Data matching method, apparatus, device, and medium

Abstract

The application discloses a data matching method and device, equipment and medium. Based on a plurality of preset matching modes, the matching degrees between candidate data and user input data under each preset matching field are calculated respectively, and weighted summation is performed according to preset weight values corresponding to each preset matching field to obtain the comprehensive matching degrees of each candidate. Finally, at least one target matching item is selected from a plurality of candidate items through a preset matching degree threshold to generate a data matching result. By introducing a knowledge graph expansion matching mechanism, the data matching process has semantic understanding ability, can identify concepts with different literal forms but related semantics, effectively avoids missed judgment caused by literal differences, and significantly improves the recall rate of the matching result. Meanwhile, the weighted summation mechanism replaces the traditional serial rule chain, overcomes the unreasonable problem that a single field matching failure leads to overall rejection, makes the determination of the matching result more scientific and objective, and effectively improves the accuracy of the data matching result.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a data matching method, apparatus, device, and medium. Background Technology

[0002] Currently, most university research management systems have automatic data matching functions in their achievement modules and scholar database modules. For example, they can automatically match research achievements with international standard journal numbers and inclusion types, or match the most suitable scholars for a research task. Existing data matching typically uses keyword-based exact matching, comparing the user's input data in each matching field with keywords from candidate options in a pre-set database, outputting a binary conclusion of match / no match based on whether they are completely identical. However, existing data matching methods rely on literal keyword comparison and lack semantic understanding capabilities. They cannot identify semantically related but literally different concepts, leading to the omission of many relevant achievements or scholars, resulting in low accuracy of the matching results. Summary of the Invention

[0003] In view of this, the present invention provides a data matching method, apparatus, device and medium to solve the technical problem that the lack of semantic understanding ability in existing data matching methods leads to low accuracy of matching results.

[0004] Firstly, a data matching method is provided, which includes: In response to a matching task request, obtain the task type included in the matching task request, as well as the user's input data for at least one preset matching field corresponding to the task type; Based on the task type, candidate data for multiple candidate options are retrieved from a pre-set database; The matching degree between the candidate data and the target data of each candidate option in each preset matching field is calculated using multiple preset matching methods. Among them, the multiple preset matching methods include at least one of the following: exact matching method, fuzzy matching method and knowledge graph extended matching method. Based on the task type and the set of preset matching rules, the target weight value of each preset matching field is determined. The set of preset matching rules includes the preset weight value corresponding to each preset matching field under each task type. The matching degree of each preset matching field and the target weight value are weighted and summed to generate the comprehensive matching degree of each candidate option; Based on the preset matching threshold corresponding to the task type and the comprehensive matching degree of each candidate option, at least one target matching item is determined from multiple candidate options; Based on the overall matching degree of each target matching item, candidate data, and matching degree details of each preset matching field, data matching results are generated.

[0005] Secondly, a data matching device is provided, the device comprising: The first acquisition module is used to respond to the matching task request and acquire the task type included in the matching task request, as well as the user's input data for at least one preset matching field corresponding to the task type. The second acquisition module is used to acquire candidate data of multiple candidate items from a preset database based on the task type; The calculation module is used to calculate the matching degree between the candidate data and the target data of each candidate option on each preset matching field using multiple preset matching methods. The multiple preset matching methods include at least one of the following: exact matching method, fuzzy matching method and knowledge graph extended matching method. The first determining module is used to determine the target weight value of each preset matching field based on the task type and the preset matching rule set, wherein the preset matching rule set includes the preset weight value corresponding to each preset matching field under each task type; The first generation module is used to perform a weighted summation of the matching degree of each preset matching field and the target weight value to generate the comprehensive matching degree of each candidate option; The second determining module is used to determine at least one target matching item among multiple candidate items based on the preset matching degree threshold corresponding to the task type and the comprehensive matching degree of each candidate item; The second generation module is used to generate data matching results based on the comprehensive matching degree of each target matching item, candidate data, and matching degree details of each preset matching field.

