Financial multi-source protocol adaptive fusion system based on AI semantic understanding and knowledge graph

Abstract

The invention belongs to the technical field of financial data management, and particularly discloses a financial multi-source protocol self-adaptive fusion system based on AI semantic understanding and a knowledge graph, which comprises the steps of avoiding fusion errors caused by semantic misunderstanding through deep semantic analysis and a conflict resolution decision based on the knowledge graph; when protocol version updating is detected, incremental learning and model adjustment are carried out based on a newly added sample and historical experience, and a fusion protocol standard is dynamically updated, so that the high adaptive capacity to dynamic change of a financial protocol is realized; an exception monitoring and repairing mechanism is introduced, data missing, format errors and other exceptions are found in time and repaired online based on historical data modes and business logic, and negative influences of abnormal data on downstream risk control, transaction decision making and other key business links are avoided; through deep collaboration and information feedback among the intelligent modules, an intelligent system capable of self-learning and evolution is constructed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of financial data management, and relates to a financial multi-source protocol adaptive fusion system based on AI semantic understanding and a knowledge graph. BACKGROUND

[0002] With the acceleration of the globalization and informatization process of the financial market, various exchanges, clearing institutions and data service providers have developed diversified business protocol documents and technical specifications to ensure the standardization of business processes and the accuracy of data interaction. However, these protocols differ significantly in data format, field meaning, business logic, etc., bringing challenges of multi-source heterogeneity to the exchange and processing of financial data. Traditional methods rely on manual parsing and mapping of protocol fields, which is not only inefficient but also prone to errors, making it difficult to meet the high requirements of modern financial markets for data real-time, accuracy and consistency.

[0003] In the prior art, the processing of multi-source heterogeneous financial protocol data mainly relies on manually writing parsing scripts and maintaining static field mapping tables. Technical personnel need to manually read each exchange protocol document, identify the fields one by one and map them to the internal standard data model. For exceptions in the data stream, offline batch data cleaning programs are usually used to filter or fill according to pre-set simple rules. When the protocol version changes, the entire process needs to be manually analyzed and configured again, resulting in slow response and high maintenance cost. Although some automated tools try to parse the protocol, there are still problems such as inaccurate semantic understanding, inflexible mapping relationship, imperfect exception handling and protocol update lag.

[0004] Although the prior art has alleviated the pressure of processing multi-source heterogeneous protocol data to some extent, it still has the following shortcomings: first, the manual parsing and mapping process is tedious, time-consuming and prone to errors, making it difficult to accurately capture the subtle semantic differences of similar terms in different protocols, resulting in low data fusion quality. Second, static mapping tables cannot cope with the normality of frequent protocol updates, and a large amount of manpower is needed to maintain each change, affecting business agility. Finally, batch processing of exception detection and repair cannot meet the high timeliness requirements of financial businesses, and the repair means is single, which can easily cause information distortion or loss, reducing the value of data usability. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the application proposes the following technical solution: a financial multi-source protocol adaptive fusion system based on AI semantic understanding and a knowledge graph, comprising: a protocol parsing module: parsing the protocol documents and technical specifications of each exchange through a natural language processing model, extracting field names, business semantics, units of measurement, numerical precision and enumeration value meanings, and generating structured protocol metadata.

[0006] The graph construction module: based on the financial instrument identification, the exchange coding system and the business terminology definition, a domain knowledge graph is constructed, which comprises entity nodes, semantic constraint nodes and business logic constraint edges.

[0007] The semantic mapping module: according to the structured protocol metadata and the domain knowledge graph, the semantic similarity of cross-protocol fields is calculated, the initial field mapping relationship is generated, the field semantic conflict is detected, the conflict resolution strategy is executed based on the semantic constraints in the domain knowledge graph and the preset historical mapping statistical result library, and the corrected field mapping relationship is generated.

[0008] The exception repair module: real-time monitoring of the abnormal fields of the input data stream, automatic calling of the historical data mode to execute field repair, and output of the standard data stream.

[0009] The protocol updating module: detecting the protocol version change signal in the standard data stream, dynamically updating the natural language processing model and the conflict resolution strategy based on the newly added protocol document sample and the corrected field mapping relationship, and generating the fusion protocol standard.

[0010] Compared with the prior art, the beneficial effects of the present application are as follows: (1) the present application can accurately identify and process the subtle semantic differences and business logic conflicts between different protocols through deep semantic analysis and conflict resolution strategy based on knowledge graph, avoid fusion errors caused by semantic misunderstanding, ensure the semantic consistency and business compliance of the finally generated data, and significantly improve the accuracy and automation level of financial multi-source protocol data fusion.

[0011] (2) the present application can quickly and low-cost incremental learning and model adjustment based on a small number of new samples and historical experience when detecting protocol version update, dynamically updating the fusion protocol standard, eliminating the business interruption risk and high maintenance cost caused by large-scale manual reconfiguration of the traditional scheme after protocol change, and realizing the high self-adaptability to the dynamic changes of financial protocols, and ensuring the continuity and stability of business processing.

[0012] (3) the present application greatly improves the quality and availability of real-time data stream by introducing a streaming exception monitoring and intelligent repair mechanism, can instantly discover and repair data missing, format errors and other exceptions based on historical data mode and business logic, solve data quality problems at the source, avoid the negative impact of abnormal data on downstream risk control, transaction decision and other key business links, and improve the operation efficiency and decision reliability of the entire financial system.

