includes a strategy structure generation and simulation evaluation layer. This layer transforms strategy suggestions obtained from the rule-based reasoning engine or the RAG-enhanced generation module into executable option strategy structures and performs simulation evaluations based on different market conditions to ensure the effectiveness, operability, and risk control of the strategies. This module provides accurate option combination structures, optimized execution strategies, and reliable risk assessments, ensuring the practical feasibility of each recommended strategy.

[0214] The strategy structure generation and simulation evaluation layer includes a strategy structure parsing and standardization module. The main function of this module is to transform the generated strategy suggestions into standardized option trading structures. Strategy generation can involve single-leg options (such as call and put options) or multi-leg combinations (such as straddles and butterfly spreads). The specific implementation process of this module includes two parts: parsing and generating standardized structures. The parsing process first performs pattern recognition and mapping on the generated natural language strategy suggestions (such as "buy a call option, strike price 100, term 1 month") to determine key parameters such as option type (buy / sell), strike price, and expiration date. Then, regular expressions and pattern matching algorithms are used to extract important fields:

[0215] The standardized structure generation part includes first generating a strategy template that conforms to the specifications of the options trading platform, as shown in formula (17): (17) Then all parameters are converted into calculable values ​​for use in subsequent risk assessments and simulations.

[0216] The strategy structure generation and simulation evaluation layer includes an option portfolio optimization and execution parameter calculation module. This module is responsible for optimizing option portfolios based on market conditions and strategy objectives (such as maximizing returns and minimizing risks).

[0217] Optimization goal: Choose the option combination type that suits the current market conditions, including: recommending to buy call options (single-leg strategy) when the index is rising; or recommending to sell straddles (double-leg strategy) when the index is fluctuating.

[0218] Calculation method: The theoretical price and Greek values ​​(Delta, Gamma, Vega, etc.) of each option are calculated using option pricing models (such as the Black-Scholes model) and incorporated into the portfolio evaluation.

[0219] For multi-leg combination strategies, the net profit and risk exposure of the option combination (such as Delta hedging, Gamma hedging, Vega hedging) are calculated as shown in formula (18): (18) Where i represents each option in the portfolio.

[0220] The potential returns and maximum drawdowns of this strategy under different market scenarios are evaluated through Monte Carlo simulations or historical backtesting.

[0221] The strategy structure generation and simulation evaluation layer also includes a return simulation and risk assessment module. This module is responsible for simulating the returns of the generated strategies, evaluating their performance under different market conditions, and adjusting the strategy recommendations based on predetermined risk thresholds.

[0222] Simulation process: By using Monte Carlo simulation or historical scenario backtesting, different execution paths of the strategy are simulated to calculate the potential gains and losses of the strategy over a future period, as shown in formula (19): (19) Where t represents the market state at each moment, and T is the total duration of the simulation.

[0223] The strategy is evaluated based on indicators such as volatility, maximum drawdown, and Sharpe ratio, and risk control recommendations are generated, as shown in formula (20): (20) Based on the risk assessment results, the system will make the following decisions: if the strategy risk is too high (e.g., the maximum drawdown exceeds the set threshold), a "strategy adjustment" reminder will be issued; if the strategy meets the risk control standards, it is recommended to continue execution.

[0224] The strategy structure generation and simulation evaluation layer also includes a strategy execution optimization and strategy combination suggestion module. This module provides final strategy execution recommendations and, when needed, offers various strategy combinations to adapt to different market scenarios.

[0225] Execution parameter optimization: During strategy execution, the strategy execution optimization and strategy combination suggestion module dynamically monitors market data, such as the deviation between the market price and theoretical price of options, and the liquidity of options, to optimize the execution strategy.

[0226] For multi-leg portfolio strategies, the system uses optimization algorithms (such as genetic algorithms and particle swarm optimization) to determine the number of each leg, the strike price, and the expiration date, thereby optimizing the overall risk-reward ratio of the strategy.

[0227] Strategy portfolio recommendations: Based on the current market conditions and risk appetite, the system recommends the optimal option combination strategy and calculates the expected return and risk exposure, providing this information to the user. If market volatility is high, a strategy such as "buying option combinations" is recommended. If market volatility is low, a recommended strategy is to "sell a straddle".

