Financial thinking chain data cleaning method, device and equipment, medium and program product

Abstract

The embodiment of the invention provides a financial thinking chain data cleaning method and device, equipment, a medium and a program product, and relates to the field of financial science and technology or other related technical fields. The method comprises the following steps: dividing data of a financial thinking chain into a surface semantic layer, a logical reasoning layer and a financial semantic layer, and respectively verifying term normalization of the surface semantic layer, formula reasonability of the logical reasoning layer and alignment of a conclusion of the financial semantic layer and a financial target; and performing multi-layer consistency evaluation on each layer of data after verification, generating a weighted score, identifying an abnormal thinking chain in the financial thinking chain according to the weighted score, and repairing the abnormal thinking chain. Based on the method, an accurate financial thinking chain can be obtained, and when the model uses the financial thinking chain for derivation, an accurate result can be obtained.

Description

Technical Field

[0001] This application relates to the field of financial technology or other related technical fields, and in particular to a financial thinking chain data cleaning method, apparatus, equipment, medium and program product. Background Technology

[0002] In certain scenarios within the financial field, large language models can generate multi-step logical chains through chain-of-thought (CoT) reasoning to support the interpretability and accuracy of financial decisions. These scenarios can include portfolio optimization, risk assessment, return forecasting, asset allocation, and compliance review. For example, in portfolio optimization, the model can generate a complete reasoning chain regarding portfolio optimization based on market assumptions (such as the risk-free rate and volatility), calculation processes (such as the Sharpe ratio and risk-return ratio), and final conclusions (such as asset weight allocation).

[0003] However, financial CoT datasets often contain some problems, such as semantic inconsistencies between inference steps, logical conflicts across chains, and implicit noise, which need to be repaired through data cleaning.

[0004] However, in some implementations, data cleaning methods that use regular expressions to validate terminology formats cannot effectively achieve accurate data cleaning in the financial thought chain. Summary of the Invention

[0005] This application provides a financial thinking chain data cleaning method, apparatus, device, medium, and program product, which is used to quantitatively identify problems such as mixed terminology, incorrect formulas, and deviations in conclusions through hierarchical analysis and multi-level consistency scoring, thereby improving data consistency and logical rationality and enhancing the accuracy of data cleaning.

[0006] In a first aspect, embodiments of this application provide a method for cleaning financial thinking chain data, the method comprising:

[0007] The data in the financial thinking chain is divided into a surface semantic layer, a logical reasoning layer, and a financial semantic layer.

[0008] The standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions with financial objectives in the financial semantic layer are verified separately.

[0009] Perform a multi-level consistency assessment on the verified data at each level to generate a weighted score;

[0010] The system identifies and corrects abnormal thought chains in the financial thought chain based on weighted scoring.

[0011] Secondly, embodiments of this application provide a financial thinking chain data cleaning device, the device comprising:

[0012] The segmentation module is used to divide the data of the financial thinking chain into a surface semantic layer, a logical reasoning layer, and a financial semantic layer.

[0013] The verification module is used to verify the standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions with financial objectives in the financial semantic layer.

[0014] The scoring module is used to perform multi-level consistency evaluation on the verified data at each level and generate a weighted score.

[0015] The repair module is used to identify abnormal thought chains in the financial thought chain based on weighted scoring and to repair these abnormal thought chains.

[0016] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, causing the processor to perform the embodiments described in the first aspect above.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the embodiments described in the first aspect above.

[0018] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, implements the implementation method described in the first aspect above.

[0019] The financial thinking chain data cleaning method, apparatus, equipment, medium, and program products provided in this application embodiment can divide the financial thinking chain data into a surface semantic layer, a logical reasoning layer, and a financial semantic layer. It verifies the terminology standardization of the surface semantic layer, the rationality of the formulas in the logical reasoning layer, and the alignment of the conclusions of the financial semantic layer with financial objectives. A multi-level consistency assessment is performed on the verified data of each layer to generate a weighted score. Based on the weighted score, abnormal thinking chains in the financial thinking chain are identified and repaired. In other words, this application embodiment can verify the accuracy of each layer separately and repair abnormal thinking chains in the financial thinking chain by combining multi-level semantic consistency assessment, thereby obtaining a more accurate financial thinking chain. Therefore, when the model uses the financial thinking chain for derivation, it can obtain more accurate results. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0021] Figure 1A flowchart illustrating the financial thinking chain data cleaning method provided in this application embodiment;

[0022] Figure 2 This is a schematic diagram illustrating the working principle of the data cleaning apparatus provided in the embodiments of this application;

[0023] Figure 3 A schematic diagram of the structure of the financial thinking chain data cleaning device provided in the embodiments of this application;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0025] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0027] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.