[0006] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described data matching method.

[0007] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described data matching method.

[0008] The aforementioned data matching method, apparatus, equipment, and medium utilize three pre-defined matching methods—precise matching, fuzzy matching, and knowledge graph-extended matching—to calculate the matching degree between candidate data and user input data for each candidate option under each preset matching field. Then, a weighted sum is performed based on the preset weight values ​​corresponding to each preset matching field to obtain the comprehensive matching degree of each candidate option. Finally, at least one target matching item is selected from multiple candidate options using a preset matching degree threshold, generating a data matching result containing detailed matching degree information. By introducing a knowledge graph-extended matching mechanism, the data matching process acquires semantic understanding capabilities, enabling it to identify semantically related concepts that differ in literal meaning, effectively avoiding missed judgments due to literal differences, thereby significantly improving the recall rate of the matching results. Simultaneously, the weighted summation mechanism replaces the traditional sequential rule chain execution method, overcoming the unreasonable problem of overall rejection due to a single field matching failure, making the judgment of matching results more scientific and objective, and comprehensively improving the accuracy and rationality of the data matching results. Attached Figure Description

[0009] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating a data matching method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a data matching device in one embodiment of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in the present invention are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of the present invention.

[0011] Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Moreover, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0012] Furthermore, the embodiments described herein are merely some, not all, of the embodiments of the invention. The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0013] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of a feature subsequently declared, but does not exclude the addition of other features. It should also be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0014] Currently, the research management system of universities generally has automatic data matching functions in modules such as the results module and the scholar database module. For example, it can automatically match the International Standard Serial Number (ISSN) and the inclusion type (SCI / EI / core journal) for research results, or match the most suitable scholar for a research task (such as peer review).

[0015] However, existing data matching methods have the following significant drawbacks: First, the matching process usually relies on simple keyword exact matching. For example, when matching research directions, if the configured keyword is neural network, it cannot identify related but literally different concepts such as deep learning and CNN, resulting in a large number of related achievements or scholars being missed.

[0016] Secondly, matching rules are usually executed sequentially, and the matching result of a preceding rule directly determines whether subsequent rules are executed. This veto mechanism may cause a minor field to fail to match, thus rejecting highly relevant matches.

[0017] Third, existing data matching methods typically only output binary results of match or no match, failing to quantify and rank the matching quality. Users cannot discern which matches are highly accurate and which are merely average, reducing the operability of the matching results.

[0018] Fourth, existing data matching methods typically do not explain why a certain matching decision is made, making it difficult for users to understand and trust the matching results, and also making it impossible to optimize matching rules accordingly.

[0019] To address the aforementioned issues, this application proposes a data matching method. It constructs a lightweight scientific knowledge graph, introduces an extended matching mechanism including synonyms and hyponyms / hypernyms, and combines fuzzy matching with a standardized vocabulary to form a multi-layered intelligent matching strategy. Furthermore, it replaces the traditional Boolean matching method with a weighted comprehensive scoring mechanism, effectively improving the accuracy, recall, and overall reliability of the data matching results.

[0020] The following is a detailed description of this case, in conjunction with the relevant accompanying drawings in the instruction manual.

[0021] Please see Figure 1 This application provides a data matching method, which specifically includes the following steps: S10: In response to a matching task request, obtain the task type included in the matching task request, and the user's input data for at least one preset matching field corresponding to the task type.

[0022] It is understood that the executing entity of this invention can be a data matching device, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0023] In this step, when a user initiates a matching task in the system, the system responds to the matching task request by obtaining the task type carried in the request and the user's input data for at least one preset matching field provided for that task type. The task type identifies the specific business scenario of the current matching task, such as achievement registration or scholar matching. The preset matching fields are the data fields that need to be matched and compared under each task type. For example, in a paper-scholar matching task, the fields to be matched include research direction and publication journal. The input data is the data to be matched entered by the user in the corresponding field, such as the paper title, research direction, or journal name.