[0013] (4) The application builds an intelligent system capable of self-learning and evolution through deep cooperation and information feedback between intelligent modules, for example, the conflict resolution experience of artificial intervention can enhance the knowledge graph and decision model in reverse, and the practice of abnormal repair can optimize the historical data mode library. This continuous self-improvement mechanism enables the overall performance of the system to be continuously improved over time, realizing the leap from static processing to dynamic evolution. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0015] Figure 1 It is a schematic diagram of the system module connection of the application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0017] Please refer to Figure 1 The financial multi-source protocol adaptive fusion system based on AI semantic understanding and knowledge graph proposed in the application includes a protocol analysis module, a graph construction module, a semantic mapping module, an abnormal repair module and a protocol update module.

[0018] The connection relationship between the modules is that the protocol analysis module is connected with the graph construction module, responsible for providing structured protocol metadata as the basis for knowledge graph construction; the graph construction module is connected with the semantic mapping module, providing domain knowledge support for semantic mapping and conflict resolution; the semantic mapping module is connected with the abnormal repair module and the protocol update module, transmitting the corrected field mapping relationship to guide abnormal repair and dynamic protocol update; the abnormal repair module is connected with the protocol update module, real-time monitoring and repairing data flow anomalies, and feeding back the repair verification results to optimize the fusion protocol standard.

[0019] The protocol analysis module: through the natural language processing model, the protocol documents and technical specifications of each trading place are analyzed, the field name, business semantics, unit of measurement, numerical precision and enumeration value meaning are extracted, and structured protocol metadata is generated.

[0020] The field name is a label or name used in the protocol document to identify a specific data item. It is usually a short string used to uniquely reference the data item during data exchange or processing.

[0021] The business semantics refer to the specific business meaning or concept represented by the field, which is used to explain the actual role and significance of the field in financial transactions or business processes. For example, "transaction amount" may appear in different agreements, but its specific business semantics (such as whether it includes taxes and fees, whether it is net value, etc.) may be different.

[0022] The unit of measurement is a standard unit used to quantify the value of a field. In the financial field, common units of measurement include currency units (such as yuan, US dollars) and quantity units (such as shares, lots).

[0023] The numerical precision refers to the number of decimal places or significant digits in the field value representation, used to determine the accuracy and range of the value.

[0024] The meaning of the enumeration value refers to the explanation or description of all the predefined values ​​that the field may take in the protocol. Enumeration values ​​are usually used to represent discrete states, types or options, such as transaction direction (buy, sell), order status (not executed, partially executed, fully executed), etc.

[0025] The graph construction module constructs a domain knowledge graph containing entity nodes, semantic constraint nodes, and business logic constraint edges, based on financial instrument identifiers, exchange coding systems, and business terminology definitions.

[0026] In a preferred embodiment, the construction of a domain knowledge graph comprising entity nodes, semantic constraint nodes, and business logic constraint edges includes: collecting and standardizing financial instrument identifiers from various financial markets, such as stock codes or International Securities Identifiers (ISINs), integrating coding systems adopted by different exchanges, such as Market Identifiers (MICs), and introducing industry-recognized business terminology definitions, such as terminology libraries from the FIX protocol, integrating them into entity nodes, and then automatically creating connection edges between them through preset ontology rules, such as "financial instruments" being "traded" on "exchanges," thereby forming an entity relationship network.

[0027] Embed the parsed business semantics and units of measurement from the structured protocol metadata to generate semantic constraint nodes.

[0028] For example, for a "transaction amount" field extracted from an agreement document, if its business semantics are parsed as "total price of a single transaction" with the unit of measurement in "yuan", the system will create a semantic constraint node. This node encapsulates the two constraints: "meaning total price of a single transaction" and "unit is yuan", and connects this node to the entity node of the "transaction amount" field. This step makes the semantic constraints, which were originally implicit in the document, explicit and solidifies them into the graph structure.

[0029] Integrate the state transition logic and numerical boundary conditions from the pre-set financial business rule base to form business logic constraint edges.

[0030] The financial business rule base is used to store various logical rules in the financial transaction process. For example, for a state transition logic where "the order status can change from unexecuted to partially executed or fully executed", the system will establish directed business logic constraint edges between the entity nodes representing the three states of "unexecuted", "partially executed", and "fully executed". Similarly, for a numerical boundary condition where "the order price must be greater than zero", the system will create a business logic constraint edge connecting the "order price" field node to a logical condition node representing "greater than zero".

[0031] By executing the above steps in sequence, a domain knowledge graph that integrates entity relationship networks, semantic constraint nodes, and business logic constraint edges is constructed.

[0032] The domain knowledge graph constructed by the above-described method not only builds an entity relationship network containing financial entities and their static relationships, but also, by embedding semantic constraint nodes, accurately and machine-readablely anchors business semantics from different sources with subtle differences. Furthermore, by integrating business logic constraint edges, dynamic business process logic and numerical verification conditions are incorporated into the graph's topology. This fusion of structured, semantic, and logical elements makes the knowledge graph a powerful reasoning foundation, enabling the system to perform deep-level association analysis and logical judgments in subsequent processing. For example, it can automatically infer that fields in two different protocols not only have similar names, but also that their business constraints and the context of their respective business processes are consistent, thus achieving high-confidence semantic equivalence judgments. This provides solid and comprehensive knowledge support for subsequent intelligent mapping and conflict resolution, realizing a synergistic enhancement effect of knowledge.

[0033] The semantic mapping module performs cross-protocol field semantic similarity calculation based on structured protocol metadata and domain knowledge graph, generates initial field mapping relationships, detects field semantic conflicts, performs conflict resolution decisions based on semantic constraints in the domain knowledge graph and a preset historical mapping statistical result library, and generates corrected field mapping relationships.