[0228] The strategy structure generation and simulation evaluation layer also includes a strategy execution and real-time risk monitoring and feedback module. During strategy execution, this module monitors market fluctuations and strategy performance in real time, and provides feedback and adjustments based on the actual operation of the strategy.

[0229] Specifically, it involves real-time monitoring of option prices, volatility changes, trading volume, and other data to promptly adjust stop-loss, take-profit points, or risk exposure. In the event of abnormal market volatility, a strategy adjustment mechanism is automatically triggered, adjusting strategy execution parameters to ensure investors can cope with potential market risks. Execution instructions are sent to external trading systems via API, enabling automated execution of option strategies.

[0230] The strategy structure generation and simulation evaluation layer provides a complete closed-loop support for option strategies from generation to execution through comprehensive return simulation, risk assessment, option portfolio optimization, and real-time feedback mechanisms. This not only effectively improves the operability and market adaptability of strategy recommendations but also ensures that investors can obtain efficient and stable investment returns in different market environments through real-time risk control and optimized strategy execution.

[0231] In this embodiment, the strategy generation device further includes a strategy output and risk feedback closed-loop layer. This layer plays a crucial role in the strategy generation and execution process, ensuring the timeliness and accuracy of the strategy output and continuously optimizing the strategy execution effect through real-time risk monitoring and feedback mechanisms.

[0232] The strategy output and risk feedback closed-loop layer includes an automatic strategy output and interpretation module. This module presents the generated option strategies to users in a structured manner and provides actionable explanations to ensure users understand the strategy's logic and execution requirements.

[0233] The automatic strategy output and interpretation module includes formatted strategy output: converting generated option strategies (such as buying call options, selling straddles, etc.) into standardized option trading structures, including: Strategy type (e.g., buy, sell, straddle, etc.); Option strike price, expiration date, quantity, etc.; Specific stop-loss / take-profit points settings.

[0234] For example, the output format of the strategy recommendation: The automatic policy output and interpretation module also includes policy logic interpretation path generation: The output includes the corresponding strategy generation logic path, enabling users to understand why that strategy was chosen. For example: "The current market conditions match strategy rule R5: the index is rising, implied volatility is rising, and the volatility term structure is inverted. This strategy has performed excellently in historical backtesting with a win rate of 78%, and we recommend buying call options." The automatic strategy output and interpretation module also includes strategy operability reminders: providing operability reminders for the strategy, informing investors how to execute the strategy, such as the market liquidity requirements for option purchases, execution period, and transaction costs to be considered.

[0235] The strategy output and risk feedback closed-loop layer also includes a real-time risk monitoring and market dynamic feedback module. This module is responsible for monitoring market fluctuations and strategy execution risks in real time, promptly identifying potential market risks, and providing adjustment suggestions.

[0236] The real-time risk monitoring and market dynamic feedback module includes real-time data stream access and processing: real-time acquisition of market data (such as implied volatility of options, open interest, trading volume, index rise and fall, etc.) and external market information (such as fluctuations in US stocks, changes in the foreign exchange market, etc.) via API.

[0237] The real-time risk monitoring and market dynamic feedback module includes volatility and risk indicator calculation: it calculates risk indicators for option strategies based on real-time data, such as Delta (i.e.,...). The specific calculation method is shown in formula (21), Gamma risk, Vega (volatility), Theta (time decay) and other Greek values, as well as the maximum drawdown of the strategy (i.e., the maximum drawdown, the specific calculation method is shown in formula (22)).

[0238] (twenty one) (twenty two) The real-time risk monitoring and market dynamic feedback module includes abnormal volatility detection and risk warning: it uses adaptive thresholds to detect abnormal market volatility, such as sharp changes in implied volatility of options or excessive index volatility, triggering a risk warning. After a risk warning is triggered, the system automatically calculates strategy adjustment suggestions, such as changing the stop-loss point or selecting a hedging strategy.

[0239] The strategy output and risk feedback closed-loop layer also includes a risk adjustment and strategy dynamic feedback module. This module automatically adjusts the strategy based on preset rules or real-time data when significant market changes occur or deviations occur during strategy execution, ensuring effective risk control.