[0024] S20: Based on the task type, retrieve candidate data for multiple candidate options from a preset database.

[0025] In this step, the preset database is a central scientific research database that has undergone one-time data cleaning and normalization, storing structured information on scholars, research achievements, and journals. Based on the task type, multiple candidate options are queried in the preset database, and candidate data for each option is obtained. The candidate data includes the field values ​​of each candidate option in each preset matching field, which are used for subsequent matching and comparison with the input data.

[0026] For example, in the paper-scholar matching task, the profile data of all scholars are obtained from a preset database as a candidate data set.

[0027] S30: Calculate the matching degree between the candidate data and the target data of each candidate option in each preset matching field using multiple preset matching methods.

[0028] Among them, the preset matching methods include at least one of the following: exact matching, fuzzy matching, and knowledge graph extended matching.

[0029] In this step, multiple matching methods are pre-defined based on the properties of different fields. This allows for flexible selection of the appropriate matching method during the matching process, based on the properties of each pre-defined matching field, to calculate the matching degree between the candidate data and the target data for each pre-defined matching field. For example, in the achievement registration task, the ISSN field can use exact matching to ensure the uniqueness of the identifier, while the journal name field can use fuzzy matching to accommodate abbreviations, capitalization, and spelling differences. In the scholar matching task, the research direction field can use knowledge graph extended matching to identify semantically related concepts, while the institution field can use fuzzy matching to handle different expression forms. By configuring differentiated matching methods for different matching fields, high-precision matching can be ensured while maximizing recall, effectively overcoming the shortcomings of existing technologies where a single matching method cannot adapt to multiple types of fields. In particular, the application of knowledge graph extended matching method enables the system to have semantic understanding capabilities. Even if the target data and candidate data are completely different in wording, as long as they have synonyms, hierarchical or related relationships in the knowledge graph, the system can still identify their semantic association and calculate a high degree of matching. This fundamentally solves the technical problem that the traditional keyword precise matching method causes a large number of related results or scholars to be missed due to word differences, and significantly improves the accuracy and comprehensiveness of the matching results.

[0030] Optionally, the various preset matching methods include at least precise matching, fuzzy matching, and knowledge graph extended matching. Precise matching is used to determine whether two field values ​​are completely identical, and is suitable for fields with unique identifiers such as ISSN numbers. Fuzzy matching calculates text similarity based on edit distance or Jaccard similarity coefficient, and is suitable for handling spelling differences, abbreviation variations, and inconsistencies between Chinese and English expressions. Knowledge graph extended matching obtains a set of extended concepts that are semantically related to the input data by querying a pre-constructed knowledge graph of the research field, and then calculates the semantic similarity between the extended concept set and the candidate field values, making it suitable for fields requiring semantic understanding, such as research directions and disciplines.

[0031] In practical applications, a hybrid strategy combining top-down and bottom-up approaches is employed to construct a lightweight knowledge graph for scientific research. First, requirements analysis and ontology design are conducted to clarify the graph's application scenarios (e.g., scholar recommendations, attribution of results), and core concept types are defined accordingly. For example, subject-specific terms, journal names, and research methods are set as node types in the graph. Simultaneously, the types of relationships between nodes are pre-planned, including synonym relationships (e.g., neural networks and deep networks), hierarchical relationships (e.g., artificial intelligence and machine learning), and correlation relationships (e.g., convolutional neural networks and image recognition). The next stage is data acquisition and preprocessing. Raw materials for constructing the graph are obtained from diverse and heterogeneous scientific research data sources, including authoritative subject-specific thesaurus / terminology dictionaries, academic literature databases, and existing structured scientific research data. After data acquisition, text cleaning and noise reduction preprocessing operations are performed to improve the quality of subsequent information extraction. Then, nodes and relationships are automatically extracted from the preprocessed text data. Specifically, various scientific research concepts (e.g., subject-specific terms, journal names, etc.) that conform to the ontology definition are extracted from the text. Simultaneously, synonyms, hyponyms, and related relationships exist between concepts. To reduce construction costs and achieve lightweight design, information extraction methods based on lightweight large language models (SLM / LLM) can be used, or existing structured knowledge bases can be leveraged for keyword extraction and graph structure construction. The extracted raw results also need to undergo entity disambiguation and conflict resolution to address issues such as homonyms or synonyms with different names. Afterward, the verified node and relationship data are imported into a graph database (such as Neo4j) for storage, forming a queryable knowledge graph. Furthermore, to ensure the long-term availability of the graph, an incremental update and maintenance mechanism can be established to support dynamic supplementation and correction of the graph as new data is introduced.