[0034] In a preferred embodiment, the step of calculating cross-protocol field semantic similarity based on structured protocol metadata and domain knowledge graph to generate initial field mapping relationships includes: extracting structured protocol metadata as input features for similarity calculation.

[0035] By using a pre-trained natural language processing model, the input features are vectorized to generate low-dimensional dense vectors of the fields.

[0036] Pre-trained natural language processing models such as BERT, Word2Vec, and GloVe can decompose text into subwords or lexical units and capture contextual dependencies (e.g., "amount" in a financial context may refer to "transaction amount" rather than a general quantity). They map textual data (such as field names and business semantic descriptions) to a low-dimensional dense space through multi-layered neural networks, generating fixed-dimensional feature vectors (such as BERT's 768-dimensional or 384-dimensional output). Because natural language processing models can capture contextual dependencies, the generated vectors contain the contextual information of the fields within the agreement document. This representation helps computers better understand and process textual information.

[0037] The vectorized representation can convert text data into numerical form.

[0038] The low-dimensional dense vector is a compact representation generated by mapping textual field features (such as field names and business semantics) to a continuous numerical space through a natural language processing model. Each dimension is a non-zero floating-point number to carry semantic information, such as transaction type, amount, and time attribute, and supports quantifying the semantic similarity between fields through vector distance.

[0039] For example, there is a simplified 3-dimensional vector to represent the "transaction amount" field (for ease of illustration, the actual dimensions are higher): [0.12, -0.45, 0.78].

[0040] The first dimension (0.12) represents the semantic features of this field related to “transaction size”.

[0041] The second dimension (-0.45) reflects the negative correlation between this field and "fee inclusion status" (such as whether taxes are included).

[0042] The third dimension (0.78) captures the importance of this field in the "high-frequency trading" scenario.

[0043] Based on the entity relationship network and semantic constraint nodes in the domain knowledge graph, the cosine similarity between low-dimensional dense vectors of fields is calculated as a quantitative indicator of semantic similarity. These quantitative indicators help the system to more accurately identify similar or equivalent fields in different protocols, thereby generating more accurate field mapping relationships.

[0044] The formula for calculating cosine similarity is: ,in and These represent low-dimensional dense vectors of the two fields, Representing vectors and The dot product (inner product) of . and Representing vectors respectively and The norm (length) of the cosine similarity ranges from [-1, 1], and the closer the value is to 1, the more semantically similar the meaning.

[0045] Example: The "Transaction Amount" field parsed from the protocol document of Exchange A is converted into a low-dimensional dense vector by a natural language processing model and used as the vector representation of field A; the "Total Transaction Amount" field parsed from the protocol document of Exchange B, or a field with semantic similarity to "Transaction Amount" that already exists in the domain knowledge graph, is also converted by the model to obtain a vector representation as field B; the vector cosine similarity between field A ("Transaction Amount") and field B ("Total Transaction Amount") is 0.92, indicating a high degree of similarity.

[0046] Based on the semantic similarity ranking among all low-dimensional dense vectors of all fields, the system automatically recommends field pairs with semantic similarity no lower than a preset similarity threshold, such as mapping “trade_amount” of protocol X to “transaction_value” of protocol Y, thereby generating the initial field mapping relationship.

[0047] In a further preferred embodiment, the detection of semantic conflicts in the fields is based on semantic constraints in the domain knowledge graph and a preset historical mapping statistical result library to perform conflict resolution decisions, including: defining that there is a semantic conflict when the semantic similarity of two cross-protocol fields is lower than a preset similarity threshold, and then identifying the type of semantic conflict in the fields, including unit mismatch, enumeration value ambiguity, or business rule contradiction.

[0048] Specifically, the preset semantic similarity threshold is adjusted according to the actual application scenario and needs to ensure that true semantic differences can be captured while avoiding too many false alarms.

[0049] When identifying the types of semantic conflicts between fields, the structured protocol metadata of a pair of fields with mapping conflicts are compared and analyzed: by comparing the unit of measurement attributes of the two, unit mismatch conflicts can be identified; by comparing the meaning definitions of their enumeration values, enumeration value ambiguity conflicts can be identified; by verifying the attributes of the field to be mapped with the business logic constraint edges associated with the target field in the domain knowledge graph, business rule contradictions can be identified.

[0050] Starting with the entity node corresponding to the semantically conflicting field in the domain knowledge graph, the query retrieves the associated semantic transformation path and business logic constraints in the domain knowledge graph.

[0051] Specifically, the semantic transformation path refers to the path connecting two entity nodes with different semantics but representing similar or equivalent concepts in a domain knowledge graph. It typically includes the transformation factors, formulas, or logical steps required to transform the semantics of one entity into the semantics of another entity.

[0052] The business logic constraints refer to the restrictions on the relationships between entity nodes or the attribute values ​​of entity nodes in the domain knowledge graph. These constraints reflect the actual business rules in the financial transaction process, such as state transition logic and numerical boundary conditions.

[0053] For unit mismatch conflicts, a semantic transformation path connecting the two unit nodes is sought, which typically includes a transformation factor or formula. For enumeration value ambiguity conflicts, semantic equivalence relations or mapping rules associated with the enumeration value field in the domain knowledge graph are queried. These relations or rules may exist in the form of synonyms, equivalence classes, or transformation tables to determine whether there is a feasible path to accurately map the enumeration value of one protocol to the enumeration value of another protocol. For business rule contradictions, all business logic constraints related to the target field are extracted, such as numerical ranges or state preconditions, to assess whether there is feasible correction logic.