[0240] The risk adjustment and strategy dynamic feedback module includes automatic strategy adjustment: through real-time monitoring of the strategy, the system can automatically adjust strategy parameters. For example, when market volatility is high, the system may suggest reducing option positions, adjusting strike prices, or shortening expiration dates. Through adaptive adjustment models (such as reinforcement learning models based on neural networks), the system learns and optimizes the strategy in real time based on market feedback.

[0241] The risk adjustment and strategy dynamic feedback module includes dynamic strategy portfolio optimization: when the system detects that the risk of a certain strategy is too high, it will automatically suggest combining the strategy with other strategies (such as selling put options, butterfly spreads, etc.) to reduce the overall risk exposure.

[0242] The risk adjustment and strategy dynamic feedback module includes automatic adjustment of stop-loss / take-profit points: based on real-time market fluctuations, it automatically adjusts stop-loss and take-profit points to ensure that investors' risk control is within an acceptable range.

[0243] The strategy output and risk feedback closed-loop layer also includes a backtesting and feedback optimization module. This module compares and analyzes the strategy output with actual market feedback, and uses the backtesting results to optimize the strategy generation process.

[0244] The backtesting and feedback optimization module includes the strategy backtesting process: backtesting the strategy on historical market data (such as past option prices, implied volatility data, etc.), evaluating its return and risk performance, and generating a historical backtesting report for the strategy. Backtesting methods can be based on Monte Carlo simulation or historical scenario simulation to calculate the strategy's performance under different market conditions.

[0245] The backtesting and feedback optimization module includes strategy performance feedback and model updates: it compares backtesting results with real-time market data to calculate the strategy's Alpha and Beta (i.e.,...). ):

[0246]

[0247] Among them, R p R represents the strategy return rate. m This is the market benchmark rate of return, which is the output of the Capital Asset Pricing Model (CAPM).

[0248] Adjust the policy parameters based on the backtesting results, and incrementally optimize the generative model using reinforcement learning or gradient descent methods to improve the long-term stability and returns of the policy.

[0249] The backtesting and feedback optimization module includes strategy performance report generation: automatically generating backtesting reports, including data such as the strategy's historical performance, maximum drawdown, profit / loss ratio, and Sharpe ratio, to provide investors with decision-making support.

[0250] The strategy output and risk feedback closed-loop layer also includes a strategy execution and feedback closed loop. During the actual execution of the strategy, the strategy execution and feedback closed loop ensures that the strategy is updated and adjusted in sync with market changes.

[0251] The strategy execution and feedback loop includes the issuance of execution instructions: based on the instructions generated by the strategy, the system sends execution instructions to the external trading platform through the API interface to ensure that the options strategy can be executed quickly.

[0252] The strategy execution and feedback loop includes real-time feedback collection and risk reassessment: collecting real-time market data (such as option strike prices, trading volume, etc.) and comparing it with strategy backtesting and historical simulation data to assess the risks and returns during the execution process.

[0253] The strategy execution and feedback loop includes strategy re-optimization: continuously collecting feedback information and feeding the feedback data into the model as incremental data to update the model and optimize the strategy, forming a feedback loop and further improving the accuracy of strategy recommendations.

[0254] The closed-loop strategy output and risk feedback layer, through real-time strategy output, risk monitoring, dynamic adjustment, and backtesting optimization, not only enhances the executability and market adaptability of option strategies but also ensures that investors can flexibly adjust their strategies in constantly changing market environments, maximizing investment returns and minimizing potential risks. This significantly improves the intelligence, operability, and long-term effectiveness of option investment decisions.