[0032] In one embodiment of this application, a specific field matching degree calculation scheme is provided. In S30, based on multiple preset matching methods, the matching degree between the candidate data and the target data of each candidate option on each preset matching field is calculated, specifically including the following steps S31-S34: S31: For any candidate, obtain the first field value of the input data in each preset matching field, and the second field value of the candidate data in each preset matching field.

[0033] In this step, for each candidate option, the specific content of the input data in each preset matching field is obtained and recorded as the first field value. Simultaneously, the specific content of the candidate data for that option in each preset matching field is obtained and recorded as the second field value. By obtaining the first and second field values ​​for each preset matching field, a data foundation is provided for subsequent field matching degree calculations.

[0034] For example, in the scholar matching task, the input data is the paper information entered by the user, and the first field value is the research direction keywords of the paper; the candidate data is the scholar's profile data, and the second field value is the scholar's research direction keywords.

[0035] S32: Get the field type of the first field value of each preset matching field.

[0036] The field types include at least one of the following: identifier fields and text fields.

[0037] In this step, the first field values ​​of the preset matching fields are identified and classified to determine the field type of each first field value. Field types include, but are not limited to, identifier fields and text fields. Identifier fields are data fields with uniqueness and standardization; their values ​​are typically standardized codes or numbers, such as ISSN numbers or scholar IDs. These fields require high matching accuracy and generally do not allow for deviations. Text fields are data fields composed of natural language; their values ​​are typically descriptive text content, such as research directions, journal names, and paper titles. These fields often have multiple expressions, spelling differences, or abbreviation variations in reality. Identifying the field type of the first field values ​​provides a basis for selecting the appropriate matching method for different fields.

[0038] S33: For any preset matching field, if the value of the first field is an identifier field, calculate the matching degree between the value of the first field and the value of the second field using the exact matching method.

[0039] In this step, the values ​​of identifier fields are uniquely linked throughout the entire database. Any slight difference (including character additions or deletions, order changes, or format variations) means pointing to completely different entity objects. Therefore, for identifier fields, an exact match method must be used to ensure the absolute accuracy of the matching results, allowing no deviation or ambiguity. For any preset matching field, when the first field value is determined to be an identifier field, the first field value and the second field value are compared character by character using an exact match method to determine if the two field values ​​are completely identical. If they are completely identical, the matching degree of the field is determined to be 100%; if they are not completely identical, the matching degree of the field is determined to be 0%.

[0040] For example, when matching ISSN numbers in the results registration task, if the ISSN number entered by the user does not match the ISSN number of a journal in the database by even one digit, it is determined that the two do not match rather than being approximately matched. Otherwise, it will lead to incorrect journal association and misjudgment of inclusion type. By using an exact matching method for identifier fields, the system can ensure absolute accuracy in matching unique fields while avoiding the risk of mismatches introduced by fuzzy matching methods, effectively guaranteeing the system's matching reliability on key fields.

[0041] S34: When the value of the first field is a text-based field, calculate the matching degree between the value of the first field and the value of the second field using fuzzy matching and knowledge graph extended matching methods.