[0054] Access the preset historical mapping statistics database, and match the distribution of resolution schemes for the same conflict type in the historical mapping statistics database based on the current conflict type, the semantic transformation path involved, and business logic constraints. This distribution quantifies the success probability of various solutions in history.

[0055] The historical mapping statistics database is used to record all past semantic conflict events in fields, as well as the final resolution schemes and success rates adopted. For example, 80% of unit mismatch conflicts are resolved through currency conversion.

[0056] The system integrates deterministic data from the domain knowledge graph and empirical data from the historical mapping statistics database to calculate the confidence level of each potential resolution solution. The highest confidence level is then compared with a preset confidence threshold. When the highest confidence level exceeds the preset confidence threshold, the corresponding transformation logic is automatically executed to complete the conflict resolution. Otherwise, it indicates that the automated processing is at high risk, triggering a manual intervention instruction. This instruction includes conflict details, candidate solutions, and their confidence level analysis for human decision-making.

[0057] The deterministic data from the domain knowledge graph includes semantic constraint nodes, business logic constraint edges, and entity relationship networks. The financial business rules, semantic constraints, and entity relationships are explicitly encoded in a structured manner, possessing clear logic and interpretability.

[0058] The historical mapping statistics database contains empirical data that records past conflict resolution cases and their effects, reflecting success patterns and probabilities in actual business scenarios. This includes conflict type distribution, resolution success rates, and contextual characteristics. For example, unit mismatch conflicts account for 60% of all conflicts, with 80% resolved through currency conversion; Solution A has a 92% success rate in similar conflicts, while Solution B has only 45%; in the "CSI 300 Index Futures" agreement, Solution C has a 30% higher success rate than other scenarios.

[0059] Confidence level of the resolution scheme The calculation formula is: ,in This represents the support level of the knowledge graph. If there are direct, unambiguous semantic transformation paths or business logic constraints in the domain knowledge graph to support the solution, then... The value is 1 if it is set to 1, and 0 otherwise. Represents the historical success probability, and its value is obtained by matching from the historical mapping statistics database, with a range of 0 to 1; and These are preset weighting coefficients, representing the relative importance of knowledge and experience in decision-making, and .

[0060] This invention combines the deterministic business logic provided by a domain knowledge graph with practical experience from historical mapping statistics. It enables precise and reliable automatic repair of conflicts with clearly defined rules, while in scenarios with ambiguous or missing rules, it draws on historical success stories to make the most probable intelligent judgment. This synergy of knowledge and experience, along with the intelligent triage of automation and human intervention based on confidence levels, significantly improves the accuracy and efficiency of conflict resolution, ensures the quality of data fusion, and minimizes the need for human intervention, achieving a high degree of intelligence and robustness in the data processing workflow.

[0061] In a further preferred embodiment, the triggering of the manual intervention instruction includes: recording the final conflict resolution scheme confirmed and executed by the manual intervention instruction, and updating the historical mapping statistics database.

[0062] Specifically, when a semantic conflict in a field triggers a manual intervention instruction because its confidence level is lower than a preset confidence threshold, the system will record in detail the final conflict resolution solution confirmed and executed by the domain expert. This record contains the complete context of the conflict and the solution that has been verified as correct, and is immediately used to update the system's historical mapping statistics database, such as "convert the 'price' field of Agreement A from 'yuan' to 'dollar'".

[0063] When a similar conflict occurs again, increase the confidence weight of the corresponding resolution solution.

[0064] Specifically, the updating of the historical mapping statistical results database directly affects the future decision-making process. When the same or similar types of conflicts occur in the future, the confidence weight of the resolution scheme will be increased because there are successful precedents in the historical mapping statistical results database, thereby increasing the probability that it will be automatically selected.

[0065] Natural language processing models are used to perform syntactic and semantic analysis on manually verified conflict resolution schemes, extracting semantic transformation paths and business logic constraints. This includes the specific steps or formulas for achieving semantic equivalence conversion of fields between different protocols, as well as the business rules, state transition logic, and numerical boundary conditions to be followed during the conversion process. If the scheme has fixed business logic or semantic transformation paths, such as a new unit conversion standard (e.g., converting from "yuan" to "dollar"), the system will identify this conversion path and the business logic behind it. The verified conversion logic will then be synchronized to the domain knowledge graph in a structured form. This is manifested by creating new semantic constraint nodes in the knowledge graph to explicitly express the conversion relationship between different semantic entities; or by modifying existing business logic constraint edges to update and optimize the business logic constraint conditions.

[0066] This invention, by recording, quantifying, and ultimately solidifying the decision-making knowledge of human experts, continuously transforms tacit expert experience into explicit, reusable machine knowledge. This not only creates a virtuous cycle, gradually reducing the number of complex scenarios requiring human intervention over time, but also continuously enriches and improves the depth and breadth of the domain knowledge graph. Ultimately, this mechanism enables the system's automation boundaries to dynamically expand, achieving continuous evolution from assisted decision-making to autonomous decision-making capabilities, significantly enhancing the overall system's intelligence level and long-term effectiveness.

[0067] The anomaly repair module monitors abnormal fields in the input data stream in real time, automatically calls historical data patterns to perform field repair, and outputs a standard data stream.

[0068] The real-time input data streams primarily refer to the financial transaction data streams continuously received from various exchanges, clearinghouses, and data service providers. These data streams contain real-time information on various trading activities in the financial market, such as stock trading, futures trading, and foreign exchange exchange. Specific objects include, but are not limited to, the following: Transaction instruction data: including detailed information on buy / sell orders, such as the trading instrument, trading quantity, trading price, and trading direction (buy or sell); Market data: covering real-time quotes for various financial products, including the latest transaction price, best bid price, best ask price, highest price, lowest price, etc.; Order status data: recording the current status of orders, such as unexecuted, partially executed, fully executed, cancelled, etc.