[0255] For example, such as Figure 2As shown: First, the historical financial investment information (investment advisor text) can be obtained; historical market states and corresponding historical strategies can be extracted from the historical financial investment information (strategy extraction); a knowledge graph can be established based on the correspondence between the historical market states and the historical strategies; the knowledge graph can be used as the strategy mapping rule (mapping rule construction). Then, the current market state (current market data) can be obtained; it can be determined whether the current market state triggers the strategy mapping rule. Specifically, the process of determining whether the current market state triggers the strategy mapping rule includes: establishing a multi-dimensional state vector based on the current market state (state vector construction); searching for multiple rule nodes that match the multi-dimensional state vector in the strategy mapping rule (real-time market state matching); determining multiple operation strategies corresponding to the multiple rule nodes, and multiple profit information of the multiple operation strategies; determining multiple confidence levels corresponding to the multiple operation strategies based on the multiple profit information; and determining whether the current market state triggers the strategy mapping rule based on the multiple confidence levels (mapping rule triggering and strategy invocation). Without triggering the strategy mapping rules, multiple historical sample cases similar to the current market state are retrieved from the structured semantic scenario library. These historical sample cases and the current market state are then input into a large language model to obtain the target strategy suggestion (RAG semantic retrieval and strategy generation). Finally, the target strategy suggestion is parsed using the preset structured information output format to obtain parsed information. A profit and risk assessment is performed based on the parsed information to obtain an assessment result. The target strategy is generated based on the assessment result (strategy structured suggestion output). Subsequently, the market performance information of the target strategy is monitored (profit analysis). If the market performance information meets preset warning conditions, the target strategy is marked (execution / adjustment feedback). Before retrieving multiple historical sample cases similar to the current market state from the structured semantic scenario library, historical market states, historical strategy suggestions, and corresponding historical execution results are also obtained. The structured semantic scenario library (historical market scenario library) is established based on the historical market states, historical strategy suggestions, and historical execution results.

[0256] Understandably, when the strategy generation device determines that the current market state has not triggered the strategy mapping rules, it uses an enhanced generation mechanism to determine the target strategy suggestion corresponding to the current market state. Then, according to the preset structured information output format, the target strategy suggestion is transformed to obtain the target strategy corresponding to the market state. This allows the target strategy to be determined based on the enhanced generation mechanism even when multiple conditions interact and cause the strategy mapping logic to fail or have a blind spot, thereby improving the accuracy of strategy generation.

[0257] Based on the same inventive concept as the above-mentioned strategy generation method, this application provides a strategy generation device 1, corresponding to a strategy generation method; Figure 3 A schematic diagram of the composition structure of a strategy generation device provided in this application embodiment. Figure 1 The strategy generation device 1 may include: Acquisition unit 11 is used to acquire the current market status; The determining unit 12 is used to determine whether the current market state triggers the strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; if the strategy mapping rule is not triggered, the target strategy suggestion corresponding to the current market state is determined based on the enhanced generation mechanism. The conversion unit 13 is used to convert the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format.

[0258] In some embodiments of this application, the apparatus further includes an establishment unit and a search unit; The establishment unit is used to establish a multi-dimensional state vector based on the current market state. The search unit is used to search for multiple rule nodes in the policy mapping rules that match the multidimensional state vector: The determining unit 12 is used to determine multiple operation strategies corresponding to the multiple rule nodes, and multiple revenue information of the multiple operation strategies; determine multiple confidence levels corresponding to the multiple operation strategies based on the multiple revenue information; and determine whether the current market state triggers the strategy mapping rule based on the multiple confidence levels. In some embodiments of this application, the determining unit 12 is used to determine the index development trend, volatility trend, implied volatility term structure slope, and market sentiment indicators of the target based on the current market state. The establishment unit is used to establish the multidimensional dynamic vector based on the index development trend, volatility trend, implied volatility term structure slope, and market sentiment indicator.

[0259] In some embodiments of this application, the determining unit 12 is configured to determine that the current market state has not triggered a strategy mapping rule when each of the plurality of confidence levels is less than a first threshold; and to determine that the current market state has triggered a strategy mapping rule when each of the plurality of confidence levels is greater than or equal to the first threshold. In some embodiments of this application, the determining unit 12 is used to determine that the current market state has not triggered the strategy mapping rule if no rule node matching the multidimensional state vector is found in the strategy mapping rule. In some embodiments of this application, the apparatus further includes a retrieval unit and an input unit; The retrieval unit is used to retrieve multiple historical sample cases similar to the current market state from the structured semantic scene library; The input unit is used to input the multiple historical sample cases and the current market state into the large language model to obtain the target strategy suggestion.