[0042] In this step, when the first field value is identified as a text-based field, the matching degree between the first field value and the second field value is calculated based on at least one of fuzzy matching and knowledge graph extended matching methods. This effectively identifies semantically related concepts that differ literally while maintaining matching accuracy, significantly improving the recall and accuracy of the matching results.

[0043] In one embodiment of this application, a specific scheme for calculating field matching degree using fuzzy matching and knowledge graph extended matching is provided. In S34, that is, when the first field value is a text field, the matching degree between the first field value and the second field value is calculated based on fuzzy matching and knowledge graph extended matching, specifically including the following steps S341-S343: S341: Determine whether there are common words between the values ​​of the first field and the second field.

[0044] In this step, firstly, the first and second field values ​​are segmented into word units, dividing the continuous text string into several independent word units. Then, the segmented results are iterated and compared one by one to determine if there is at least one shared word between them. When there is a shared word between the first and second field values, it indicates a certain degree of direct overlap in their literal meaning; when there is no shared word, it indicates that no effective association can be established at the literal level.

[0045] For example, taking the research direction field as an example, the first field value is "neural network" and the second field value is "deep neural network". The first field value "neural network" is segmented into two lexical units: "neural" and "network". The second field value "deep neural network" is segmented into three lexical units: "depth", "neural", and "network". Then, the segmentation results for each field value are iterated and compared one by one to determine if there is at least one shared word. Since both "neural" and "network" appear in the segmentation results of "deep neural network", it is determined that there is a shared word between the first and second field values, namely "neural" and "network". If the first field value is "neural network" and the second field value is "deep learning", after segmentation, the former is "neural" and "network", and the latter is "depth" and "learning", and there are no shared lexical units between them, then it is determined that there is no shared word.

[0046] By using the above method, the most suitable matching method for the current field can be quickly determined before the matching calculation, which avoids unnecessary knowledge graph query overhead and ensures that effective matching can still be achieved through semantic association even when there is no overlap in the literal meaning.

[0047] S342: When there are common words between the first field value and the second field value, calculate the text similarity between the first field value and the second field value by fuzzy matching, and use the text similarity as the matching degree.

[0048] In this step, if there are common words between the first and second field values, it indicates a certain degree of direct overlap in their literal expressions. In this case, fuzzy matching can effectively calculate the text-level similarity. Specifically, the text similarity between the first and second field values ​​is calculated based on edit distance or the Jaccard similarity coefficient, and the calculated text similarity is used as the matching degree for that field. Since the two already share common words, the literal overlap provides a basic basis for similarity. At this point, fuzzy matching can accurately quantify the text-level similarity while ensuring computational efficiency. It does not require complex semantic reasoning using a knowledge graph, has low computational overhead, and reliable accuracy. It is suitable for handling spelling differences, abbreviation variations, word order adjustments, etc. It can achieve accurate text-level measurement while ensuring matching efficiency.

[0049] S343: In the absence of common words between the first field value and the second field value, the matching degree between the first field value and the second field value is calculated by the knowledge graph extended matching method.

[0050] In this step, if there are no common words between the first and second field values, it indicates that they are completely different in meaning and cannot be effectively compared at the text level. However, there may still be synonyms, hyponyms, or highly related semantic relationships between them, so semantic matching at the knowledge graph level is required. Specifically, the first field value is used as the query term to query a pre-built scientific research domain knowledge graph to obtain a set of extended concepts that are semantically related to the first field value. Then, the semantic similarity between each extended field value in this set of extended concepts and the second field value is calculated, and the calculated semantic similarity is used as the matching degree of that field. By introducing the knowledge graph extended matching method, semantic relationships between concepts can be discovered even when there is no overlap in meaning. This effectively identifies concepts such as neural networks and deep learning that are different in meaning but related in meaning, thereby avoiding missed judgments due to literal differences and significantly improving the recall rate of the matching results. It maximizes the coverage of semantic recognition while ensuring matching efficiency.