[0069] In a preferred embodiment, the real-time monitoring of abnormal fields in the input data stream and the automatic invocation of historical data patterns to perform field repair includes: defining an abnormal field as one whose value in the data stream is missing, out of bounds, or has an abnormal format; extracting the relevant field value sequence within the adjacent time window in the real-time data stream; and generating a spatiotemporal feature matrix.

[0070] Specifically, the adjacent time window refers to a set containing the abnormal field itself and several consecutive fields before and after it. The system not only extracts the numerical sequence of the abnormal field itself, but also simultaneously extracts the sequences of other fields with high business relevance to that field. This relevance information is predefined in the domain knowledge graph. These numerical values ​​from multiple fields spanning multiple time points are then integrated to generate a two-dimensional spatiotemporal feature matrix. The rows of this matrix represent time steps, and the columns represent different business fields, thereby capturing the local dynamic patterns of the data.

[0071] Synchronously query the compliant value range associated with the abnormal field in the domain knowledge graph. The compliant value range is defined by the constraints of the field's value or format when constructing the knowledge graph, based on the protocol metadata and business logic. These constraints define the legal boundaries of the field's value or format, such as upper and lower limits of the value or regular expression template.

[0072] The spatiotemporal feature matrix is ​​used as the main input, and the compliant value range is used as a constraint verification for the repair operation: if the detected abnormal field is a missing field, the mathematical model in the sequence interpolation algorithm is started to calculate an estimated value to fill in the missing field.

[0073] The mathematical model in the sequence interpolation algorithm refers to a set of statistical or machine learning methods based on spatiotemporal feature analysis. It takes the spatiotemporal feature matrix as input and its core objective is to generate missing value estimates that conform to business logic by mining the numerical patterns of time windows before and after abnormal fields in the data stream.

[0074] If the detected abnormal field is a value that is out of bounds, the abnormal handling strategy in the associated business logic constraint edge will execute the repair process.

[0075] Specifically, the system locates the entity node corresponding to the out-of-bounds field in the domain knowledge graph and retrieves the encapsulated exception handling strategy along the business logic constraint edge connected to that node. This business logic constraint edge explicitly defines the corrective measures to be taken when a specific field's value goes out of bounds. For example, for an out-of-bounds price field, the strategy might be to clamp it to the nearest dynamic boundary value; while for an abnormally high value in the trading volume field, the strategy might be to trigger a more complex historical data pattern comparison and correction. The system then strictly executes the retrieved exception handling strategy to intelligently correct the out-of-bounds value.

[0076] If the detected abnormal field is a format error, a preset regular expression template is invoked, and this template is then applied to the field value with the incorrect format. By matching and recombining characters, the field is corrected to a compliant format. The regular expression template defines the correct text format that the field should follow, and is obtained from the domain knowledge graph or protocol metadata.

[0077] The final result of all repair operations must meet the compliance value range constraints obtained from the domain knowledge graph, ensuring that the repaired data is not only complete in form but also conforms to business logic in content.

[0078] This invention constructs a spatiotemporal feature matrix, leveraging the continuity of data over time and its relevance to business operations. This ensures that the repair results closely align with the immediate context and local dynamic trends of the data. More importantly, by introducing a compliant value range provided by a domain knowledge graph, it ensures that the data repair process always operates within the framework of business logic. This deep synergy between a data-driven historical model and knowledge-driven business logic guarantees that the repaired data not only fills gaps or corrects formats but also maintains a high degree of numerical authenticity and logical correctness, thereby significantly improving the overall quality and credibility of the output standard data stream.

[0079] In a further preferred embodiment, the definition of an abnormal field is defined when the value of a certain field in the data stream is missing, out of bounds, or has an abnormal format. This includes: checking the data packet for missing fields in real time through the list of required fields defined by the structured protocol metadata. For example, when the "transaction price" field in the transaction report is empty, a completion mechanism is triggered.

[0080] The list of required fields refers to the set of fields that are explicitly marked in the structured protocol metadata and must be included in data interaction.

[0081] Based on the definition of numerical precision in the structured protocol metadata and the context information extracted from the real-time input data stream, dynamic value domain boundary conditions are generated. Then, a value in a field in the real-time input data stream that exceeds the dynamic value domain boundary conditions is defined as a numerical out-of-bounds condition. The numerical precision definition includes the static format requirements of the field (specifying the number of decimal places that certain numerical fields should retain, such as retaining two decimal places for monetary data and using cents as the unit) and business logic constraints.

[0082] For example, for a stock price field, its dynamic range boundary conditions are not only greater than zero, but its upper and lower limits may also depend on the closing price of the trading instrument on the previous trading day and the preset price fluctuation limits. Therefore, its boundary conditions change dynamically with the market context, rather than remaining fixed.

[0083] By comparing the real-time data stream with the protocol format templates stored in the domain knowledge graph, abnormal field formats can be identified, including but not limited to format type mismatch (such as a field that should transmit an integer transaction amount containing a decimal) and out-of-bounds enumeration values ​​(such as an undefined "pending confirmation" status in the order status field).