[0028] Furthermore, the technical solution involved in this application, which involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.) and the use of artificial intelligence technology for automated decision-making, and makes decisions that have a significant impact on personal rights based on the results of automated decision-making, provides users with corresponding operation entry points for users to choose to agree to or reject the results of automated decision-making; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0029] In this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0030] In the embodiments of this application, the use of terms such as "first" and "second" is to distinguish between identical or similar items that have essentially the same function and effect. For example, "first electronic device" and "second electronic device" are merely used to distinguish different electronic devices and do not limit their order of execution. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply that they are different.

[0031] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0032] It should be noted that the financial thinking chain data cleaning method, apparatus, equipment, media and program products provided in this application can be used in the field of financial technology, or in any field other than financial technology. The application field of the financial thinking chain data cleaning method, apparatus, equipment, media and program products in this application is not limited.

[0033] The following is an explanation of some terms used in the embodiments of this application:

[0034] Financial thought chain: This refers to data that contains the step-by-step reasoning process for problem-solving in the financial field. Using this data, the model can simulate the human thought process when dealing with financial problems, breaking down complex financial issues into multiple interconnected intermediate steps, and ultimately arriving at a conclusion through step-by-step reasoning. In other words, financial thought chain data is the model's input. By learning the reasoning patterns and logical relationships in this data, the model learns how to think and reason step-by-step when facing financial problems, thereby generating corresponding multi-step logical chains as output, providing users with analytical results or decision-making suggestions. Financial thought chain data itself may contain multiple thought chains (also called logical chains), and the specific content is not limited.

[0035] In scenarios such as portfolio optimization, risk assessment, return forecasting, asset allocation, and compliance review, models can generate multi-step logical chains through financial CoT data inference. The complexity and multi-dimensionality of financial CoT data make data quality a decisive factor in model performance. Data quality includes, for example, cross-step semantic dependencies and cross-chain logical consistency requirements.

[0036] In practical applications, the need for cleaning financial CoT data is particularly urgent. For example, if there are issues such as mixed terminology, incorrect formulas, or conflicting assumptions across chains within a particular chain of financial CoT data, it may lead to model training bias, reducing the reliability of predictions and risk control capabilities. Therefore, the financial sector urgently needs a CoT cleaning method that can systematically handle multi-dimensional data problems to improve the theoretical performance and practical application value of large models in financial decision-making.

[0037] In existing technologies, some methods involve manually cleaning financial CoT data, but this relies on human experience and the accuracy is difficult to control. Other methods may use regular expressions to check terminology format, but they cannot detect semantic inconsistencies caused by mixed terminology, or they only focus on the standardization of terminology or the correctness of logical formulas, lacking a systematic analysis of the consistency of data at different levels. This results in low accuracy of the financial thought chain and makes it difficult to meet the needs of large financial models for high-quality training data.

[0038] In view of this, embodiments of this application provide a method for cleaning financial thought chains. This method can divide the data of financial thought chains into a surface semantic layer, a logical reasoning layer, and a financial semantic layer. It verifies the terminology standardization of the surface semantic layer, the rationality of the formulas in the logical reasoning layer, and the alignment of the conclusions with financial objectives in the financial semantic layer. A multi-level consistency assessment is performed on the verified data of each layer to generate a weighted score. Abnormal thought chains in the financial thought chain are identified based on the weighted score, and these abnormal thought chains are repaired. That is, embodiments of this application can verify the accuracy of each layer separately, and combine this with a multi-level semantic consistency assessment to repair abnormal thought chains in the financial thought chain, thereby obtaining a more accurate financial thought chain. Therefore, when the model uses the financial thought chain for derivation, it can obtain more accurate results.

[0039] The technical solutions of this application will be described in detail below with reference to specific embodiments. The specific embodiments described below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0040] The method provided in this application can be applied to applications, websites, or mini-programs that have financial thinking chain data cleaning task processing functions. The financial thinking chain data cleaning task processing function is implemented on the application, website, or mini-program. For example, a computer with a financial thinking chain data cleaning application deployed can implement the financial thinking chain data cleaning task processing function by running the application. Another example is a terminal electronic device, such as a mobile phone, with a financial thinking chain data cleaning mini-program deployed, which can implement the financial thinking chain data cleaning task processing function by running the mini-program.