[0260] In some embodiments of this application, the acquisition unit 11 is used to acquire historical market status, historical strategy suggestions, and historical execution results corresponding to the historical strategy suggestions; The establishment unit is used to establish the structured semantic scenario library based on the historical market status, the historical strategy suggestions, and the historical execution results. In some embodiments of this application, the apparatus further includes an output unit, an evaluation unit, and a generation unit; The output unit is used to parse the target strategy suggestion using the output format of the preset structured information to obtain parsed information; The evaluation unit is used to perform a benefit-risk assessment based on the parsed information and obtain an evaluation result. The generation unit is used to generate the target strategy based on the evaluation results. In some embodiments of this application, the apparatus further includes an extraction unit; The establishment unit is used to establish strategy mapping rules based on historical financial investment information; and to establish a knowledge graph based on the correspondence between the historical market state and the historical strategy. The acquisition unit 11 is used to acquire the historical financial investment information; The extraction unit is used to extract historical market states and historical strategies corresponding to the historical market states from the historical financial investment information. The determining unit 12 is used to use the knowledge graph as the policy mapping rule. In some embodiments of this application, the apparatus further includes a parsing unit; The input unit is used to input the historical financial investment information into the information extraction model to obtain the initial historical market state and the initial historical strategy; The parsing unit is used to parse the correlation between the initial historical market state and the initial historical strategy to obtain the historical market state and the historical strategy. In some embodiments of this application, the device further includes an adjustment unit; The adjustment unit is used to adjust the historical strategy using a preset strategy template format to obtain the adjusted historical strategy. The establishment unit is used to establish a knowledge graph based on the correspondence between the historical market state and the adjusted historical strategy.

[0261] In some embodiments of this application, the device further includes a monitoring unit and a tagging unit; The monitoring unit is used to monitor the market performance information of the target strategy; The marking unit is used to mark the target strategy based on the market performance information meeting preset warning conditions. In some embodiments of this application, the apparatus further includes an updating unit and a receiving unit; The receiving unit is used to receive annotation information for the target strategy; The update unit is used to update the structured semantic scene library based on the annotation information and the target strategy.

[0262] It should be noted that, in practical applications, the acquisition unit 11, determination unit 12 and conversion unit 13 mentioned above can be implemented by the processor 14 on the electronic device 1, specifically by a CPU (Central Processing Unit), MPU (Microprocessor Unit), DSP (Digital Signal Processor) or Field Programmable Gate Array (FPGA), etc.; the data storage mentioned above can be implemented by the memory 15 on the electronic device 1.

[0263] This application also provides an electronic device 1, such as... Figure 4 As shown, the electronic device 1 includes a processor 14, a memory 15, and a communication bus 16. The memory 15 communicates with the processor 14 through the communication bus 16. The memory 15 stores programs executable by the processor 14. When the program is executed, the processor 14 executes the strategy generation method as described above.

[0264] In practical applications, the aforementioned memory 15 can be volatile memory, such as random-access memory (RAM); or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 14.

[0265] This application provides a computer-readable storage medium having a computer program thereon, which, when executed by a processor 14, implements the strategy generation method as described above.

[0266] For example, this application also provides a computer program product, including a computer program that can be executed by a processor 14 to complete the steps described in the aforementioned strategy generation method.

[0267] Understandably, when the strategy generation device determines that the current market state has not triggered the strategy mapping rules, it uses an enhanced generation mechanism to determine the target strategy suggestion corresponding to the current market state. Then, according to the preset structured information output format, the target strategy suggestion is transformed to obtain the target strategy corresponding to the market state. This allows the target strategy to be determined based on the enhanced generation mechanism even when multiple conditions interact and cause the strategy mapping logic to fail or have a blind spot, thereby improving the accuracy of strategy generation. Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0268] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0269] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0270] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0271] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A strategy generation method, characterized in that, The method includes: Obtain the current market status; Determine whether the current market state triggers the strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; Without triggering the aforementioned strategy mapping rules, the target strategy recommendation corresponding to the current market state is determined based on the enhanced generation mechanism; The target strategy suggestion is transformed according to the preset structured information output format to obtain the target strategy corresponding to the market state.

2. The method according to claim 1, characterized in that, Determining whether the current market state triggers the strategy mapping rule includes: Establish a multi-dimensional state vector based on the current market state; Search for multiple rule nodes in the policy mapping rules that match the multidimensional state vector: Determine multiple operation strategies corresponding to the multiple rule nodes, and multiple benefit information of the multiple operation strategies; Based on the multiple revenue information, determine multiple confidence levels corresponding to the multiple operational strategies; Based on the multiple confidence levels, determine whether the current market state triggers the strategy mapping rule.

