Information matching method and device, equipment and storage medium

Abstract

The present disclosure provides an information matching method and device, equipment and a storage medium, relates to the technical field of computers, in particular to the technical field of big data, data processing, entity disambiguation, artificial intelligence and the like. The specific implementation scheme is as follows: obtaining to-be-processed business data and a source identifier of the to-be-processed business data; extracting at least one initial attribute information of a to-be-processed entity from the to-be-processed business data; performing a matching operation on the at least one initial attribute information and a preset dictionary to obtain an attribute set corresponding to each of the at least one initial attribute information; and sequentially processing attribute sets corresponding to each of a plurality of matchers in a matcher sequence corresponding to the source identifier, to perform a matching operation on the to-be-processed entity and a candidate entity set.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of big data, data processing, entity disambiguation, and artificial intelligence. Background Technology

[0002] Entity disambiguation (ED) is a key technology in natural language processing and knowledge graph construction, aiming to solve the problem of multiple representations of the same entity during the digital representation of real-world entities. Due to the diversity and ambiguity of entity references in natural language, the same entity may correspond to multiple different representations. For example, the same entity may have a standard name or alias; different entities may have the same standard name; and even the same name may point to multiple entities with vastly different categories, attributes, and semantic connotations.

[0003] Entity disambiguation is a prerequisite for ensuring the accuracy of semantic understanding and is crucial for building high-precision structured knowledge bases. In downstream applications such as information retrieval, intelligent question answering, and data mining, failure to eliminate entity ambiguity will lead to broken knowledge associations, index redundancy, and inference biases, severely impacting the system's recall and accuracy. Especially in scenarios involving the fusion of multi-source heterogeneous data, entity disambiguation can effectively clarify the unique identity boundaries of entities, solve the data silo problem caused by naming conflicts, and provide reliable semantic support for upper-layer artificial intelligence applications. Summary of the Invention

[0004] This disclosure provides an information matching method, apparatus, device, and storage medium.

[0005] According to one aspect of this disclosure, an information matching method is provided, comprising: Obtain the pending business data and its source identifier; Extract at least one initial attribute information of the entity to be processed from the business data to be processed; Match at least one initial attribute with a preset dictionary to obtain the attribute set corresponding to each of the at least one initial attribute; Based on multiple matchers in the matcher sequence corresponding to the source identifier, the attribute sets corresponding to each matcher are processed sequentially to perform matching operations on the entity to be processed and the candidate entity set.

[0006] According to another aspect of this disclosure, an information matching apparatus is provided, comprising: The acquisition module is used to acquire the business data to be processed and the source identifier of the business data to be processed. The extraction module is used to extract at least one initial attribute information of the entity to be processed from the business data to be processed. The matching module is used to match at least one initial attribute information with a preset dictionary to obtain the attribute set corresponding to each of the at least one initial attribute information. The processing module is used to process the attribute sets corresponding to each matcher in the matcher sequence based on the source identifier in order to perform matching operations between the entity to be processed and the candidate entity set.

[0007] According to another aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described in the present disclosure.

[0008] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform any of the methods according to embodiments of this disclosure.

[0009] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the methods according to embodiments of this disclosure.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a flowchart illustrating an information matching method according to an embodiment of the present disclosure; Figure 2 This is a flowchart illustrating the process of sequentially processing the attribute set corresponding to each matcher in a matcher sequence based on the source identifier, according to an embodiment of the present disclosure. Figure 3 This is a schematic flowchart illustrating the process of obtaining first path encoding information according to an embodiment of the present disclosure; Figure 4 This is a schematic diagram of a process for performing rule matching and traffic splitting on first path encoding information based on preset traffic splitting rules according to an embodiment of the present disclosure; Figure 5 This is a flowchart illustrating abnormal problems and potential causes in the location matching process according to an embodiment of the present disclosure; Figure 6 This is a schematic diagram of the structure of an information matching device according to another embodiment of the present disclosure; Figure 7 This is a block diagram of an electronic device used to implement the information matching method of the embodiments of this disclosure. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] The terms “first,” “second,” etc., used in this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion, such as including a series of steps or units. A method, system, product, or apparatus is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0014] It should be noted that, unless it is explicitly stated that there is a sequential order of execution between different operations, or that there is a sequential order of execution between different operations in terms of technical implementation, the execution order between multiple operations may not be significant, and multiple operations may be executed simultaneously.

[0015] Entity disambiguation is a key technology in natural language processing with numerous applications. Taking pharmaceutical intelligence analysis and commercial data analysis as examples, entity disambiguation technology possesses crucial application value. Data sources, such as HCPs (Healthcare Professionals) and marketed products, typically encompass multimodal and heterogeneous information including academic literature, clinical trial reports, medical insurance databases, and social media. Due to a lack of naming conventions, the same HCP may be represented by its full name, abbreviation, or a combination of professional titles. Similarly, the same product may exist in multiple referential forms, such as generic name, trade name, and R&D code. By implementing entity disambiguation, the system can accurately align referential items scattered across unstructured data from different sources to a unique HCP or product identifier, thereby constructing a complete and accurate entity profile.

[0016] This technology is crucial for ensuring the quality of data-driven decision-making. In product monitoring, taking pharmaceuticals as an example, eliminating ambiguity enables seamless aggregation of data throughout the entire drug lifecycle, avoiding underreporting of adverse reactions or distortion of market competition analysis due to confusing terminology.

[0017] In related technologies, entity disambiguation solutions often employ static, fixed rules, and some technologies customize the execution logic of entity messages for specific business scenarios. These technologies all lack a unified adaptation and scheduling mechanism for multi-source, heterogeneous business scenarios.

[0018] Taking medical data products as an example, business data such as papers, clinical trials, academic societies, conferences, guidelines, grants, and news need to be accurately correlated with the HCP master data. This process faces challenges such as multi-source heterogeneity, missing fields, name aliases, and inconsistencies in institutional hierarchy. Mainstream methods mostly focus on disambiguating authors of English papers, relying on string matching or general vector retrieval, which are insufficient for covering more complex information, such as contextual features like departmental hierarchy, geographical distribution, collaborator networks, and conference series history.

[0019] In view of this, this disclosure proposes an information matching method for entity disambiguation, which can also be understood as providing an entity disambiguation method. This method obtains the business data to be processed and its source identifier, extracts the initial attribute information of the entity to be processed, combines it with a preset dictionary to obtain the corresponding attribute set, and processes each attribute set sequentially based on the matcher sequence corresponding to the source identifier, thereby achieving matching between the entity to be processed and the candidate entity set. This method is not only applicable to simple scenarios with few feature dimensions, but also to complex scenarios with multiple heterogeneous sources, many entity feature dimensions (e.g., greater than a preset number of dimensions), and dependency relationships among feature dimensions.

[0020] like Figure 1 The diagram shown is a flowchart illustrating the information matching method provided in this embodiment of the disclosure, including the following: S101, Obtain the business data to be processed and the source identifier of the business data to be processed.

[0021] Business data to be processed refers to the raw data that needs to be identified and matched in the current business process. For example, taking HCP as an example, business data to be processed can be business records from different information sources such as papers, clinical trials, academic societies, conferences, guidelines, grants, news, or external HCP master data.

[0022] The source identifier is an identifier of the information source, used to uniquely mark the source of the business data to be processed. It can be in the form of a string or a numeric code.

[0023] S102, extract at least one initial attribute information of the entity to be processed from the business data to be processed.

[0024] An entity to be processed refers to the main body that carries the core business object in the business data to be processed. For example, an entity to be processed can be a target HCP involved in the business data, a product identifier in the business data, etc.

[0025] Initial attribute information refers to the basic feature fields used to describe the entities to be processed, which are directly parsed from the business data to be processed. These feature fields are not standardized and retain the original form of the data source. For example, taking a paper as an example, the initial attribute information may include at least one of the following: original personnel text, original institution text, original department text, original region text, original disease / research field text, original collaborator text, and original series of activities text.

[0026] In implementation, at least one initial attribute information of the entity to be processed can be extracted from the business data to be processed through predefined field paths or regular expression matching. The specific information extraction method is not limited in this embodiment.

[0027] S103, perform a matching operation between at least one initial attribute information and a preset dictionary to obtain attribute sets corresponding to at least one initial attribute information.

[0028] A predefined dictionary is a standardized mapping tool built for different attribute types, used to convert non-standard, ambiguous, or colloquial initial attribute information into standardized, uniform representations. The dictionary type can be flexibly configured according to business needs. For example, taking HCP disambiguation as an example, the predefined dictionary may include at least one of the following: The database comprises at least one of the following: Institutional Dictionary, Departmental Dictionary, Regional Dictionary, Disease Dictionary, Subject Dictionary, HCP Master Data Dictionary, and Dynamic Relationship Feature Database. Specifically, the Institutional Dictionary is a standardized set of codes and attributes for legally practicing medical service institutions. The Departmental Dictionary is a standardized classification and coding of clinical / medical technology / administrative departments within a medical institution. The Regional Dictionary is a standardized dataset that hierarchically and coded regional administrative divisions. The Disease Dictionary is a standardized set of classifications and codes for health issues such as diseases, symptoms, and injuries. The Subject Dictionary is a standardized classification of medical professional fields and research directions. The HCP Master Data Dictionary is a collection of unique identifiers and standardized descriptions for personnel entities engaged in medical, pharmaceutical, and public health professional services. The Dynamic Relationship Feature Database is generated from matched business data and manually reviewed results and is continuously updated with new data.