[0041] Figure 1 This is a flowchart illustrating the financial thinking chain data cleaning method provided in this application embodiment. The executing entity of this method can be an electronic device with corresponding data storage and computing capabilities, such as a computer, server, or server cluster. Figure 1 As shown, the method includes:

[0042] S101 divides the data of the financial thinking chain into a surface semantic layer, a logical reasoning layer, and a financial semantic layer.

[0043] In this embodiment, the data of the financial thinking chain includes, for example, multiple financial-related thinking chains. Each thinking chain may include a problem description, reasoning steps, and a final conclusion. The surface semantic layer may be the understanding of the literal meaning of the text, the logical reasoning layer may be the process of deriving implicit relationships based on surface information, and the financial semantic layer may be the semantics focusing on specific concepts and rules in the financial field.

[0044] For example, financial thinking chain data can be divided into a surface semantic layer, a logical reasoning layer, and a financial semantic layer using a pre-trained model. The input to this pre-trained model can be the financial thinking chain data, and the output can be the surface semantic layer, logical reasoning layer, and financial semantic layer of the financial thinking chain data. The specific implementation is not limited.

[0045] S102 verifies the standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions with financial objectives in the financial semantic layer.

[0046] In this embodiment, the surface semantic layer primarily checks terminology, symbols, and format. The logical reasoning layer primarily verifies the reasoning steps and formulas. The financial semantic layer primarily verifies whether the conclusions align with financial objectives, such as return prediction and risk matching.

[0047] For example, a directed graph of reasoning can be constructed using a Natural Language Processing (NLP) parser, regular expressions, and graph structure modeling to verify the standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions with financial objectives in the financial semantic layer.

[0048] For example, when validating the surface semantic layer, NLP parsers can be used to focus on the standardization of textual expressions, such as the consistency of terminology, the correctness of syntax and format, and the standardization of symbols; when validating the logical reasoning layer, regular expressions can be used to check the computational logic and the rationality of assumptions in the reasoning steps, such as verifying the appropriateness of the application of financial formulas or the selection of market parameters; when validating semantic layer analysis, graph structure modeling can be used to construct a directed graph of reasoning, and vector similarity can be used to judge the conclusions to evaluate the degree of alignment between the entire chain and the initial financial goal, such as ensuring that the reasoning process does not deviate from the expected prediction or decision-making needs.

[0049] S103 performs a multi-level consistency assessment on the verified data at each level and generates a weighted score.

[0050] For example, the verification scores of the surface semantic layer, the logical reasoning layer, and the financial semantic layer can be weighted to generate a weighted score.

[0051] For example, if the weighted score is Score, the validation score of the surface semantic layer is Ssemantic, the validation score of the logical reasoning layer is Slogic, and the validation score of the financial semantic layer is Ssyntax, then the following formula can be satisfied: Score = α * Ssemantic + β * Slogic + γ * Ssyntax. Here, α, β, and γ can be weighting coefficients, and their specific values ​​are not limited.

[0052] It should be noted that when performing multi-level consistency assessment on the verified data at each level and generating weighted scores, a linear relationship can be used instead of other methods, without any restrictions.

[0053] S104, Identify abnormal thought chains in the financial thought chain based on weighted scoring, and repair abnormal thought chains.

[0054] For example, when the weighted score is below a threshold, the thought chain can be marked as an abnormal thought chain, and the abnormal thought chain can be repaired.

[0055] The embodiments of this application do not limit the specific form of the abnormal thought chain or the repair method.

[0056] In summary, this application embodiment can divide the data of the financial thinking chain into a surface semantic layer, a logical reasoning layer, and a financial semantic layer. It verifies the terminology standardization of the surface semantic layer, the rationality of the formulas in the logical reasoning layer, and the alignment of the conclusions of the financial semantic layer with the financial objectives. A multi-layer consistency assessment is performed on the verified data of each layer to generate a weighted score. Based on the weighted score, abnormal thinking chains in the financial thinking chain are identified and repaired. That is, this application embodiment can verify the accuracy of each layer separately, and combine multi-layer semantic consistency assessment to repair abnormal thinking chains in the financial thinking chain, thereby obtaining a more accurate financial thinking chain. Therefore, when the model uses the financial thinking chain for derivation, it can obtain more accurate results.