3. The method according to claim 2, characterized in that, The step of establishing a multi-dimensional state vector based on the current market state includes: Based on the current market conditions, determine the underlying index trend, volatility trend, implied volatility term structure slope, and market sentiment indicators. The multidimensional dynamic vector is established based on the index development trend, volatility trend, implied volatility term structure slope, and market sentiment indicator.

4. The method according to claim 2, characterized in that, The step of determining whether the current market state triggers the strategy mapping rule based on the multiple confidence levels includes: If each of the multiple confidence levels is less than the first threshold, it is determined that the current market state has not triggered the strategy mapping rule; If each of the plurality of confidence levels is greater than or equal to a first threshold, the current market state trigger strategy mapping rule is determined.

5. The method according to claim 2, characterized in that, After establishing the multidimensional state vector based on the current market state, the method further includes: If no rule node matching the multidimensional state vector is found in the strategy mapping rules, it is determined that the current market state has not triggered the strategy mapping rules.

6. The method according to claim 1, characterized in that, The target strategy recommendations based on the enhanced generation mechanism to determine the current market state include: Retrieve multiple historical sample cases similar to the current market state from the structured semantic scene library; By inputting the multiple historical sample cases and the current market state into the large language model, the target strategy suggestion is obtained.

7. The method according to claim 6, characterized in that, Before retrieving multiple historical sample cases similar to the current market state from the structured semantic scene library, the method further includes: Obtain historical market conditions, historical strategy recommendations, and the historical execution results corresponding to the historical strategy recommendations; The structured semantic scenario library is established based on the historical market conditions, the historical strategy recommendations, and the historical execution results.

8. The method according to claim 1, characterized in that, The step of transforming the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format includes: The target strategy suggestion is parsed using the output format of the preset structured information to obtain parsed information; Based on the analyzed information, a risk-reward assessment is performed to obtain the assessment result; The target strategy is generated based on the evaluation results.

9. The method according to claim 1, characterized in that, Before obtaining the current market status, the method further includes: Obtain the aforementioned historical financial investment information; Extract historical market conditions and corresponding historical strategies from the historical financial investment information; A knowledge graph is established based on the correspondence between the historical market states and the historical strategies. The knowledge graph is used as the policy mapping rule.

10. The method according to claim 9, characterized in that, The step of extracting historical market states and corresponding historical strategies from the historical financial investment information includes: The historical financial investment information is input into the information extraction model to obtain the initial historical market state and the initial historical strategy. The relationship between the initial historical market state and the initial historical strategy is analyzed to obtain the historical market state and the historical strategy.

11. The method according to claim 9, characterized in that, The step of establishing a knowledge graph based on the correspondence between the historical market states and the historical strategies includes: The historical strategy is adjusted using a preset strategy template format to obtain the adjusted historical strategy; A knowledge graph is established based on the correspondence between the historical market conditions and the adjusted historical strategies.

12. The method according to claim 1, characterized in that, After converting the target strategy suggestion according to a preset structured information output format to obtain the target strategy corresponding to the market state, the method further includes: Monitor the market performance information of the target strategy; If the market performance information meets the preset early warning conditions, the target strategy is marked.

13. The method according to claim 1, characterized in that, After converting the target strategy suggestion according to a preset structured information output format to obtain the target strategy corresponding to the market state, the method further includes: Receive annotation information for the target strategy; The structured semantic scene library is updated based on the annotation information and the target strategy.

14. A strategy generation apparatus, characterized in that, The device includes: The acquisition unit is used to acquire the current market status; A determining unit is used to determine whether the current market state triggers a strategy mapping rule; the strategy mapping rule is determined based on historical financial investment information; if the strategy mapping rule is not triggered, a target strategy suggestion corresponding to the current market state is determined based on an enhanced generation mechanism. The conversion unit is used to convert the target strategy suggestion into the target strategy corresponding to the market state according to a preset structured information output format.

15. An electronic device, characterized in that, The electronic device includes: The system includes a memory, a processor, and a communication bus, wherein the memory communicates with the processor via the communication bus, and the memory stores a policy-generated program executable by the processor. When the policy-generated program is executed, the processor performs the method as described in any one of claims 1 to 13.

16. A storage medium storing a computer program thereon, used in a strategy generation apparatus, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 13.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 13.

Images