[0084] This invention generates dynamic value boundary conditions, enabling the monitoring mechanism to accurately adapt to the rapidly changing business scenarios of the financial market. This significantly reduces false positives and false negatives caused by the use of static, rigid rules, thus improving the effectiveness of anomaly detection. More importantly, when a value is detected to be out of bounds, the system performs repair by associating it with business logic constraints in the domain knowledge graph. This ensures that the repair action itself conforms to the specific business logic of that field, rather than being a blind or generic process. This close collaboration between dynamic detection and knowledge-driven repair makes data quality assurance no longer an isolated technical operation, but an intelligent self-healing process integrated into the business logic. This ensures that the final output data is not only complete in form but also possesses a high degree of business rationality and logical consistency in content.

[0085] The protocol update module detects protocol version change signals in the standard data stream, dynamically updates the natural language processing model and conflict resolution decisions based on newly added protocol document samples and the corrected field mapping relationship, and generates a fusion protocol standard.

[0086] In a preferred embodiment, the protocol version change signal is detected by crawling the version identifier of the exchange protocol release page in real time. The version identifier includes, but is not limited to, version number, release date, or file hash value.

[0087] Specifically, the core of this mechanism is a continuously running web crawling task configured to periodically access the pages on the official websites or developer portals of various target financial exchanges that specifically publish technical specifications and protocol documents. On each visit, the task uses preset web element location rules, such as XPath or CSS selectors, to identify and extract specific text or metadata on the page that identifies the protocol version—the version identifier. The system maintains a local database that records the identifiers of the latest versions of all currently known protocols.

[0088] After successfully retrieving the page version identifier each time, it is compared with the corresponding protocol version identifier stored in the local database. If the two do not match, it is determined that the protocol version has changed, and an internal protocol version change signal is immediately generated. The generation of this signal will trigger subsequent automated processing.

[0089] This invention transforms passive notification reception into proactive real-time crawling. This method ensures the system can capture any protocol version updates immediately, significantly reducing the delay from protocol release to system awareness and preventing data processing errors or interruptions caused by information lag. Furthermore, by automatically generating version difference summaries, this mechanism can quickly focus on the core changes, providing precise guidance for subsequent intelligent update processes and reducing the enormous workload of manually analyzing new documents. This proactive and accurate change detection capability is a key prerequisite for ensuring the entire intelligent fusion system can keep pace with the market and achieve seamless dynamic adaptation, thereby ensuring the system's long-term stability and reliability in the ever-changing financial environment.

[0090] In a further preferred embodiment, the step of dynamically updating the natural language processing model and conflict resolution decision based on the newly added protocol document samples and the corrected field mapping relationship to generate a fusion protocol standard includes: inputting the newly added protocol document samples into the incremental training unit and adjusting the semantic parsing weights of the natural language processing model.

[0091] The incremental training unit does not completely reconstruct the original natural language processing model. Instead, it uses transfer learning or fine-tuning techniques to adjust the internal parameters and semantic parsing weights of the natural language processing model by utilizing the new terms, changed semantic descriptions, or format specifications contained in the new protocol document samples. This step allows the model to quickly grasp the characteristics of the new protocol version without consuming a large amount of resources for full retraining.

[0092] When a protocol version change signal is detected, the system automatically downloads the complete protocol document associated with the new version identifier according to the preset download link rules, in order to obtain the structured protocol metadata of the newly added protocol document sample.

[0093] The preset download link rules refer to a set of predefined logic or criteria within the system, used to guide the system on how to automatically locate and download the complete protocol document associated with the new version identifier. These rules ensure that when the system detects a protocol version change, it can quickly and accurately obtain the latest protocol document, providing a basis for subsequent processing and analysis.

[0094] For example, the target URL construction rule is to dynamically construct a specific download URL pointing to the new version of the protocol document based on the known URL structure of the exchange's official website or developer portal, combined with the protocol version identifier (such as version number, release date, or file hash value).

[0095] Download source priority settings: For the same protocol version, there may be multiple download sources (such as the main site, mirror sites, CDN, etc.). The preset download link rules may include download source priority settings to ensure that the system prioritizes downloading protocol documents from reliable and high-speed sources, thereby improving download efficiency and success rate.

[0096] After downloading, the structured protocol metadata of the two protocol versions is compared, and a version difference summary containing the main changes is automatically extracted and generated. This summary aims to quickly locate the core content of the protocol changes, generate richer low-dimensional dense vectors, and provide accurate input for subsequent dynamic updates to optimize the calculation of cross-protocol field semantic similarity.

[0097] The key changes include, but are not limited to, the addition and deletion of fields, changes in field meaning and business semantics, and adjustments to units of measurement and numerical precision. These changes are used as inputs to optimize semantic similarity calculations, enabling more accurate assessment of the correlation between fields in the old and new protocols.

[0098] By merging all historical conflict resolution records and incorporating the characteristics of the new protocol, the priority ranking of conflict resolution decision rules is restructured. For example, if the new protocol clearly identifies a unit that has caused ambiguity, the priority of the conflict resolution decision rules related to that unit will be increased, thereby optimizing the entire decision-making logic.

[0099] This invention, through incremental learning and dynamic optimization, can rapidly absorb new knowledge and adapt to new rules without interrupting core business operations, ensuring the real-time performance and accuracy of the converged protocol standard. This dynamic adaptive mechanism not only significantly reduces the long-term maintenance costs of the system but also guarantees business continuity and data processing reliability in dynamic market environments, giving the entire intelligent converged system sustained vitality and competitiveness.

[0100] In a further preferred embodiment, the generation of the fusion protocol standard further includes: continuously monitoring the output standard data stream, automatically extracting and structuring all executed anomaly repair records from it, and feeding them back to the historical data pattern library.