[0029] In implementation, each initial attribute information corresponds to its own attribute set. For each initial attribute information, its attribute set refers to the information set containing multiple representations of that initial attribute information, generated after dictionary mapping. For example, for the initial attribute information "name," its attribute set may include different representations of "name," "alias," etc.

[0030] In implementation, the initial attribute information in the business data to be processed can first be converted into matchable standardized entity features, and then matched with a preset dictionary to obtain at least one attribute set corresponding to each initial attribute information. Alternatively, the standardized entity features of each initial attribute information can be obtained by preprocessing the business data to be processed.

[0031] S104, based on multiple matchers in the matcher sequence corresponding to the source identifier, process the attribute sets corresponding to each matcher in sequence to perform matching operations on the entity to be processed and the candidate entity set.

[0032] A matcher sequence is a pre-configured, ordered chain of multiple matchers arranged according to corresponding business logic, based on a source identifier. Because different sources have varying data quality, field integrity, and business importance, different matcher sequences can be set for different source identifiers. This matcher sequence defines the execution order of each matcher.

[0033] A matcher is an atomic processing unit that encapsulates matching logic and business rules. Each matcher is an independent logical unit with a defined unified input and output interface. For example, a matcher may include at least one of the following: NameMatcher, OrganizationMatcher, DepartmentMatcher, LocationMatcher, DiseaseMatcher, CollabMatcher, SeriesMatcher, and LLMMatcher.

[0034] For each matcher, its candidate entity set is the set of known entities that the matcher needs to process and filter and verify.

[0035] During implementation, the corresponding matcher sequence can be read based on the obtained source identifier, the execution order of the matcher sequence can be determined, and the attribute sets corresponding to each matcher can be processed in sequence to perform matching operations on the entity to be processed and the candidate entity set.

[0036] The matchers provided in this disclosure are atomic matchers. Each matcher includes editable parameters, and the parameter values ​​can be generated manually. The same matcher can be reused in multiple scenarios, and the execution order of the matchers can be arranged according to requirements for business data from different source identifiers.

[0037] In this embodiment, by acquiring the business data to be processed and its source identifier, and extracting multiple initial attribute information of the entities to be processed, the original characteristics and business source attributes of multi-source business data can be completely preserved, providing a complete data foundation for differentiated matching and adapting to multi-source heterogeneous business data. By matching and verifying the initial attribute information with a preset dictionary, the initial attribute information is standardized and normalized, which can uniformly resolve problems such as aliases, abbreviations, former names, hierarchical confusion, and parsing errors in the original data. By handing over the standardized attribute set after dictionary standardization to the dedicated matcher sequence corresponding to the source identifier for step-by-step processing, the corresponding hierarchical precision matching logic can be adapted to the feature differences of different business source data, overcoming the drawbacks of traditional general matching schemes that uniformly process multi-source data and have poor adaptability. This processing architecture, which first standardizes and normalizes through preset fields and then performs source-specific matching based on the matcher sequence, realizes standardized preprocessing and scenario-based differentiated matching of multi-source heterogeneous medical business data, which can effectively improve the accuracy and stability of matching entities to be processed and candidate entities, and improve the efficiency of entity disambiguation.

[0038] Taking HCP disambiguation as an example, the problem of matching deviation caused by the high difficulty of standardization of medical master data and the inconsistency of fields can be effectively solved by pre-setting fields. Through the chain matching path of the matcher sequence, funnel-shaped matching can be achieved, and information can be filtered step by step to achieve entity disambiguation.

[0039] In this embodiment of the disclosure, in order to perform entity disambiguation more effectively, each matcher in the matcher sequence includes multiple matching conditions with different precision levels; the matching conditions are used to define the conditions that belong to the same entity.

[0040] Accuracy level is a quantitative indicator that measures the strictness of matching conditions. A higher level indicates stricter constraints on the matching conditions and higher accuracy in the judgment. A lower level indicates looser constraints on the matching conditions and greater tolerance for errors.

[0041] Matching criteria are used to define the criteria for determining whether the business data to be processed and the standard candidate entity (i.e., the known candidate entity) belong to the same entity.

[0042] During implementation, matching conditions with gradient precision levels can be pre-configured for each level of matcher, forming an entity matching judgment rule system that adapts step by step from high precision to low precision.

[0043] During implementation, based on multiple matchers in the matcher sequence corresponding to the source identifier, the attribute sets corresponding to each matcher are processed sequentially, such as... Figure 2 As shown, the following can be executed separately for each matcher: S201, obtain the candidate entity set and the attribute set corresponding to the matcher as the target attribute set.

[0044] In implementation, when multiple initial attribute information is extracted from the business data to be processed, each initial attribute information yields its own attribute set. A single matcher supports matching the corresponding initial attribute information. For each matcher, the attribute set corresponding to its initial attribute information is the target attribute set of that matcher. For example, when the current matcher is OrganizationMatcher, the target attribute set can be the attribute set corresponding to the organization, which may include information such as the organization's standard name and a list of organization aliases.

[0045] During implementation, for the current level matcher, the candidate entity set filtered by the previous level matcher can be read, and the target attribute set corresponding to the matcher can be retrieved to perform a matching operation, so as to continue funnel-shaped information filtering from the candidate entity set.

[0046] S202, in descending order of accuracy level, the target attribute set and the candidate entity set are compared according to the matching conditions, and the candidate entity corresponding to the matching condition that matches in the first comparison is determined as the successor entity.

[0047] During implementation, the matching condition of each level of precision in the matcher is traversed from high to low. The target attribute set and the candidate entity set are matched and verified one by one. After the first precision rule that meets the matching condition is hit, the precision downgrade traversal stops and the corresponding candidate entity is determined as the successor entity.

[0048] S203, if multiple successor entities are matched and the matcher is not the last matcher in the matcher sequence, the multiple successor entities are taken as the candidate entity set of the next matcher of the matcher.

[0049] During implementation, if multiple successor entities are obtained in the current matching and filtering process, and the current matcher is an intermediate-level matcher, then all successor entities obtained by the matcher are packaged as the candidate entity set of the next-level matcher and enter the next round of attribute matching and filtering process.

[0050] In this embodiment, multiple matching conditions with different accuracy levels are configured for each matcher, enabling the construction of a multi-level judgment standard from strict to lenient. This provides a stepped decision path for entity recognition, avoiding over-matching or under-matching problems caused by a single threshold. Comparing the target attribute set and candidate entity set sequentially in descending order of accuracy ensures the accuracy of entity matching. Furthermore, determining subsequent entities based on the initial matching result reduces unnecessary redundant calculations and improves the execution efficiency of entity matching, facilitating the search for candidate entities with the highest possible matching accuracy. When multiple subsequent entities are matched and the current matcher is not the last matcher, the resulting set is passed to the next matcher for iterative filtering. This enables cascading disambiguation across matchers, addressing the shortcomings of existing technologies in providing unified decision-making capabilities across business data sources and inconsistent results from multi-source data association.

[0051] During implementation, the matcher also includes conflict determination rules; based on the conflict determination rules, candidate entities in the candidate entity set that meet the target conditions can be filtered out. The target conditions are: there is an information conflict between the candidate entity and the target attribute set.

[0052] Conflict determination rules refer to pre-defined standardized rules used to identify content contradictions and semantic inconsistencies in dimensional expression between business data features and candidate entity attributes.

[0053] Information conflict refers to an abnormal correspondence in which the attribute data of a candidate entity conflicts with any attribute in the target attribute set, or with half or more of the attributes in the target attribute set. In other words, it is determined that the two have contradictory, mismatched or mutually exclusive abnormal correspondences.

[0054] During implementation, before the matcher filters subsequent entities according to accuracy level, a preset conflict determination rule can be invoked to check the conflict between each candidate entity in the matcher's candidate entity set and the target attribute set, filter out candidate entities with information conflicts and complete the filtering and elimination, and retain valid candidate entities without conflicts for subsequent accuracy matching and hierarchical filtering processes.

[0055] In this embodiment of the disclosure, candidate entities that conflict with the target attribute set are filtered out based on conflict determination rules. This can quickly eliminate invalid entities with conflicts, such as interfering items with contradictory field values, logical mutual exclusion, or business rule inconsistencies. This can further improve the accuracy and reliability of entity matching and increase the efficiency and precision of entity disambiguation.

[0056] In the process of performing multi-precision matching of the entities to be processed and obtaining subsequent entities through the step-by-step convergence of each matcher, in order to fully record the combination of matchers called in this entity matching, the order of matching execution, and the characteristics of the matching chain, and to achieve traceability, replayability, and quantifiable analysis of the matching process, the matchers actually involved in the matching operation and their matching process can be encoded to obtain the first path encoding information. In implementation, such as... Figure 3 As shown, it may include the following: S301, For each enabled matcher in the matcher sequence, generate a matcher subcode for the matcher; the matcher subcode includes at least the matcher identifier of the matcher.

[0057] The enabled matcher is the valid matcher in the matcher sequence that participates in this match. If the initial attribute information is missing in the business data to be processed, the matcher for that initial attribute information will not be enabled during the matching process. For example, if the business data to be processed does not contain geographic information, but the matcher sequence includes a geographic matcher, then the geographic matcher (LocationMatcher) will not participate in this match.

[0058] A matcher subcode is an encoding unit used to record the decision result of a single matcher. The matcher subcode carries at least the matcher identifier of the corresponding matcher and can uniquely represent the execution subject and rule type of the matcher.

[0059] The matcher identifier is a unique coded identifier used to distinguish each matcher, matching dimension, and matching rule.