[0057] In one possible implementation, the abnormal thought chain is repaired according to a repair strategy, which includes: correcting local anomalies in the abnormal thought chain by completing missing parameters based on a regression model; and / or reconstructing the reasoning steps using a sequence generation model to resolve chain anomalies in the abnormal thought chain; and / or resolving cross-chain contradictions in the abnormal thought chain by verifying the consistency of conclusions of multiple thought chains through a knowledge graph.

[0058] In this embodiment, the abnormal types of the abnormal thought chain can be divided into local abnormalities, chain abnormalities, and cross-chain abnormalities. Local abnormalities can refer to errors in a single step, such as omitting a time factor. Chain abnormalities can refer to the propagation of incorrect assumptions within the chain, such as deriving subsequent cash flows based on an incorrect risk-free interest rate, causing the entire chain to deviate. Cross-chain abnormalities can refer to contradictory conclusions from multiple chains in the same task, such as inconsistent return predictions for the same investment portfolio from different chains.

[0059] In practice, a graph database can be used to trace the error propagation path and divide the abnormal thought chain. The specific implementation is not limited.

[0060] In this embodiment of the application, a repair network driven by a pre-trained model can be used to process the abnormal thought chain according to the abnormality type. Specifically:

[0061] For local anomalies, the missing parameters can be supplemented based on the regression model to correct the local anomalies in the abnormal thinking chain and correct single errors, such as supplementing missing time factors or adjusting incorrect parameters. In this way, the integrity of the financial thinking chain data can be restored and subsequent calculation errors caused by missing parameters can be avoided.

[0062] For chain anomalies, sequence generation models can be used to reconstruct the reasoning steps to resolve the chain anomalies in the abnormal thought chain. For example, cash flow can be re-derived based on the corrected interest rate. This can restore the coherence of the chain logic and reduce the propagation of cascading errors.

[0063] For cross-chain anomalies, cross-chain contradictions in anomalous thought chains can be resolved by verifying the consistency of conclusions across multiple thought chains using knowledge graphs. This includes coordinating assumptions and conclusions across multiple chains, such as unifying market parameters and adjusting related inferences. Furthermore, in cross-chain anomaly repair, consensus can be reached through voting mechanisms or maximum likelihood estimation, without limitation. This approach can resolve multi-chain contradictions and improve the overall consistency of the dataset.

[0064] In this embodiment, a differentiated repair strategy can be adopted according to the type of anomaly to achieve accurate repair.

[0065] In some implementations, if there are multiple types of anomalous thought chains within the financial thought chain, each type of anomalous thought chain can be repaired according to priority. For example, the repair priority mechanism could be: prioritize local anomalies, then chain anomalies, and finally cross-chain anomalies. In specific implementations, the repair method can be optimized through reinforcement learning. For instance, the strategy in reinforcement learning can be set to prioritize repairing anomalous thought chains that have a significant impact on prediction accuracy, thereby improving the repair effect.

[0066] In one possible implementation, the method further includes: recording the anomaly rate decline curve after the anomaly thought chain is repaired, and dynamically adjusting the repair strategy based on the anomaly rate decline curve.

[0067] In this embodiment, the repaired financial thinking chain can be verified again, for example, by performing N iterations, where N is a natural number, such as 3–5, or a dynamically adjusted value. The iteration steps can be referred to as S101 to S104, and can be summarized as: consistency assessment → anomaly tracing → dynamic repair → verification. In each iteration, the anomaly rate decline curve can be recorded, and the repair strategy can be dynamically adjusted based on the anomaly rate decline curve.

[0068] For example, if the anomaly rate decline curve indicates that the anomaly repair criteria are not met, the repair process continues, and the parameters of each anomaly repair thought chain are adjusted until the anomaly rate decline curve indicates that the anomaly repair criteria are met. This can include adjusting the weights of α, β, and γ, or the repair priority, without limitation.

[0069] In this way, by repeatedly performing consistency assessment, anomaly tracing, repair, and verification processes, data quality is gradually improved through multiple iterations. Each iteration aims to reduce potential anomalies and optimize repair results until the theoretically preset quality standard is reached.

[0070] In one possible implementation, the repair strategy is dynamically adjusted based on the anomaly rate decline curve, including: determining that the cleaning of the financial thought chain has reached a preset quality standard when the anomaly rate decline curve decays exponentially.