[0051] In one embodiment of this application, a specific scheme for calculating field matching degree using the knowledge graph extended matching method is provided. In S343, that is, when there are no common words between the first field value and the second field value, the matching degree between the first field value and the second field value is calculated by the knowledge graph extended matching method, specifically including the following steps S3431-S3434: S3431: Set the concept node corresponding to the first field value in the preset knowledge graph as the center node.

[0052] In this step, the pre-defined knowledge graph contains multiple concept nodes, each corresponding to a concept term in a research field, such as academic terminology, journal name, or research method. Using the first field value as the query condition, node matching is performed in the knowledge graph to find the concept node corresponding to the first field value. This node is then designated as the central node, serving as the starting point for subsequent graph traversal.

[0053] S3432: Obtain at least one extended field value at the associated concept node within a preset number of hops, centered on the central node, using a graph database traversal algorithm.

[0054] In this step, a graph database traversal algorithm is used to expand outwards layer by layer along the relational edges in the preset knowledge graph, starting from the central node, to obtain all associated concept nodes whose path length to the central node is within a preset number of hops. Each associated concept node stores the corresponding concept vocabulary, which is the extended field value.

[0055] Optionally, the preset number of hops can be configured according to actual matching needs. For example, setting the number of hops to 1 hop only retrieves concept nodes directly adjacent to the central node, while setting it to 2 hops can further retrieve indirectly related concept nodes. By limiting the range of preset hops, the breadth of knowledge graph expansion can be effectively controlled, avoiding the introduction of irrelevant noise concepts due to excessive expansion, thus improving recall while ensuring the accuracy of semantic association and computational efficiency.

[0056] S3433: Calculate the semantic similarity between each extended field value and the second field value respectively.

[0057] In this step, each extended field value and the second field value are converted into semantic vector representations in a high-dimensional space. Then, the cosine similarity between the semantic vectors of each extended field value and the semantic vectors of the second field value is calculated, and the value of the cosine similarity is used as the semantic similarity between the extended field value and the second field value.

[0058] S3434: The maximum value in the semantic similarity is determined as the matching degree between the first field value and the second field value.

[0059] In this step, when the first field value is expanded into multiple extended field values ​​through the knowledge graph, a semantic similarity is calculated between each extended field value and the second field value. The maximum value among all semantic similarities is selected as the final matching degree between the first and second field values. For example, if the first field value is "neural network", after knowledge graph expansion, it yields three extended field values: "deep learning", "CNN", and "machine learning". Semantic similarity is calculated between each of these extended field values ​​and the second field value "deep learning". The semantic similarity between "deep learning" and "deep learning" itself is 100%, the semantic similarity between "CNN" and "deep learning" is 85%, and the semantic similarity between "machine learning" and "deep learning" is 70%. The maximum value of 100% is selected as the matching degree for this field. By using the maximum value strategy, the path that best matches the second field value among multiple semantic association paths can be selected as the final result of the field matching degree. This fully utilizes the semantic expansion capability of the knowledge graph and effectively avoids the problem of underestimating the field matching degree due to the inaccuracy of a certain association path, making the matching result more reasonable and reliable.

[0060] In one embodiment of this application, a specific weight adjustment scheme is provided, which, after calculating the matching degree between the candidate data and the target data of each candidate item on each preset matching field based on multiple preset matching methods, further includes the following steps: Get the preset matching threshold for each preset matching field; If the matching degree of any preset matching field is less than the preset matching threshold, generate a matching error message based on the abnormal matching field and its corresponding matching degree. Send a matching error message to the user's terminal; Obtain the matching degree description of the abnormal matching field provided by the user in response to the matching error message; Based on the matching degree description, determine whether the preset weight value needs to be adjusted; If it is confirmed that the preset weight value needs to be adjusted, the weight adjustment parameter is determined based on the matching degree description, and the preset weight value of the abnormal preset matching field under the task type in the preset matching rule set is adjusted through the weight adjustment parameter.