[0060] In implementation, different matcher identifiers can be represented by different letters. For example, the matcher identifier for NameMatcher can be "N", the matcher identifier for OrganizationMatcher can be "O", the matcher identifier for DepartmentMatcher can be "D", and the matcher identifier for LocationMatcher can be "L".

[0061] S302, perform a concatenation operation on the subcodes of the enabled matchers in the matcher sequence to obtain the first path encoding information.

[0062] The first path encoding information refers to the string formed by concatenating the subcodes of each enabled matcher in the matcher sequence in a predetermined order. The order strictly follows the order of the matchers in the matcher sequence.

[0063] For example, if the source is identified as a paper, the matcher sequence can be: NameMatcher->OrganizationMatcher->DepartmentMatcher->DiseaseMatcher->CollaborationMatcher, and the corresponding first path encoding information can be represented as [N#][O#][D#][E#][C#].

[0064] When the source identifier is "conference", the matcher sequence can be: NameMatcher->OrganizationMatcher->SeriesMatcher->DepartmentMatcher->DiseaseMatcher->CollabMatcher->LocationMatcher. The corresponding first path encoding information can then be represented as: [N#][O#][S#][D#][E#][C#][L#].

[0065] When the source identifier is "Society", the matcher sequence can be: NameMatcher->OrganizationMatcher->SeriesMatcher->DepartmentMatcher->CollabMatcher->LocationMatcher. The corresponding first path encoding information can then be represented as: [N#][O#][S#][D#][C#][L#].

[0066] When the source identifier is a guide, the matcher sequence can be: NameMatcher->OrganizationMatcher->DepartmentMatcher->DiseaseMatcher->CollabMatcher. The corresponding first path encoding information can then be represented as: [N#][O#][D#][E#][C#].

[0067] When the source identifier is a fund, the matcher sequence can be: NameMatcher->OrganizationMatcher->DiseaseMatcher. The corresponding first path encoding information can then be represented as: [N#][O#][E#].

[0068] When the source is identified as a clinical trial, the matcher sequence can be: NameMatcher->OrganizationMatcher->DiseaseMatcher->CollabMatcher. The corresponding first path encoding information can then be represented as: [N#][O#][E#][C#].

[0069] In this embodiment, a matcher sub-code containing a matcher identifier is generated for each enabled matcher in the matcher sequence. This abstracts each level of matching operation into a traceable, structured marker, effectively solving the problem of fragmented and chaotic process information and the lack of dedicated identifiers for each level of matching action in traditional entity matching. By sequentially concatenating each matcher sub-code according to the execution order of the matcher sequence, a well-organized first path encoding information is generated. This allows for a complete and orderly recording of the hierarchical execution path of the overall entity matching, preserving the execution trajectory and matcher call status of each round of matching and filtering. This facilitates subsequent compliance review and problem tracing of entity matching results.

[0070] To ensure that the generated matcher subcode not only identifies the matcher type but also records the actual matching precision level during a single match, achieving dual traceability across both the matching link and matching precision dimensions, and improving the granularity and interpretability of path encoding information, the generation method of the matcher subcode can be refined during implementation. The generation of the matcher subcode can be achieved based on the following steps: Step A1: If the matcher matches a subsequent entity, it generates a precision identifier for the precision level corresponding to the matching condition that matches the match. Precision identifiers are symbolic markers used to quantify the precision level of the current matcher. For example, they can be represented using non-negative integers (such as "0", "1", "2", "3") or specific status characters. "0" corresponds to the highest precision level, such as exact match of the current employer or exact match of a standard name alias. "1 / 2 / 3" correspond to progressively more relaxed precision levels, such as match of a former employer, generalized match of a parent institution, or match of a related subject. Larger numbers indicate looser constraints and lower confidence levels.

[0071] In practice, when the matcher matches a subsequent entity, it first determines the precision level based on the matching type, and then uses the symbol corresponding to that precision level as the precision identifier.

[0072] Step A2: Generate the matcher subcode based on the matcher identifier and the precision identifier.

[0073] In implementation, the matcher identifier and precision identifier can be concatenated in the format "[matcher identifier + precision identifier]" to generate the matcher subcode. For example, if the precision identifier corresponding to NameMatcher is 0 and the matcher identifier is N, then the corresponding matcher subcode can be [N0].

[0074] In this embodiment, when the matcher successfully matches a subsequent entity, a precision level identifier corresponding to the matching condition is generated. This enables an explicit association between the matching result and its confidence level, providing a quantitative basis for subsequent quality assessment. Generating the matcher sub-code based on the matcher identifier and precision identifier retains the processing logic of the matcher dimension while reflecting the granularity of the condition dimension, thus improving the refinement and parsingability of path coding information.

[0075] During the matching process of candidate entity comparison and filtering, there may be special scenarios where all candidate entities conflict with the target attribute set, and no effective subsequent entity matches. To fully record such abnormal matching states and conflict matching characteristics, enrich the state dimension of the first path encoding information, and realize the source tracing and classification statistics of matching failure scenarios, the matching subcode of the matcher can be generated based on the following steps: Step B1: Based on the conflict determination rules of the matcher, if all candidate entities in the candidate entity set have information conflicts with the target attribute set, a conflict identifier is generated. A conflict identifier is a specific symbol used to characterize an irreconcilable logical contradiction between the business data features and the inherent attributes of the candidate entity under the current matching dimension. For example, it can be represented by "X".

[0076] Step B2: Generate the matcher subcode based on the matcher identifier and the conflict identifier.

[0077] In implementation, the matcher identifier and conflict identifier can be concatenated in the format "[match identifier + conflict identifier]" to generate the matcher subcode. For example, if DepartmentMatcher detects departmental mutual exclusion, the generated matcher subcode can be represented as "[DX]".

[0078] In this embodiment, a conflict identifier is generated when the conflict determination rules based on the matcher determine that all candidate entities in the candidate entity set conflict with the target attribute set. This explicitly marks the global unmatchable state under the current matcher, capturing specific conflict situations and preventing the system from getting stuck in invalid iterations or misjudgments when there are zero valid candidates. Generating a matcher subcode based on the matcher identifier and conflict identifier makes matching scenarios with no valid successor entities or information conflicts explicit and coded. This allows the matching process to not only trace the normal matching path but also to fully trace the abnormal scenarios of matching failure and data contradictions. This effectively solves the problems of black-box matching processes, lack of record of conflict causes, and difficulty in locating abnormal nodes in related technologies.

[0079] In scenarios where a unique successor entity is obtained through step-by-step filtering using a matcher sequence, the entity to be processed often retains some initial attribute information that was not involved in the current matching operation. These unused attributes still have value for entity feature verification. To fully leverage the verification role of initial attributes, avoid matching errors caused by single-dimensional matching, and further improve the accuracy and reliability of subsequent entity matching results, a secondary verification of successfully matched successor entities can be performed using unused initial attributes. This includes the following: Step C1: If a single successor entity is matched and at least one unused initial attribute exists in at least one initial attribute information, select the initial attribute from the unused initial attribute as the successor attribute. Post-hoc attributes refer to redundant or auxiliary attribute fields that are not used by the earlier matchers in the entity matching process, but are logically useful for verifying the correctness of the obtained matching results. Examples include auxiliary fields such as department, region, disease, and collaborators carried in the business data to be processed.

[0080] During implementation, after executing some or all matchers in the matcher sequence and uniquely identifying a successor entity, it is possible to first check whether the currently used attribute set covers all extracted initial attribute information. If it is determined that there are remaining initial attributes that have not yet participated in the matching operation, a priority selection method can be used to select the highest priority and unused initial attribute from the remaining initial attributes as the successor attribute.

[0081] Step C2: Based on the matchers corresponding to the posterior attributes in the matcher sequence, validate the successor entity to obtain the validation result.

[0082] It should be noted that this step is an optional reinforcement verification process, and you can flexibly choose to perform post-verification or directly generate the encoding result according to the preset rules.

[0083] During implementation, when posterior verification is required, the dedicated matcher corresponding to the posterior attribute dimension in the matcher sequence is invoked. The posterior attribute data is loaded to verify the unique successor entity obtained after matching, and the corresponding dimension verification result is output, completing the secondary reinforcement verification of the matching result. This verification result contains the matcher subcode of the matcher corresponding to the posterior attribute.

[0084] If no post-verification is required, the post-verification process can be skipped directly, and subsequent path encoding generation steps can be performed directly based on the single matched successor entity.

[0085] Understandably, whether to perform subsequent verification can be flexibly configured based on the source identifier of the business scenario. For example, the post-verification function can be enabled or disabled. When the function is enabled, post-verification is performed.

[0086] In this embodiment of the disclosure, when a single successor entity is matched and there are unused initial attributes, the remaining attributes are introduced as post-validation attributes into subsequent verification. This forms a secondary verification mechanism for the matching results, effectively screening for implicit matching deviations, field alias conflicts, and hierarchical mismatches in a single high-precision match. It avoids mismatches caused by local attribute consistency but overall information mismatch, and realizes in-depth mining and closed-loop utilization of business data value. Thus, without excessively increasing the comparison cost, it enhances the reliability, security, and fault tolerance of entity parsing results.

[0087] In practice, after multi-precision comparison and conflict filtering of all matchers in the matcher sequence, if multiple successor entities output by the current matcher cannot achieve unique locking, arbitration can be performed based on the large model. The specific processing logic is as follows: Step D1: When multiple successor entities are matched and the matcher is the last matcher, input the attribute sets corresponding to the multiple successor entities and at least one initial attribute information into the large model to obtain the matching result of the large model. Large models are trained on massive amounts of text, images, and other types of data. They have a huge number of parameters and can autonomously understand semantics, perform logical reasoning, content creation, knowledge question answering, and multimodal processing. They are artificial intelligence foundational models that can adapt to various complex intelligent tasks.