[0071] In this embodiment, the exponential decay of the anomaly rate decline curve indicates that the cleaning of the financial thought chain has reached a preset quality standard. Of course, in addition to exponential decay, any other method can be used to determine whether the anomaly rate decline curve indicates that the anomaly repair indicator is met, such as the anomaly rate decreasing from 30% to below 1%, etc., without limitation.

[0072] By recording the anomaly rate decline curve in each iteration and dynamically adjusting the repair strategy, we can ensure that the anomaly rate gradually converges and improve data quality.

[0073] In one possible implementation, verifying the terminology standardization of the surface semantic layer includes: using a natural language processing (NLP) parser and regular expressions to verify the terminology consistency and grammatical correctness of the surface semantic layer.

[0074] For example, an NLP parser can be used to segment and identify entities in the surface semantic layer of text, extract financial terms, and then build a financial terminology dictionary. The dictionary can contain synonyms, standard expressions, etc. Regular expressions can be used to match the terms in the text to check for non-standard or confusing terms.

[0075] In NLP, when processing text, the parser can perform syntactic analysis on sentences, identifying whether subject-verb-object structures, tenses, and quantifier collocations conform to grammatical rules. It uses regular expressions to specifically validate the formatting unique to financial texts, automatically marking non-compliant formats and thus obtaining a validation score. This improves the accuracy of validating the terminology standardization at the surface semantic layer.

[0076] In one possible implementation, verifying the rationality of the formulas in the logical reasoning layer includes: constructing a directed graph of reasoning through graph structure modeling, and verifying the correctness of the formulas and the rationality of the assumptions in the logical reasoning layer.

[0077] For example, the key information of the logical reasoning layer can first be decomposed into graph components, clarifying the meaning and attributes of each component. Then, based on the attributes of each component, the components are divided into nodes and directed edges. Nodes are used to record node attributes, and directed edges are used to define the logical relationships between nodes and clarify the reasoning flow.

[0078] Furthermore, based on the aforementioned component definitions, the reasoning process can be transformed into a structured directed graph through element association, edge connection, and hierarchical sorting. In element association, the text is traversed to identify facts, formulas, and assumptions, corresponding nodes are created, and attributes are populated. In edge connection, directed edges are added to nodes according to the reasoning logic. In hierarchical sorting, nodes are layered according to the order of reasoning, ensuring that directed edges only point from lower to higher levels.

[0079] Graph structures enable automatic validation. For example, they can check if the input edges of formula nodes cover all necessary parameters, call the formula parsing engine, substitute the specific values ​​of the input nodes, verify whether the formula calculation result matches the true value, and check whether the units of the input parameters match the units required by the formula based on the unit information in the node attributes. This allows for accurate validation of the logical reasoning layer.

[0080] The method of the embodiments of this application has been described above. The following is in conjunction with… Figure 2 The following description explains the data cleaning devices that can be adapted to the financial thinking chain data cleaning method provided in the embodiments of this application.

[0081] like Figure 2 As shown, the data cleaning device may include a data parsing module, a consistency assessment module, an anomaly tracing module, a dynamic repair module, a cross-chain verification module, an optimization control module, and an output module.

[0082] The data parsing module is responsible for decomposing the financial thought chain into a surface semantic layer, a logical reasoning layer, and a financial semantic layer; the consistency assessment module is used to perform multi-level quality analysis; the anomaly tracing module is used to identify the source and scope of anomalies; the dynamic repair module is used to implement intelligent repair based on the financial context; the cross-chain verification module is used to verify the logical consistency between chains; the optimization control module is used to manage the iteration process and ensure convergence; and the output module is used to generate the cleaned dataset. Specific implementation details can be found in the description of the aforementioned embodiments and will not be elaborated further.

[0083] The following example illustrates the implementation of this application. For instance, the input dataset contains CoT data analyzing 1500 financial portfolios, covering return predictions and risk assessments for stocks, bonds, and derivatives. Each thought chain contains an average of eight reasoning steps, covering market assumptions (such as the risk-free rate and volatility), calculation processes (such as the Sharpe ratio), and conclusions. Initial analysis may reveal potential problems in approximately 30% of the chains.