[0061] In this embodiment, a preset matching threshold is obtained for each preset matching field. The preset matching threshold is a pre-configured matching degree threshold for each preset matching field, used to determine whether the matching result of that field meets the minimum acceptable standard. For example, in a scholar matching task, the preset matching threshold for the research direction field can be set to 60%, the preset matching threshold for the published journal field can be set to 50%, and the preset matching threshold for the affiliated institution field can be set to 40%. After calculating the field matching degree for each preset matching field, the field matching degree is compared with the preset matching threshold. When the matching degree of a preset matching field is greater than or equal to the preset matching threshold corresponding to that field, it indicates that the similarity of that field in this matching has reached the preset minimum acceptable standard. Conversely, when the matching degree is lower than the preset matching threshold, it indicates that the similarity of that field in this matching has not reached the preset minimum acceptable standard. There are generally three reasons why the matching degree might fall below the threshold: First, there is a significant difference between the user input data and the candidate data in that field. For example, if the input research direction is "neural network" while the candidate scholar's research direction is "Tang and Song literature," even knowledge graph extended matching cannot establish an effective semantic association. Second, the matching method is inappropriate. For example, fuzzy matching is used instead of knowledge graph extended matching to identify semantic associations, resulting in the inability to identify semantically related concepts that are different in wording. Third, the preset matching threshold for that field is set too high, exceeding the reasonable range achievable by actual matching capabilities, causing results that should be considered qualified matches to be misjudged as unqualified. When the matching degree of any preset matching field is detected to be lower than its preset matching threshold, a matching anomaly message is generated and sent to the user terminal based on the field name, actual matching degree, matching method, and preset matching threshold of the abnormal matching field. This alerts the user to the matching anomaly of that field and allows them to judge the actual importance of the field in the current task scenario based on their own business experience. If the user believes that the field is not actually important, it is recommended to reduce the preset weight of the field to reduce its negative impact on the overall matching score. If the user believes that the field is still important but the threshold setting is too strict, it is recommended to appropriately reduce the preset matching threshold of the field to relax the matching criteria. If the user believes that the matching method is inappropriate, they can switch to a more suitable matching method and recalculate the matching score. Users can generate matching score explanations based on the received prompts. For example, users can provide feedback on abnormal matching of the research direction field, such as "Although the scholar's research direction and the paper topic are different in wording, they are highly relevant in actual scientific research. It is recommended to increase the weight of this field to 70%" or "The low matching score of this field is due to the scholar's overly broad description and should not be used as a basis for rejection."

[0062] Furthermore, based on the matching degree description provided by the user, semantic analysis is performed on the matching degree description using natural language processing technology to extract the user's opinion on the current weight configuration, thereby obtaining the direction and magnitude of weight adjustment for that field, which serve as weight adjustment parameters. Subsequently, according to the weight adjustment parameters, the preset weight value of the abnormally matched field is increased, decreased, or left unchanged, and the adjusted weight value is updated to the corresponding position in the preset matching rule set for use in the next matching task. For example, when the matching degree of the research direction field is below the threshold and the user feedback indicates that the actual importance of this field is high and suggests adjusting it to 60%, the preset weight value of this field can be increased from 50% to 60%.

[0063] In practical applications, matching error messages can be pushed to the user's current terminal device via web front-end, mobile application, or email, and displayed on the user interface in the form of pop-ups, notification bars, or highlighted error fields, so that users can know the specific details of the matching error in a timely manner and respond accordingly.

[0064] By using the above methods, the weight configuration of each field can be optimized based on the actual matching situation and user suggestions, so that the matching rules gradually approach the optimal configuration in different business scenarios, effectively overcoming the shortcomings of existing technologies where the weight configuration is fixed and cannot be dynamically optimized based on the actual matching effect.

[0065] S40: Based on the task type and the set of preset matching rules, determine the target weight value of each preset matching field.