[0088] During implementation, at least one set of attributes corresponding to initial attribute information can be input into a large model. By leveraging the large model's ability to call external tools (such as retrieval tools), its ability to understand and process massive amounts of information, and its deep understanding of complex semantics, implicit associations, and weak feature signals, the final entity matching decision result can be output from multiple subsequent entities.

[0089] During implementation, the large model can also output the decision rationale for the matching results, which is used to understand the basis for the large model to give the matching results.

[0090] Understandably, the matching result of a large model can be either a match for a unique successor entity or a match for no unique successor entity.

[0091] Step D2: Add the model identifier of the large model and the result identifier of the matching result to the first path encoding information.

[0092] A model identifier is a letter designation used to distinguish large model types. For example, a model identifier can be represented by "M" (Model).

[0093] Result identifiers are symbols used to characterize the decision state of a large model, following a coding convention similar to that of the aforementioned matcher: "0" represents that the large model hits the only candidate with high confidence; "1" represents that the large model gives a tendency ranking but there is still uncertainty; "X" represents that the large model cannot distinguish between multiple successor entities or there is a logical conflict; "empty" represents that the large model was not called or the call failed.

[0094] In implementation, the model identifier of the large model and the result identifier of the matching result can be combined into the matcher subcode corresponding to the large model, such as [M0]. Then, the matcher subcode is appended to the end of the first path encoding information.

[0095] In this embodiment, when multiple successor entities are matched and the current matcher is at the end of the sequence, the candidate entity set and the full attribute set are input into a large model for processing. This leverages the large model's deep understanding of complex semantics, implicit relationships, and weak feature signals, effectively solving the problem that traditional fixed rules are ill-suited to the complex aliases, diverse hierarchies, and significant cross-source differences in medical data. Appending the model identifier of the large model and the result identifier of the matching result to the first path encoding information makes the data-driven black-box reasoning process transparent and structured. This ensures that the entire matching chain simultaneously encompasses the deterministic path of rule matching and the decision nodes of the large model's reasoning, forming an auditable and reproducible complete chain of evidence.

[0096] After generating the first path encoding information carrying the complete matcher execution order and matching accuracy information at each level using the above method, to achieve differentiated traffic management, categorized archiving and traceability of different first path encoding information, and to improve the regularity, retrieval, and operational efficiency of business matching records, rule matching and traffic distribution can be performed on the first path encoding information based on preset traffic distribution rules, and targeted association storage of matching data can be completed, such as... Figure 4 As shown, it may include the following: S401, match the first path encoding information with multiple preset traffic splitting rules to determine the target rule corresponding to the first path encoding information.

[0097] Traffic splitting rules are pre-configured logical criteria used to perform differentiated post-processing of business data based on the matching path characteristics of the first path encoding information. Each traffic splitting rule defines the mapping relationship between a specific path encoding pattern and the corresponding processing strategy.

[0098] The target rule is a specific traffic splitting strategy that is matched from the rule base and is applicable to the current first path encoding information.

[0099] During implementation, the target rule corresponding to the first path encoding information can be determined by matching based on the level combination (such as whether it contains a high confidence prefix such as "N0O0"), conflict information, the size of the candidate entity set (whether it is 1), the source identifier, and preset threshold parameters and multiple preset diversion rules in the first path encoding information.

[0100] For example, for records whose first path encoding information belongs to a high-confidence combination, is unique as a candidate, and has no conflict in the supplementary dimensions, the target rule can be determined as "directly entering the database".

[0101] For records where the first path encoding information contains multiple candidates or has conflicts in key dimensions, the target rule can be determined as "manual review".

[0102] For records in the first path encoding information that contain a small number of candidates and are difficult to judge manually but have high business value, the target rule can be determined as "model arbitration". That is, when there are multiple successor entities and a large model is required for processing, the data suitable for processing by the large model can be filtered out first through the diversion rule, and then the large model can be used to filter from multiple successor entities.

[0103] For records where there are no candidates, insufficient evidence, or low path credibility in the first path encoding information, the target rule can be set as "not to be processed for the time being".

[0104] S402, associate and store the first path encoding information and the matching record of the matcher sequence for the business data to be processed into the target queue corresponding to the target rule.

[0105] A target queue refers to a physical or logical storage unit determined according to the routing rules, used to hold specific types of matching results. For example, a target queue may include at least one of a direct entry queue, a manual review queue, a model arbitration queue, and a queue that is not processed temporarily.

[0106] A matching record refers to all contextual information, excluding the first path encoding information, used to support subsequent processing. This matching record may include: the candidate entity set one level above the stopping node, the final candidate entity set, detailed explanatory information, a summary of the original business data, and a source identifier.

[0107] Associated storage refers to the process of using the first path encoding information as an index key, packaging it with the aforementioned matching records, and writing it into the storage medium corresponding to the target queue.

[0108] In implementation, based on the target rule corresponding to the determined first path encoding information, the matching records of the first path encoding information and the matcher sequence for the business data to be processed can be associated and stored in the target queue corresponding to the target rule. For example, if the target rule is "direct entry into the database", it can be associated and stored in the direct entry into the database queue.

[0109] In this embodiment, by matching the first path encoding information with multiple preset diversion rules, the target rule adapted to the current entity matching process is accurately located. This enables automated and refined classification and diversion of business data based on standardized matching execution paths, solving the problems of mixed storage of traditional entity matching data, ambiguous scenario classification, and inability to manage data differently according to matching links. By associating the first path encoding information with the complete matching record corresponding to the matcher sequence and storing it in the target queue corresponding to the target rule, integrated binding and archiving of encoding paths, matching process data, and diversion rules are achieved. This allows for partitioned and organized storage of business data with different matching logics, different matching paths, and different levels of difficulty. This ensures the traceability, statistical compatibility, and optimizability of entity matching decision logic under different business scenarios, continuously improving the stability, standardization, and iterative upgrade capabilities of the medical personnel entity unification matching system.

[0110] After batch business data completes entity matching and generates multiple complete first path encoding information, the execution status of the current matching strategy (including the preset dictionary and / or matcher) can be reflected based on the large amount of generated first path encoding information, so as to promptly identify problems and correct the matching strategy. To achieve abnormal behavior identification and iterative optimization of matching rules in the entity matching process, embodiments of this disclosure can locate abnormal problems and potential causes in the matching process by quantitatively comparing the differences between batch encoding features and the standard baseline model. In implementation, a baseline model can be established first. The baseline model is pre-built and used to characterize the feature expression under normal operating conditions of the matching strategy.

[0111] During implementation, a baseline model can be built based on historical matching data. The specific implementation steps are as follows: Step E1: Obtain historical reference information; the historical reference information includes multiple first path encoding information and the metadata corresponding to the first path encoding information; Historical reference information consists of multiple first path encoding information and their metadata collected by the matching strategy within the target runtime. Each first path encoding information is stored in the aforementioned [N#][O#][D#][L#][E#][C#][S#][M#] format and associated with metadata. The metadata may include at least one of the following: source identifier, execution timestamp of each matcher, input and output candidate entity sets of each matcher, manual review result tags, and queue identifier of the target queue to which the traffic belongs.

[0112] Step E2: Construct a baseline model based on historical reference information.

[0113] In this embodiment of the disclosure, a baseline model is constructed by retrieving historical reference information containing first path encoding information and corresponding metadata. This allows the baseline model to be constructed based on real and massive historical data, thereby improving the accuracy of the baseline model construction.

[0114] In implementation, the baseline model may include at least one of the following: pattern distribution baseline, attribute-level probability baseline, and path transformation funnel baseline. The construction method of each baseline model is described below.

[0115] 1) Constructing a pattern distribution baseline can be achieved based on the following steps: Step F1: Count the number of occurrences of each path pattern in the historical reference information as observation data.

[0116] Each piece of business data receives a unique first path code. However, duplicate information can appear within this massive dataset. After deduplication, the path pattern is obtained. During implementation, the pattern distribution baseline represents the distribution characteristics of the path patterns. To improve the accuracy of the pattern distribution baseline construction, reusable path coding templates can be obtained through frequent itemset mining of path patterns. Path patterns in frequent itemsets occur more frequently than a preset number of times.

[0117] The frequency of occurrence is the number of times each path pattern is repeated in the historical reference information.

[0118] During implementation, the path codes of all samples in the historical reference information are traversed, the actual frequency of occurrence of each path pattern is matched and counted, and structured observation data is obtained. This observation data can be described by expression (1): (1) In expression (1), This represents the obtained observation data; This represents the total number of times path pattern k appears.

[0119] Step F2: Based on the observed data, update the Dirichlet prior distribution to obtain the Dirichlet posterior distribution; The Dirichlet prior distribution is a pre-defined prior probability model used to characterize the initial statistical regularity of the path pattern distribution.

[0120] During implementation, the Dirichlet prior distribution is updated to obtain the Dirichlet posterior distribution, which can be described by expression (2): (2) In expression (2), This indicates that the Dirichlet posterior distribution has been obtained; Represents the Dirichlet prior distribution; This represents the obtained observation data.

[0121] Step F3: Extract the mode vector of the Dirichlet posterior distribution to obtain the baseline of the pattern distribution.

[0122] The mode vector is the probability vector that corresponds to the maximum value of the Dirichlet posterior probability density function. It is used to represent the distribution state of the most likely path pattern under given observation data.