[0084] Local anomalies (approximately 15%): For example, if a step assumes "annualized volatility of 20%" but incorrectly states "2%", it may lead to a deviation in risk assessment. Chain anomalies (approximately 10%): For example, calculating the Sharpe ratio and investment weights based on incorrect volatility may make the conclusions of the entire reasoning chain unreliable. Cross-chain anomalies (approximately 5%): For example, three chains of the same investment portfolio may predict "annualized return of 7%", "annualized return of 8%", and "annualized return of 6%" respectively, and the market assumptions (such as risk-free interest rates set at 2%, 3%, and 4% respectively) are inconsistent.

[0085] The data cleaning device can first perform layered analysis on the data: at the surface semantic layer, it may identify inconsistencies in the use of terminology, such as "volatility" and "standard deviation" being used interchangeably in different steps without a clear conversion relationship;

[0086] At the logical reasoning level, potential problems in the calculation process may be discovered, such as the Sharpe ratio not being correctly divided by volatility, leading to an abnormally amplified result; at the financial semantic level, some links in the chain may be detected as deviating from the target, such as introducing macroeconomic assumptions unrelated to the portfolio when predicting returns.

[0087] Furthermore, the data cleaning device traces the scope of the anomaly's impact. For example, an incorrect "2% volatility" might cause the Sharpe ratio to change from the theoretical value of 1.5 to an unreasonable 15, further affecting the derivation of investment weight allocation. The dynamic repair module may take the following measures:

[0088] For local anomalies, adjust "2%" to "20%" and ensure consistency in related statements. For chain anomalies, re-derive the Sharpe ratio and weight allocation based on the corrected volatility, for example, adjust the Sharpe ratio to 1.5 and the stock weight from the possible 80% to 60%. For cross-chain anomalies, unify the risk-free rate to a theoretically reasonable 3% and coordinate the forecast results of all chains to a consistent value (e.g., 7.5%), while adjusting related assumptions and procedures.

[0089] Through multiple rounds of iterative optimization, the data cleaning device can handle most local anomalies in rounds 1-2, reconstruct chain problems within the chain in rounds 3-4, and complete cross-chain coordination in rounds 5-6, ultimately reducing the potential anomaly rate of the dataset from the initial 30% to the theoretical 1%. The cleaned dataset is expected to improve the theoretical performance of large financial models in return prediction and risk assessment, such as reducing prediction bias or improving decision consistency.

[0090] Figure 3 This is a schematic diagram of the structure of the financial thinking chain data cleaning device provided in the embodiments of this application, as shown below. Figure 3 As shown in the figure, this application provides a financial thinking chain data cleaning device, which includes:

[0091] The segmentation module 301 is used to divide the financial thinking chain data into a surface semantic layer, a logical reasoning layer, and a financial semantic layer.

[0092] The verification module 302 is used to verify the standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions with financial objectives in the financial semantic layer, respectively.

[0093] The scoring module 303 is used to perform multi-level consistency evaluation on the verified data at each level and generate a weighted score.

[0094] Repair module 304 is used to identify abnormal thought chains in the financial thought chain based on weighted scoring and to repair abnormal thought chains.

[0095] In one possible implementation, the repair module 304 is specifically used for:

[0096] Local anomalies in abnormal thought chains are corrected by completing missing parameters based on regression models; and / or,

[0097] Using sequence generation models to reconstruct reasoning steps resolves chain anomalies in abnormal thought processes; and / or,

[0098] By using knowledge graphs to verify the consistency of conclusions across multiple thought chains, cross-chain contradictions in abnormal thought chains can be resolved.

[0099] In one possible implementation, an adjustment module is also included, which records the anomaly rate decline curve after the anomaly thought chain repair is completed, and dynamically adjusts the repair strategy based on the anomaly rate decline curve.

[0100] In one possible implementation, the adjustment module is specifically used to: determine whether the cleaning of the financial thought chain has reached a preset quality standard when the abnormality rate decline curve decays exponentially.

[0101] In one possible implementation, the verification module 302 is specifically used to: use a natural language processing (NLP) parser and regular expressions to verify the terminology consistency and grammatical correctness of the surface semantic layer.

[0102] In one possible implementation, the verification module 302 is specifically used to: construct a directed graph of reasoning through graph structure modeling, and verify the correctness of the formulas and the rationality of the assumptions in the logical reasoning layer.

[0103] In one possible implementation, the scoring module 303 is specifically used to: perform weighted calculations on the verification scores of the surface semantic layer, the logical reasoning layer, and the financial semantic layer to obtain a weighted score.