[0101] Specifically, the historical data pattern library is used to store all executed anomaly repair records. Each anomaly repair record contains detailed information on the anomaly type (missing field, out-of-bounds value, format anomaly), the context data at the time of occurrence, the specific repair strategy adopted, and the result after repair (such as missing fields being filled in using a sequence interpolation algorithm). This is used to enrich and update the knowledge in the library about anomaly patterns and their effective countermeasures.

[0102] When the same anomaly is detected again in the continuously monitored real-time data stream, the system automatically queries the dynamically updated historical data pattern library. By analyzing the success rate and application frequency of various repair strategies in the historical data pattern library, the system dynamically strengthens the priority weight of the corresponding repair strategy, so that it will be given priority in future automated decision-making.

[0103] Specifically, the success rate of the repair strategy = (number of successful repairs with the corresponding repair strategy / total number of calls) × 100%; the application frequency of the repair strategy = α × success rate of the corresponding repair strategy + β × recent call weight, where α and β refer to the preset weight coefficients of success rate and recent call weight, α + β = 1, usually set α = 0.7 and β = 0.3 to highlight the dominance of success rate, and recent call weight refers to the call frequency within the most recent set time (such as the proportion of the number of calls in the past 1 hour to the total number of calls).

[0104] Combining the success rate of the repair strategy and the weight of recent calls, the following formula is used to determine the dynamically adjusted call priority weight of the repair strategy. : ,in and The preset weighting coefficients satisfy... ,Will Set it to 0.7 to highlight the dominance of success rate. Set to 0.3 to take into account timeliness; the specific value can be adjusted according to actual needs. This represents the proportion of priority impact after normalization, mapping the success rate to the range of 0 to 1. This refers to the priority influence ratio that maps the application frequency to the range of 0 to 1 after normalization processing; the normalization processing may use the Sigmoid function. Example: According to the calculation results, the priority weight of repair strategy A (0.69) is higher than that of repair strategy B (0.65). Therefore, in future automated decision-making, the system will prioritize repair strategy A.

[0105] The repair verification results will be synchronously updated to the business logic constraint edges of the domain knowledge graph.

[0106] Specifically, the system continuously verifies the repair results, such as by verifying the final effect through feedback from downstream systems or logical consistency checks with subsequent data. For repair verification results that are verified to be effective in the long term, the system extracts them into more fundamental business logic. For example, if the value of a certain field is frequently corrected to a value outside the original definition range but verified as correct, the system considers this to reflect an actual change in business logic and automatically updates this repair verification result to the domain knowledge graph. This is manifested by adjusting or adding a related business logic constraint edge, thereby solidifying the knowledge learned from practice into the system's core knowledge system.

[0107] This invention enables the system to self-optimize its repair strategies by feeding back anomaly repair records to a historical data pattern library, making its anomaly handling capabilities more accurate and efficient over time. Furthermore, by synchronizing validated repair results to a domain knowledge graph, the system achieves a deeper cognitive leap; it not only learns how to better solve problems but also how to revise and refine its definition of problems. This two-way feedback and co-evolution between repair practices and the knowledge system constructs a powerful virtuous cycle of self-healing and self-reinforcing, ensuring the system maintains long-term adaptability and robustness in the face of constantly changing data and business environments, ultimately achieving continuous endogenous growth in data quality assurance capabilities.

[0108] It should be noted that the above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A financial multi-source protocol adaptive fusion system based on AI semantic understanding and knowledge graph, characterized in that, include: Protocol parsing module: Parses the protocol documents and technical specifications of various exchanges using a natural language processing model, extracts field names, business semantics, units of measurement, numerical precision and the meaning of enumeration values, and generates structured protocol metadata; Graph construction module: Based on financial instrument identifiers, exchange coding systems, and business terminology definitions, construct a domain knowledge graph containing entity nodes, semantic constraint nodes, and business logic constraint edges; Semantic mapping module: Calculates cross-protocol field semantic similarity based on structured protocol metadata and domain knowledge graph, generates initial field mapping relationship, detects field semantic conflicts, performs conflict resolution decision based on semantic constraints in the domain knowledge graph and a preset historical mapping statistical result library, and generates corrected field mapping relationship; Anomaly Repair Module: Monitors abnormal fields in the input data stream in real time, automatically calls historical data patterns to perform field repair, and outputs a standard data stream; Protocol update module: Detects protocol version change signals in the standard data stream, dynamically updates the natural language processing model and conflict resolution decisions based on newly added protocol document samples and the corrected field mapping relationship, and generates a fusion protocol standard.

2. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 1, characterized in that, The construction of the domain knowledge graph, which includes entity nodes, semantic constraint nodes, and business logic constraint edges, includes: Collect and standardize the identifiers of financial instruments from various financial markets, integrate the coding systems adopted by different exchanges, and introduce industry-recognized business terminology definitions to integrate them into entity nodes, thereby forming an entity relationship network. Embed the parsed business semantics and units of measurement from the structured protocol metadata to generate semantic constraint nodes; Integrate the state transition logic and numerical boundary conditions from the pre-set financial business rule base to form business logic constraint edges; By executing the above steps in sequence, a domain knowledge graph that integrates entity relationship networks, semantic constraint nodes, and business logic constraint edges is constructed.

3. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 2, characterized in that, The step of calculating cross-protocol field semantic similarity based on structured protocol metadata and domain knowledge graphs to generate initial field mapping relationships includes: Extract structured protocol metadata as input features for similarity calculation; By using a pre-trained natural language processing model, the input features are vectorized to generate low-dimensional dense vectors of the fields. Based on the entity relationship network and semantic constraint nodes in the domain knowledge graph, the cosine similarity between low-dimensional dense vectors of fields is calculated as a quantitative indicator of semantic similarity. Based on the semantic similarity ranking among all field low-dimensional dense vectors, the system automatically recommends field pairs with semantic similarity no lower than a preset similarity threshold, thereby generating initial field mapping relationships.

4. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 1, characterized in that, The detection field semantic conflict is resolved by making conflict resolution decisions based on semantic constraints in the domain knowledge graph and a preset historical mapping statistical result database, including: Define a semantic conflict when the semantic similarity between two cross-protocol fields is lower than a preset similarity threshold, and then identify the type of semantic conflict, including unit mismatch, enumeration value ambiguity, or business rule contradiction. Starting from the entity node corresponding to the semantically conflicting field in the domain knowledge graph, query the associated semantic transformation path and business logic constraints in the domain knowledge graph; Access the preset historical mapping statistics result library, and match the distribution of resolution schemes for the same conflict type in the historical mapping statistics result library based on the current conflict type, the semantic transformation path involved, and business logic constraints. The system integrates deterministic data from the domain knowledge graph and empirical data from the historical mapping statistics database to calculate the confidence level of each potential resolution scheme. The highest confidence level is selected and compared with a preset confidence threshold. When the highest confidence level exceeds the preset confidence threshold, the corresponding transformation logic of the scheme is automatically executed to complete the conflict resolution; otherwise, a manual intervention instruction is triggered.

5. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 4, characterized in that, The command to trigger manual intervention includes: Record the final conflict resolution plan confirmed and executed by manual intervention instructions, and update the historical mapping statistics database; When a similar conflict occurs again, increase the confidence weight of the corresponding resolution solution; Syntactic and semantic analysis is performed on manually confirmed conflict resolution solutions to extract semantic transformation paths and business logic constraints. If the solution has a fixed business logic or semantic transformation path, the verified transformation logic is synchronized to the domain knowledge graph in a structured form. This is manifested in creating new semantic constraint nodes or modifying existing business logic constraint edges in the knowledge graph to update and optimize the business logic constraints.

6. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 1, characterized in that, The real-time monitoring of abnormal fields in the input data stream automatically calls historical data patterns to perform field repair, including: An abnormal field is defined as one whose value in a field in the data stream is missing, out of bounds, or has an abnormal format. The relevant field value sequences within adjacent time windows in the real-time data stream are extracted to generate a spatiotemporal feature matrix. Synchronously query the compliant value range associated with abnormal fields in the domain knowledge graph; The spatiotemporal feature matrix is ​​used as the main input and the compliant value range is used as the constraint verification for the repair operation: if the detected abnormal field is a missing field, the mathematical model in the sequence interpolation algorithm is started to calculate an estimated value to fill in the missing field; If the detected abnormal field is a value that is out of bounds, the abnormal handling strategy in the associated business logic constraint edge will execute the repair process. If the detected abnormal field is a format error, a preset regular expression template is invoked, and this template is then applied to the field value with the incorrect format. By matching and recombining characters, the field is corrected to a compliant format. The regular expression template defines the correct text format that the field should follow.

7. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 6, characterized in that, An abnormal field is defined as a field in a data stream that exhibits any of the following characteristics: missing field, out-of-bounds value, or abnormal format. By using the list of required fields defined by the structured protocol metadata, we can check in real time whether there are any missing fields in the data packets; Based on the definition of numerical precision in the structured protocol metadata and the context information extracted from the real-time input data stream, dynamic value domain boundary conditions are generated. Then, a value of a field in the real-time input data stream that exceeds the dynamic value domain boundary conditions is defined as a numerical out-of-bounds condition. The numerical precision definition includes the static format requirements and business logic constraints of the field. By comparing the real-time data stream with the protocol format templates stored in the domain knowledge graph, abnormal field formats can be identified.

8. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 1, characterized in that, The protocol version change signal is detected in the following ways: Real-time crawling of the version identifier from the exchange's protocol release page; After successfully retrieving the version identifier of the page each time, it is compared with the version identifier of the corresponding protocol stored in the local database. If the two are inconsistent, it is determined that the version of the protocol has changed, and an internal protocol version change signal is immediately generated.

9. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 1, characterized in that, The process of dynamically updating the natural language processing model and conflict resolution decisions based on newly added protocol document samples and revised field mapping relationships to generate a fusion protocol standard includes: The newly added protocol document samples are input into the incremental training unit to adjust the semantic parsing weights of the natural language processing model; When a protocol version change signal is detected, the system automatically downloads the complete protocol document associated with the new version identifier according to the preset download link rules, in order to obtain the structured protocol metadata of the new protocol document sample. After downloading, the structured protocol metadata of the two protocol versions is compared, and a version difference summary containing the main changes is automatically extracted and generated to optimize the calculation of cross-protocol field semantic similarity. By merging all historical conflict resolution records and combining them with the characteristics of the new protocol, the priority ranking of conflict resolution decision rules is reconstructed.

10. The adaptive fusion system for multi-source financial protocols based on AI semantic understanding and knowledge graphs according to claim 9, characterized in that, The generation fusion protocol standard also includes: The standard output data stream is continuously monitored, and all executed anomaly repair records are automatically extracted and structured and fed back to the historical data pattern library. When the same type of anomaly is detected again in the continuously monitored real-time data stream, the dynamically updated historical data pattern library is automatically queried. By analyzing the success rate and application frequency of various repair strategies in the historical data pattern library, the priority weight of the corresponding repair strategy is dynamically strengthened. The repair verification results will be synchronously updated to the business logic constraint edges of the domain knowledge graph.

Images