[0123] In practice, the mode vector of the Dirichlet posterior distribution is extracted and can be described by expression (3): (3) In expression (3), Let represent the mode estimate of the probability of occurrence of the i-th type of path pattern under the Dirichlet posterior distribution, which is also the baseline of the pattern distribution; Let represent the Dirichlet posterior distribution corresponding to the i-th path pattern; i represents the index of a path pattern whose probability is to be calculated; j represents the summation index of all path patterns; and K represents the total number of all path patterns.

[0124] In this embodiment, the Dirichlet prior distribution is updated based on path pattern observation data to obtain the posterior distribution. This allows for continuous correction of probability distribution parameters using massive historical matching data, fitting the natural distribution patterns of various matching path patterns in real-world business scenarios, and achieving data-driven iterative optimization of the baseline model. By extracting the mode vector of the Dirichlet posterior distribution to construct the pattern distribution baseline, a standardized benchmark distribution that fits actual working conditions can be obtained, characterizing the steady-state distribution features of path patterns under normal matching strategies.

[0125] 2) Constructing attribute-level probability baselines can be achieved based on the following steps: Step G1: Divide the historical reference information according to the statistical window and information source to obtain multiple subsets of historical information; Step G2 is performed separately for multiple statistical windows of each information source: For each statistical window, the proportion of historical reference information with set attributes marked as conflict markers in the historical information subset within the statistical window is calculated to obtain the conflict ratio; Setting attributes refers to the matching dimensions that need to monitor the conflict rate. For example, institutions (O), departments (D), regions (L), and diseases (E) can be used as setting attributes.

[0126] A conflict flag refers to the state in which a given attribute is marked as conflict (X) in the corresponding path encoding information.

[0127] The conflict ratio refers to the proportion of path-coded information containing the conflict marker of this attribute in the total number of path-coded information in the historical information subset.

[0128] During implementation, all statistical windows under each information source are traversed, and the number of samples with conflict markers for the set attributes is counted for each window. The proportion of each sample in the historical information subset of the current window is calculated to obtain the conflict ratio for each window.

[0129] Step G3: Combine multiple statistical windows from the same information source based on the conflict ratio of a set attribute to obtain a conflict ratio sequence; A conflict ratio sequence is a discrete time series formed by arranging the conflict ratios calculated from different statistical windows under the same information source in chronological order.

[0130] During implementation, the window-by-window conflict ratios of single sources and single attributes are sorted and summarized in time sequence to form a complete and continuous time series of conflict ratios.

[0131] Step G4: Determine the mean and variance of the conflict ratio requirement to obtain the attribute-level probability baseline corresponding to the set attribute.

[0132] In practice, a Gaussian distribution can be used to fit the mean and variance of the conflict ratio to form an attribute-level probability baseline corresponding to each attribute and the source information.

[0133] In this embodiment, by splitting historical reference information according to statistical windows and information sources, the distribution characteristics of different attribute dimensions can be obtained for each information source. By statistically analyzing the proportion of attribute conflict markers set window by window, the corresponding conflict ratio can be obtained, which can characterize the distribution status of a certain attribute from the same source in different time periods. The mean and variance are calculated based on the conflict ratio sequence to construct an attribute-level probability baseline, which can simultaneously take into account the overall steady-state level and fluctuation characteristics, forming a normal distribution characteristic of a single dimension that is adapted to different sources.

[0134] 3) Constructing the path transformation funnel baseline can be achieved based on the following steps: Step H1: Divide the historical reference information into multiple subsets according to the matcher sequence; During implementation, the historical reference information is classified and collected according to the matcher sequence identifier associated with the historical reference information, and then split into multiple independent subsets of historical information corresponding to different matching execution links.

[0135] Step H2, performed separately for each matcher sequence: Step H21: Based on each business data sample in the subset of the matcher sequence: calculate the retention rate of each matcher on the candidate entity set of the matcher to obtain the retention rate sequence of the matcher sequence; The retention rate is the ratio of the number of remaining candidate entities after a single-layer matcher completes dimensional matching filtering to the total number of candidate entities before filtering. It is used to represent the candidate filtering and retention capability of a single-level matching link.

[0136] The retention rate sequence is a time-series statistical sequence composed of multiple hierarchical retention rates obtained by matching all business data samples layer by layer within the same matcher sequence dimension. For example, the matcher sequence includes matcher 1, matcher 2, and matcher 3. For business data 1, the retention rate of matcher 1 is P1, the retention rate of matcher 2 is P2, and the retention rate of matcher 3 is P3. Therefore, the final retention rate requirement for the matcher sequence for business data 1 is P1 P2 P3 arranged in sequence.

[0137] The retention rate sequence of the matcher sequence may include: the retention rate sequence for business data 1, the retention rate sequence for business data 2, ... the retention rate sequence for business data n.

[0138] Therefore, for the same matcher sequence, based on the processing of different business data, the retention rate sequence of the matcher sequence can be obtained.

[0139] Step H22: Based on the retention rate sequence of multiple business data samples in the subset, obtain the average retention rate of the same matcher; The mean retention rate is the steady-state mean obtained by statistically averaging the retention rates of all historical samples for a single matcher within the same matcher sequence.

[0140] During implementation, for each level of the matcher sequence, the retention rate of all business samples at each level is collected, and the mean retention rate of each level of matcher is calculated by averaging.

[0141] Step H23: Construct a tolerance interval based on the mean retention rate of each matcher to obtain the path transformation funnel baseline of the matcher sequence.

[0142] The tolerance range is a normal fluctuation range constructed based on the long-term fluctuation characteristics of the retention rate of each level of matcher, and is used to define the reasonable fluctuation range of the matching retention rate.

[0143] The path transition funnel baseline is a hierarchical funnel baseline model composed of the average retention rate of each level of matching and the corresponding tolerance interval. For example, the matcher sequence includes: matcher 1, whose tolerance interval is interval 1 determined by its average retention rate; matcher 2, whose tolerance interval is interval 2 determined by its average retention rate; and matcher 3, whose tolerance interval is interval 3 determined by its average retention rate. The final path transition funnel baseline for this matcher sequence is expressed as (interval 1, interval 2, interval 3). Each interval corresponds to its respective matcher.

[0144] During implementation, the standard deviation and historical fluctuation characteristics of the retention rate at each level can be combined to configure a corresponding tolerance range for the mean retention rate at each level. The benchmark mean and tolerance range of all levels can be integrated to form a complete path conversion funnel baseline that adapts to each matcher sequence.

[0145] In this embodiment, the mean retention rate of each matcher is calculated based on the retention rate sequence of batch samples, which effectively eliminates the random fluctuation interference of a single matching sample and obtains a quantitative indicator that can characterize the steady-state screening capability of the matcher. By constructing a tolerance range for the mean retention rate of each matcher, a path transition funnel baseline adapted to multi-level progressive matching logic is formed, which can fully cover the reasonable fluctuation range of the matching link under normal operating conditions. The final path transition funnel baseline can express the normal state characteristics from the dimension of the matcher sequence.

[0146] After obtaining the baseline model, subsequent matching strategies can be monitored and analyzed to detect anomalies in a timely manner. During implementation, by quantitatively comparing the differences between batch encoded features and the baseline model, anomalies in the matching process and potential causes can be identified. During implementation, such as... Figure 5 As shown, it may include the following: S501, counts the first path encoding information generated in real time within the current time window.

[0147] The current time window is a fixed time segment interval for batch aggregation and statistics of multiple first path encoding information, which can be divided into minute-level or hour-level sliding window methods.

[0148] It should be noted that each first path encoding information follows the aforementioned [N#][O#][D#][L#][E#][C#][S#][M#] format and is associated with metadata.

[0149] S502 generates coding distribution features based on the coding information of multiple first paths.

[0150] The coding distribution characteristics are a set of multidimensional quantifiable features derived from multidimensional statistical analysis of multiple first path coding information collected within a time window.

[0151] This feature may include at least one of three sub-features: (1) The path pattern distribution vector is generated in the same way as the pattern distribution baseline, and will not be described again here; The path pattern distribution vector is used to compare with the pattern distribution baseline. The similarity between the two can be calculated to determine whether they match. If the similarity is greater than the first similarity threshold, the two are determined to match.

[0152] (2) The conflict rate features by source and dimension are generated in the same way as the attribute-level probability baseline, except that the statistical objects are different, which will not be elaborated here.

[0153] Source-specific and dimension-specific conflict rate features are used to express attribute characteristics of a single attribute. The source-specific and dimension-specific conflict rate sequence is compared with the attribute-level probability baseline. The source-specific and dimension-specific conflict rate features of the same attribute can be compared with the normal range expressed by the mean and variance of that attribute in its attribute-level probability baseline. If the source-specific and dimension-specific conflict rate features are not within this normal range, it indicates a mismatch between the two in that attribute dimension.

[0154] (3) The retention rate of the path transformation funnel is generated in the same way as the baseline of the path transformation funnel, so it will not be described again here.

[0155] The path conversion funnel retention rate feature is used to compare with the path conversion funnel baseline. It can determine whether the path conversion funnel retention rate feature is within the tolerance range expressed by the path conversion funnel baseline in the corresponding path mode. If it exceeds the tolerance range, it indicates that the two do not match.

[0156] S503 matches the coding distribution features with the baseline model to obtain the matching results.

[0157] During implementation, a preset quantification difference algorithm can be used to compare the deviation between the coding distribution characteristics and the standard statistical laws of the corresponding baseline model to obtain the matching result. The matching result is used to indicate whether the coding distribution characteristics match the corresponding baseline model.

[0158] S504, the matching result indicates that when the coding distribution characteristics do not match the baseline model, root cause analysis is performed based on the coding information of multiple first paths.