[0104] The financial thinking chain data cleaning device provided in this application embodiment can be used to execute the technical solution of the financial thinking chain data cleaning method in any of the above embodiments of this application. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0105] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 4As shown, the electronic device of this embodiment may include: at least one processor 401; and a memory 402 communicatively connected to the at least one processor; wherein the memory 402 stores instructions that can be executed by the at least one processor 401, and the instructions are executed by the at least one processor 401 to cause the electronic device to perform the method as described in any of the above embodiments.

[0106] Optionally, the memory 402 can be either standalone or integrated with the processor 401.

[0107] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0108] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method of any of the foregoing embodiments.

[0109] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the foregoing embodiments.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0111] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0112] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU) or other general-purpose processors. The processor can also be a Digital Signal Processor (DSP) or an Application Specific Integrated Circuit (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0113] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), disks or optical discs.

[0114] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Examples of storage media include Static Random-Access Memory (SRAM) or Electrically Erasable Programmable Read Only Memory (EEPROM).

[0115] Storage media can be, for example, erasable programmable read-only memory (EPROM) or programmable read-only memory (PROM). Storage media can also be read-only memory (ROM), magnetic storage, flash memory, magnetic disks, or optical disks. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0116] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside within an application-specific integrated circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components within an electronic device or host device.

[0117] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0118] The sequence numbers of the embodiments in this application are merely for description and do not represent the superiority or inferiority of the embodiments. Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0119] Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0120] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0121] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0122] It should be further noted that although the steps in the flowchart are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0123] Furthermore, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0124] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0125] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0126] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for cleaning up financial thought chains, characterized in that, The method includes: The data in the financial thinking chain is divided into a surface semantic layer, a logical reasoning layer, and a financial semantic layer. The terminology of the surface semantic layer, the formula rationality of the logical reasoning layer, and the alignment of the conclusions of the financial semantic layer with the financial objectives are verified respectively. Perform a multi-level consistency assessment on the verified data at each level to generate a weighted score; Abnormal thought chains in the financial thought chain are identified based on the weighted score, and the abnormal thought chains are repaired.

2. The method according to claim 1, characterized in that, The abnormal thought chain is repaired according to the repair strategy, which includes: The local anomalies in the abnormal thought chain are corrected by completing missing parameters based on the regression model; and / or, The abnormal chain of thought is resolved by reconstructing the reasoning steps using a sequence generation model; and / or, The cross-chain contradictions of the abnormal thought chain are resolved by verifying the consistency of conclusions from multiple thought chains using a knowledge graph.

3. The method according to claim 2, characterized in that, The method further includes: After the abnormal thought chain is repaired, the abnormality rate decrease curve is recorded, and the repair strategy is dynamically adjusted based on the abnormality rate decrease curve.

4. The method according to claim 3, characterized in that, The dynamic adjustment of the repair strategy based on the anomaly rate decline curve includes: When the anomaly rate decline curve decays exponentially, it is determined that the cleaning of the financial thought chain has reached the preset quality standard.

5. The method according to any one of claims 1-4, characterized in that, The verification of the terminology normativity of the surface semantic layer includes: Using a Natural Language Processing (NLP) parser and regular expressions, the terminology consistency and grammatical correctness of the surface semantic layer are verified.

6. The method according to any one of claims 1-4, characterized in that, Verifying the rationality of the logical reasoning layer formulas includes: A directed graph of reasoning is constructed by modeling a graph structure to verify the correctness of the formulas and the rationality of the assumptions in the logical reasoning layer.

7. The method according to any one of claims 1-4, characterized in that, The process of performing multi-level consistency evaluation on the verified data at each layer to generate a weighted score includes: The verification scores of the surface semantic layer, the logical reasoning layer, and the financial semantic layer are weighted and calculated to obtain the weighted score.

8. A financial thinking chain data cleaning device, characterized in that, The device includes: The segmentation module is used to divide the data of the financial thinking chain into a surface semantic layer, a logical reasoning layer, and a financial semantic layer. The verification module is used to verify the standardization of terminology in the surface semantic layer, the rationality of formulas in the logical reasoning layer, and the alignment of conclusions in the financial semantic layer with financial objectives. The scoring module is used to perform multi-level consistency evaluation on the verified data at each level and generate a weighted score. The repair module is used to identify abnormal thought chains in the financial thought chain based on the weighted score, and to repair the abnormal thought chains.

9. An electronic device, characterized in that, include: Memory and processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images