[0159] Root cause analysis is a failure localization process that is carried out based on the first path coding information and structured metadata within the current window after a mismatch is detected between the coding distribution characteristics and the baseline model.

[0160] In this embodiment of the disclosure, by statistically analyzing multiple first path coding information generated in real time within a preset time window and generating coding distribution features, it is possible to transform single discrete entity matching path data into quantifiable and statistical batch operation features, thereby realizing the abnormal perception of the running status of the matching strategy based on the first path coding information, and performing root cause analysis accordingly.

[0161] To identify the core business attributes causing disambiguation matching anomalies and locate hidden failures at the data level, root cause analysis can be performed based on the first path encoding information. This can be achieved using a feature replay comparison method, with the specific implementation steps as follows: Step R1: Obtain the audit tags for multiple first path encoding information; An audit tag is a qualitative marker attached to each first path code, verified by manual review or subsequent feedback, used to define the correctness of the matching result. This audit tag can include different types, such as "matching result is correct" or "matching result contains an error." Each first path code is uniquely associated with one audit tag, stored in the metadata field of the path code log. First path codes that have not undergone manual review may not have an audit tag.

[0162] During implementation, the metadata bound to all first path encoding information in the current window can be read, and the audit tag field corresponding to each record can be extracted.

[0163] Step R2: Filter out the first path encoding information that indicates an error in the matching result due to the review label, and obtain the error sample set; Step R3: Filter out the first path encoding information that indicates a correct matching result based on the audit tags to obtain the normal sample set; Step R4: For the specific attribute to be analyzed, determine the similarity between the original business data of each sample in the error sample set and the matched candidate entity on the specific attribute, and obtain the first similarity distribution. Specific attributes can be any business dimension feature, such as institution name, department name, disease description, location text, etc.

[0164] The first similarity distribution is a statistical distribution sequence consisting of all similarity values ​​of the erroneous samples.

[0165] During implementation, the similarity under a specific attribute can be calculated for each sample based on the original business data characteristics and corresponding candidate entity characteristics of each sample in the error sample set, and all similarity values ​​can be summarized to form the first similarity distribution.

[0166] Step R5: Determine the similarity of the original business data of each sample in the normal sample set with the matched candidate entities on specific attributes to obtain the second similarity distribution; The second similarity distribution is a standard distribution consisting of similarity values ​​calculated from normal samples under the same specific attribute.

[0167] In practice, the similarity under specific attributes can be calculated for each sample based on the original business data features and corresponding candidate entity features of the normal sample set, and the results can be aggregated to obtain the second similarity distribution.

[0168] Step R6: Based on the first similarity distribution and the second similarity distribution, if a specific attribute meets the preset significance condition, the root cause analysis result is determined to be: in the information dimension corresponding to the specific attribute, the preset field has missing information or the matching strategy of the matcher is ineffective.

[0169] The pre-defined saliency criteria are used to determine, through statistical testing, whether there is a non-accidental substantial difference between the first and second similarity distributions, thereby identifying the attribute dimensions that cause matching errors. The specific determination process is as follows: First, the continuous similarity values ​​in the first and second similarity distributions are discretized and divided into intervals according to the preset numerical range. The frequency of samples in each similarity interval is counted to determine the proportion of the error sample set and the normal sample set in different similarity intervals.

[0170] Secondly, an association table is constructed with similarity intervals as rows and sample categories as columns, including both erroneous and normal samples. A chi-square test is then performed on the association table to obtain the chi-square statistic and its corresponding p-value.

[0171] When the p-value obtained from the chi-square test is less than the preset significance level (the preset significance level is the preset value), it proves that the difference between the two distributions is statistically significant, and the proportion of the erroneous sample set in the low similarity interval is higher than that of the normal sample set. It is then determined that the specific attribute to be analyzed meets the preset significance condition, and the root cause analysis result can be determined as follows: on the information dimension corresponding to the specific attribute, the preset field has missing information or the matching strategy of the matcher is ineffective.

[0172] In this context, "missing field information" refers to the absence of specific attributes or their associated aliases in the preset dictionary. This indicates the emergence of new automatic fields, requiring maintenance and updates to the preset dictionary.

[0173] Matching strategy failure means that the matching rules and parameters in the matcher used to match a specific attribute may not be able to cope with the current new situation and need to be optimized.

[0174] In this embodiment, for specific attributes, the similarity of business data and candidate entities between erroneous and normal samples is statistically analyzed to form differentiated first and second similarity distributions. This allows for a quantitative comparison of the feature differences between correct and incorrect matches on a single attribute dimension, transforming previous experience-based qualitative investigations into data-driven quantitative analysis. When the dual similarity distributions meet a significant condition, the root cause analysis results can be determined. This approach can identify potential problems based on the first path encoding information, helping to optimize the matching strategy.

[0175] To identify dimensional conflicts and disambiguation performance degradation caused by external dictionary updates, root cause analysis can be performed based on the first path encoding information. The dictionary update association analysis method can be used, and the specific implementation steps are as follows: Step T1: The baseline model includes an attribute-level conflict probability baseline. When an abnormal attribute is identified based on the attribute-level conflict probability baseline, the proportion of the second path coding information in each time window is statistically analyzed to obtain the conflict rate time series. The second path coding information is: the first path coding information of the abnormal attribute with conflict identifier. Abnormal attributes are attribute dimensions whose conflict rates deviate from the normal range and exhibit abnormal fluctuations, as determined by the attribute-level conflict probability baseline.

[0176] The second path encoding information refers to the first path encoding information that carries the conflict identifier "X" in the corresponding time window for this anomaly attribute dimension.

[0177] The conflict rate time series is a continuous time series statistical sequence formed by taking a preset time window as the statistical unit, statistically analyzing the proportion of the second path code information corresponding to the abnormal attribute in the total path code information in the corresponding window, and then arranging them in chronological order.

[0178] Step T2: Based on the change point detection algorithm, identify the mutation time when the anomaly begins to occur from the conflict rate time series; Change point detection algorithms are time series analysis algorithms capable of identifying abrupt changes in the statistical characteristics of time series data. Algorithms that can be used include PELT (PrunedExactLinearTime), CUSUM (cumulative sum), or Bayesian change point detection.

[0179] The mutation time is the precise point in time within the conflict rate time series where the conflict probability abruptly increases from the normal steady-state level, producing a statistical anomaly. The mutation time represents the starting point at which the disambiguation performance of the matching strategy begins to degrade. The mutation time is the temporal representation of the time window that produces the mutation. It can be either the start or end time of that time window.

[0180] Step T3: Construct a time range based on the mutation time; The time range is a time screening interval formed by extending the preset buffer time forward and backward with the mutation time as the core benchmark. It is used to cover all potential dictionary change behaviors before and after the mutation occurs, so as to avoid missing change operations at nearby time points.

[0181] During implementation, the target time range for screening dictionary change records is constructed by taking the identified mutation time as the center and combining it with the preset buffer window of the operation and maintenance update cycle.

[0182] Step T4: Retrieve change records within the time range from the field change log of the preset dictionary; The field change log is a retained dictionary operation log, which fully records the addition, deletion, modification, and hierarchical adjustment of various dimension dictionaries and entity field libraries. Each change record carries the operation timestamp, change dimension, and specific change content.

[0183] During implementation, based on the time range in which the database was built, the field change logs are retrieved, and all dictionary change records with timestamps falling within that range are filtered in batches.

[0184] In step T5, if the change record contains change information for the abnormal attribute, the root cause analysis result is determined to be that the dictionary change caused the abnormality.

[0185] Change information for abnormal attributes refers to the adjustments and updates made to the entity dictionary, field standards, hierarchical mapping, and alias library corresponding to the current abnormal attribute dimension in the dictionary change record.

[0186] During implementation, each change record obtained from the screening is compared with the abnormal attribute dimension. If there is a matching dimension change content and the change time is highly consistent with the conflict rate mutation time, then the root cause of this disambiguation anomaly is determined to be a dictionary change, and the corresponding root cause analysis results are output.

[0187] In this embodiment, abnormal attributes are identified based on the attribute-level conflict probability baseline, and a conflict rate time series is constructed by statistically analyzing the proportion of second path encoding information carrying conflict identifiers in each time window. This transforms discrete attribute conflict events into time-trackable quantitative change curves, characterizing the dynamic evolution of attribute conflicts during entity matching. By introducing a change point detection algorithm to analyze the conflict rate time series, the abrupt transition time of abnormal states can be located. Based on the abrupt transition time to lock the time range and match field change logs, attribute dictionary change records within abnormal periods can be specifically screened. When corresponding abnormal attribute change information exists, the root cause is determined to be a matching anomaly caused by a dictionary change. This provides guidance for optimizing matching strategies.

[0188] To effectively investigate disambiguation anomalies caused by internal component updates such as matcher rule iteration and parameter threshold adjustment, root cause analysis can be performed based on the first path encoding information. The matcher logic verification method can be used, and the specific implementation steps are as follows: Step Y1: From multiple first path coding information, filter out the first path coding information with the correct review label as the correct sample; A correct sample refers to the first set of path coding information marked by the audit label as having a correct matching result and a reliable disambiguation status. It is used as a benchmark sample to verify the logical consistency of the matcher. The original disambiguation result and path coding status corresponding to the correct sample are the standard output results under normal system operation.

[0189] During implementation, the multiple first path code information collected are traversed, and the information is filtered according to the bound manual review labels. The path code information with correct review results is collected in a unified manner to construct a correct sample set.

[0190] Step Y2: Obtain the original attribute set, the original candidate entity set, and the first path encoding information corresponding to the correct sample; The original attribute set consists of the original business input feature data corresponding to the correct sample, such as the original business field content of each matching dimension, including organization, department, disease, location, etc.

[0191] The original candidate entity set is the entity candidate set input to the matcher during the historical disambiguation process.

[0192] Step Y3: Based on the matcher sequence corresponding to the source identifier of the correct sample, process the original attribute set and the original candidate entity set to obtain the path encoding information to be matched for the correct sample. Step Y4: Match the path encoding information to be matched with the first path encoding information of the correct sample to identify the different encoding bits; The encoded bits are independent marker fields corresponding to each dimension in the path encoding, including the status identifiers of each dimension: N, O, D, L, E, C, S, and M.

[0193] The difference encoding bit is the field where the dimension marker value has changed compared to the historical baseline first path encoding information in the replay-generated matching path encoding. It is used to locate the specific processing step where the matching behavior has changed.

[0194] During implementation, the old and new path codes are compared dimension by dimension and bit by bit, and the dimension codes with numerical differences are recorded one by one to determine the specific location and dimension type where the matching behavior changes.

[0195] Step Y5: If there are update records for the differing coding bits in the update log of the matcher sequence corresponding to the source identifier of the correct sample, then the root cause analysis result is determined to be that the anomaly is caused by the matcher update.

[0196] The update log is used to fully record information on changes such as parameter adjustments, rule iterations, script updates, and threshold modifications for each matcher, and each update record corresponds to a specific change dimension and effective time.

[0197] Matcher update causing anomalies refers to changes in the matcher rules, parameters, or orchestration logic that cause the same original input data to produce different path encoding matching results, thus triggering anomalies.

[0198] During implementation, the update log of the matcher corresponding to the dimension to which the difference coding bit belongs is queried. If there is an update or change record of the matcher for the corresponding dimension during the abnormal period, the root cause of this disambiguation anomaly is determined to be the matcher update, and the corresponding root cause analysis results are output.

[0199] In this embodiment, by selecting the first path encoding information with correct verification labels as the correct sample, the root cause interference problem caused by using incorrect samples for analysis can be avoided, ensuring the reliability and accuracy of the analysis benchmark. By comparing the reproduced path encoding information to be matched with the original first path encoding information bit by bit, the encoding difference sites can be located, and subtle changes in matching can be quickly captured. When there is an update record with a difference in encoding positions in the matcher update log, the abnormal root cause is determined to be caused by the matcher update iteration. This can identify anomalies caused by matcher changes and provide guidance information for matching strategy updates.

[0200] Based on the same technical concept, this disclosure also provides an information matching device 600, such as... Figure 6 As shown, it includes: The acquisition module 601 is used to acquire the business data to be processed and the source identifier of the business data to be processed. Extraction module 602 is used to extract at least one initial attribute information of the entity to be processed from the business data to be processed; The matching module 603 is used to perform a matching operation between at least one initial attribute information and a preset dictionary to obtain the attribute set corresponding to each of the at least one initial attribute information. The processing module 604 is used to process the attribute set corresponding to each matcher in the matcher sequence based on the source identifier in order to perform matching operation between the entity to be processed and the candidate entity set.

[0201] Each matcher in the matcher sequence includes multiple matching conditions with different precision levels; The processing module is specifically used for: For each matcher, execute the following separately: Obtain the candidate entity set and the attribute set corresponding to the matcher as the target attribute set; According to the order of accuracy level from high to low, the target attribute set and the candidate entity set are compared with each other according to the matching conditions, and the candidate entity corresponding to the matching condition that matches in the first comparison is determined as the successor entity. If multiple successor entities are matched and the matcher is not the last matcher in the matcher sequence, the multiple successor entities are used as the candidate entity set for the next matcher of the matcher.

[0202] The matcher also includes conflict determination rules; the device also includes: The conflict identification module is used to filter out candidate entities in the candidate entity set that meet the target condition based on the conflict determination rules. The target condition is that there is an information conflict between the candidate entity and the target attribute set.

[0203] Also includes: The subcode generation module is used to generate a matcher subcode for each enabled matcher in the matcher sequence; the matcher subcode includes at least the matcher identifier of the matcher. The encoding module is used to concatenate the subcodes of the enabled matchers in the matcher sequence to obtain the first path encoding information.

[0204] The subcode generation module is specifically used for: If the matcher finds a subsequent entity, it generates a precision identifier for the precision level corresponding to the matching condition that matches the match. Generate the matcher subcode based on the matcher identifier and the precision identifier.

[0205] The sub-code generation module is specifically used to: generate conflict identifiers when all candidate entities in the candidate entity set and the target attribute set have information conflicts based on the conflict determination rules of the matcher; Generate matcher subcodes based on matcher identifiers and conflict identifiers.

[0206] Also includes: The filtering module is used to select an initial attribute as a subsequent attribute when a single successor entity is matched and at least one unused initial attribute exists in at least one initial attribute information. The verification module is used to verify subsequent entities based on the matchers corresponding to the posterior attributes in the matcher sequence, and obtain the verification results.

[0207] Also includes: The optimization module is used to input the attribute sets corresponding to the multiple successor entities and at least one initial attribute information into the large model when multiple successor entities are matched and the matcher is the last matcher, and to obtain the matching result of the large model. Add a module to add the model identifier of the large model and the result identifier of the matching result to the first path encoding information.

[0208] Also includes: The traffic splitting module is used to match the first path encoding information with multiple preset traffic splitting rules to determine the target rule corresponding to the first path encoding information. The storage module is used to associate and store the first path encoding information, the matching records of the matcher sequence for the business data to be processed, and the target queue corresponding to the target rule.

[0209] Also includes: The statistics module is used to collect statistics on the encoding information of multiple first path lines generated in real time within the current time window; The feature generation module is used to generate coding distribution features based on the coding information of multiple first paths; The feature matching module is used to match the encoded distribution features with the baseline model to obtain the matching results; The analysis module is used to perform root cause analysis based on the coding information of multiple first paths when the matching result indicates that the coding distribution characteristics do not match the baseline model.

[0210] Also includes: The aggregation module is used to obtain historical reference information; the historical reference information includes multiple first path encoding information and the metadata corresponding to the first path encoding information. The building module is used to construct a baseline model based on historical reference information.

[0211] The analysis module includes: The first acquisition unit is used to acquire the audit tags of multiple first path encoding information; The first filtering unit is used to filter out the first path encoding information indicating that the matching result of the audit tag is incorrect, and obtain the error sample set. The second filtering unit is used to filter out the first path encoding information that indicates the correct matching result of the audit label, and obtain the normal sample set. The first calculation unit is used to determine the similarity between the original business data of each sample in the error sample set and the matched candidate entity on the specific attribute to be analyzed, and to obtain the first similarity distribution. The second calculation unit is used to determine the similarity between the original business data of each sample in the normal sample set and the matched candidate entities on specific attributes, and to obtain the second similarity distribution. The first analysis unit is used to determine the root cause analysis result based on the first similarity distribution and the second similarity distribution, when a specific attribute meets the preset significance condition: the preset field has missing information or the matching strategy of the matcher fails in the information dimension corresponding to the specific attribute.

[0212] The analysis module includes: The second acquisition unit is used to calculate the proportion of the second path coding information in each time window to obtain the conflict rate time series when the baseline model includes the attribute-level conflict probability baseline and abnormal attributes are identified based on the attribute-level conflict probability baseline; the second path coding information is: the first path coding information of the abnormal attribute with conflict identification. The detection unit is used to identify the time of abrupt change that begins to produce anomalies from the conflict rate time series based on the change point detection algorithm. Building blocks are used to construct time ranges based on mutation times; The third acquisition unit is used to acquire change records within a time range from the field change log of a preset dictionary; The second analysis unit is used to determine that the root cause analysis result is caused by dictionary changes when the change record contains change information for abnormal attributes.

[0213] The analysis module includes: The fourth acquisition unit is used to filter out the first path code information with the correct audit label from multiple first path code information and use it as the correct sample; The fifth acquisition unit is used to acquire the original attribute set, the original candidate entity set, and the first path encoding information corresponding to the correct sample. The restoration unit is used to process the original attribute set and the original candidate entity set based on the matcher sequence corresponding to the source identifier of the correct sample, and obtain the path encoding information to be matched for the correct sample. The first matching unit is used to perform a matching operation between the path encoding information to be matched and the first path encoding information of the correct sample, and to identify the encoding bits that are different. The third analysis unit is used to determine that the root cause analysis result is that the anomaly is caused by the matcher update, in the update log of the matcher sequence corresponding to the source identifier of the correct sample, where there is an update record for the differential coding bit.

[0214] Build modules, including: The sixth acquisition unit is used to count the number of occurrences of each path pattern in the historical reference information, as observation data; The update unit is used to update the Dirichlet prior distribution based on the observed data to obtain the Dirichlet posterior distribution. The first building unit is used to extract the mode vector of the Dirichlet posterior distribution to obtain the pattern distribution baseline. The baseline model includes the pattern distribution baseline.

[0215] Build modules, including: The first partitioning unit is used to divide historical reference information according to statistical windows and information sources, resulting in multiple subsets of historical information. The first execution unit is used to perform the following for each information source's multiple statistical windows: for each statistical window, it calculates the proportion of historical reference information with set attributes marked as conflict markers in the historical information subset within the statistical window to obtain the conflict ratio; The splicing unit is used to splice multiple statistical windows from the same information source based on the conflict ratio of a set attribute to obtain a conflict ratio sequence. The second building unit is used to determine the mean and variance of the conflict ratio requirement and obtain the attribute-level probability baseline corresponding to the set attribute; the baseline model includes the attribute-level probability baseline.

[0216] Build modules, including: The second partitioning unit is used to divide the historical reference information into multiple subsets according to the matcher sequence; The second execution unit is used to execute separately for each matcher sequence: For each business data sample in a subset of the matcher sequence: calculate the retention rate of each matcher on the candidate entity set of the matcher to obtain the retention rate sequence of the matcher sequence; Based on the retention rate sequence of multiple business data samples in the subset, the average retention rate of the same matcher is obtained; For each matcher, the mean retention rate is used to construct a tolerance interval, resulting in the path transformation funnel baseline of the matcher sequence. The baseline model includes the path transformation funnel baseline.

[0217] The specific functions and examples of each module and submodule of the apparatus in this disclosure can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0218] The acquisition, storage, and application of any type of information, such as user personal information, involved in the technical solutions disclosed herein comply with relevant laws and regulations and do not violate public order and good morals.

[0219] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0220] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0221] like Figure 7As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded into random access memory (RAM) 703 from storage unit 708. The RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0222] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0223] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as the information matching method. For example, in some embodiments, the information matching method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the information matching method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the information matching method by any other suitable means (e.g., by means of firmware).

[0224] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0225] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0226] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0227] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0228] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0229] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0230] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0231] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information matching method, comprising: Obtain the business data to be processed and the source identifier of the business data to be processed; Extract at least one initial attribute information of the entity to be processed from the business data to be processed; The at least one initial attribute information is matched with a preset dictionary to obtain the attribute set corresponding to each of the at least one initial attribute information; Based on multiple matchers in the matcher sequence corresponding to the source identifier, the attribute sets corresponding to each matcher are processed sequentially to perform matching operations on the entity to be processed and the candidate entity set.

2. The method according to claim 1, wherein, Each matcher in the matcher sequence includes multiple matching conditions with different precision levels; The process involves sequentially processing the attribute sets corresponding to each matcher in the matcher sequence corresponding to the source identifier, including: For each matcher, execute the following separately: Obtain the candidate entity set and the attribute set corresponding to the matcher as the target attribute set; According to the order of accuracy level from high to low, the target attribute set and the candidate entity set are compared with each other according to the matching conditions, and the candidate entity corresponding to the matching condition that matches in the first comparison is determined as the successor entity. If multiple successor entities are matched and the matcher is not the last matcher in the matcher sequence, the multiple successor entities are taken as the candidate entity set for the next matcher of the matcher.

3. The method according to claim 2, wherein the matcher further includes conflict determination rules; the method further includes: Based on the conflict determination rule, candidate entities in the candidate entity set that meet the target condition are filtered out. The target condition is that the candidate entity has an information conflict with the target attribute set.

4. The method according to claim 2, further comprising: For each enabled matcher in the matcher sequence, generate the matcher subcode for that matcher; The matcher subcode includes at least the matcher identifier of the matcher; The matching subcodes of the enabled matching machines in the matching machine sequence are concatenated to obtain the first path encoding information.

5. The method according to claim 4, wherein, The process of generating the matcher subcode includes: When the matcher matches a subsequent entity, it generates a precision identifier for the precision level corresponding to the matching condition that matches the match. The matcher subcode is generated based on the matcher identifier and the precision identifier.

6. The method according to claim 4, wherein, The process of generating the matcher subcode includes: Based on the conflict determination rules of the matcher, if it is determined that all candidate entities in the candidate entity set have information conflicts with the target attribute set, a conflict identifier is generated. The matcher subcode is generated based on the matcher identifier and the conflict identifier.

7. The method according to claim 2, further comprising: If a single successor entity is matched and at least one unused initial attribute exists in the at least one initial attribute information, an initial attribute is selected from the at least one unused initial attribute as a subsequent attribute. Based on the matcher corresponding to the posterior attribute in the matcher sequence, the successor entity is validated to obtain the validation result.

8. The method according to claim 4, further comprising: When multiple successor entities are matched and the matcher is the last matcher, the attribute sets corresponding to the multiple successor entities and the at least one initial attribute information are input into the large model to obtain the matching result of the large model. The model identifier of the large model and the result identifier of the matching result are added to the first path encoding information.

9. The method according to claim 4, further comprising: The first path encoding information is matched with multiple preset traffic splitting rules to determine the target rule corresponding to the first path encoding information; The first path encoding information and the matching records of the matcher sequence for the business data to be processed are associated and stored in the target queue corresponding to the target rule.

10. The method of claim 4, further comprising: Statistically analyze the encoding information of multiple first path lines generated in real time within the current time window; Based on the encoding information of the multiple first paths, an encoding distribution feature is generated; The encoded distribution features are matched with the baseline model to obtain the matching results; The matching result indicates that, in the case where the coding distribution features do not match the baseline model, root cause analysis is performed based on the coding information of the multiple first paths.

11. The method of claim 10, further comprising: Obtain historical reference information; The historical reference information includes multiple first path encoding information and metadata corresponding to the first path encoding information; The baseline model is constructed based on the historical reference information.

12. The method according to claim 10, wherein, The root cause analysis based on the multiple first path encoding information includes: Obtain the review tags for the multiple first path encoding information; The first path encoding information indicating that the matching results contain errors is filtered out by the review tags, thus obtaining the set of error samples; The first path encoding information indicating a correct matching result is selected from the audit tags to obtain the normal sample set; For a specific attribute to be analyzed, the similarity between the original business data of each sample in the error sample set and the matched candidate entity on the specific attribute is determined, and a first similarity distribution is obtained. The similarity between the original business data of each sample in the normal sample set and the matched candidate entity on the specific attribute is determined to obtain the second similarity distribution; Based on the first similarity distribution and the second similarity distribution, if the specific attribute meets the preset saliency condition, the root cause analysis result is determined to be: in the information dimension corresponding to the specific attribute, the preset field has missing information or the matcher's matching strategy is ineffective.

13. The method according to claim 10, wherein, The root cause analysis based on the multiple first path encoding information includes: The baseline model includes an attribute-level conflict probability baseline. When an abnormal attribute is identified based on the attribute-level conflict probability baseline, the proportion of the second path coding information within each time window is statistically analyzed to obtain a conflict rate time series. The second path coding information is the first path coding information of the abnormal attribute with a conflict identifier. Based on the change point detection algorithm, the mutation time when the abnormality begins to occur is identified from the conflict rate time series; A time range is constructed based on the mutation time; Obtain the change records within the specified time range from the field change log of the preset dictionary; If the change record contains change information for the abnormal attribute, the root cause analysis result is determined to be that the dictionary change caused the abnormality.

14. The method of claim 10, wherein, The root cause analysis based on the multiple first path encoding information includes: From the multiple first path coding information, the first path coding information with the correct review label is selected as the correct sample; Obtain the original attribute set, the original candidate entity set, and the first path encoding information corresponding to the correct sample; Based on the matcher sequence corresponding to the source identifier of the correct sample, process the original attribute set and the original candidate entity set to obtain the path encoding information to be matched for the correct sample; The path encoding information to be matched is matched with the first path encoding information of the correct sample to identify the encoding bits that are different. If, in the update log of the matcher sequence corresponding to the source identifier of the correct sample, there is an update record for the differing coding bit, the root cause analysis result is determined to be that the anomaly is caused by the matcher update.

15. The method according to claim 11, wherein, The construction of the baseline model based on the historical reference information includes: The frequency of occurrence of each path pattern in historical reference information is counted as observation data; Based on the observed data, the Dirichlet prior distribution is updated to obtain the Dirichlet posterior distribution. The mode vector of the Dirichlet posterior distribution is extracted to obtain the pattern distribution baseline, and the baseline model includes the pattern distribution baseline.

16. The method according to claim 11, wherein, The construction of the baseline model based on the historical reference information includes: The historical reference information is divided according to statistical windows and information sources to obtain multiple subsets of historical information; For each information source, perform the following steps for multiple statistical windows: For each statistical window, calculate the proportion of historical reference information whose set attributes are marked as conflict markers within the historical information subset of the statistical window to obtain the conflict ratio; By combining the conflict ratios of multiple statistical windows from the same information source for the specified attribute, a conflict ratio sequence is obtained. The mean and variance of the conflict ratio requirement are determined to obtain the attribute-level probability baseline corresponding to the set attribute; the baseline model includes the attribute-level probability baseline.

17. The method according to claim 11, wherein, The construction of the baseline model based on the historical reference information includes: The historical reference information is divided into multiple subsets according to the matcher sequence; Execute separately for each matcher sequence: Based on each business data sample in a subset of the matcher sequence: calculate the retention rate of each matcher for the candidate entity set of the matcher, and obtain the retention rate sequence of the matcher sequence; Based on the retention rate sequence of multiple business data samples in the subset, the average retention rate of the same matcher is obtained; For each matcher, a tolerance interval is constructed based on the mean retention rate, resulting in the path transformation funnel baseline of the matcher sequence. The baseline model includes the path transformation funnel baseline.

18. An information matching device, comprising: The acquisition module is used to acquire the business data to be processed and the source identifier of the business data to be processed; The extraction module is used to extract at least one initial attribute information of the entity to be processed from the business data to be processed; The matching module is used to perform a matching operation between the at least one initial attribute information and a preset dictionary to obtain the attribute set corresponding to the at least one initial attribute information respectively; The processing module is used to sequentially process the attribute sets corresponding to each matcher in the matcher sequence corresponding to the source identifier, so as to perform a matching operation between the entity to be processed and the candidate entity set.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-17.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-17.

21. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-17.

Images