[0066] The preset matching rule set includes preset weight values ​​corresponding to each preset matching field under each task type.

[0067] In this step, traditional matching rules are typically executed sequentially in a fixed order, lacking differentiation in importance between matching fields. This means that a failure to match a minor field can negate the entire match, affecting the accuracy of the final matching result. To address this issue, this application introduces a task-type-based weight configuration mechanism. It pre-configures the weight values ​​of each preset matching field according to different task types, constructing a preset matching rule set. Based on the task type of the current matching task, it queries and loads the corresponding weight configuration from the preset matching rule set to obtain the target weight value for each preset matching field. This target weight value serves as the basis for subsequent weighted summation calculations, ensuring that subsequent matching calculations reflect the differences in the importance of each matching field under different business scenarios. This allows the quality of the matching result to comprehensively reflect the overall similarity of each field, ensuring that the matching result is reasonable and targeted across different scenarios.

[0068] Optionally, based on the diverse needs of different universities, disciplines, and business scenarios, a set of matching rules is pre-configured for each task type. Each matching rule includes multiple preset matching fields and their corresponding preset weight values. By establishing the correspondence between each preset matching field and preset weight value under each task type, a set of preset matching rules is formed. For example, in the scholar matching task, the multiple preset matching fields include at least: research direction, published journal, and affiliated institution, where the weight of research direction is configured as 50%, the weight of published journal is configured as 30%, the weight of affiliated institution is configured as 20%, and the sum of the weight values ​​of each field is 100%.

[0069] In practical applications, users can further optimize subsequent matching results by adjusting the weight configuration of matching rules or the settings of matching fields based on specific information in the matching details. This enhances the system's operability and continuous iterative optimization capabilities. A simulation testing interface is provided to receive test data samples uploaded by users and perform simulated matching based on the currently configured set of preset matching rules. The simulation results, including the number of matches, the matching rate, and an interpretability report, are then provided to the user for further rule configuration adjustments.

[0070] In one embodiment of this application, a specific target weight value determination scheme is provided. In S40, that is, based on the task type and the preset matching rule set, the target weight value of each preset matching field is determined, specifically including the following steps S41-S48: S41: Based on the task type, determine whether the matching task includes subject matching.

[0071] In this step, different task types rely on subject information to varying degrees. For example, scholar matching tasks typically require combining subject information to accurately locate scholars in the corresponding subject area, while achievement registration tasks are unrelated to subject. The task type identifier is used to determine whether a subject dimension needs to be introduced as a filtering condition for the matching rules.

[0072] S42: When the matching task includes subject matching, determine the subject to be matched based on the input data.

[0073] S43: In the preset matching rule set, obtain the matching rules corresponding to the task type and the subject to be matched.

[0074] For steps S42-S43, when the matching task includes subject matching, subject-related field values ​​are extracted from the input data. For example, in a scholar matching task, based on the paper title, keywords, or abstract entered by the user, the subject area to which the paper belongs is identified through methods such as text classification or keyword extraction, and this is used as the subject to be matched, to ensure that the selection of subsequent matching rules matches the subject background of the current paper.

[0075] Subsequently, the preset matching rule set stores the matching rule configurations corresponding to each discipline under each task type. Different disciplines can be configured with different matching fields and weight values ​​under the same task type. For example, in a scholar matching task, the matching rules for computer science may focus more on matching the research direction field, while the matching rules for humanities may focus more on matching the published works field. Using the task type and the discipline to be matched as joint query conditions, the preset matching rule set is located and the matching rules that match both are obtained.

[0076] S44: If the matching task does not include subject matching, retrieve the matching rule corresponding to the task type from the preset matching rule set.

[0077] In this step, when the matching task does not include subject matching, such as achievement registration and other task types that do not require subject differentiation, the task type is used as the query condition to obtain the corresponding matching rule from the preset matching rule set. This eliminates the need for subject screening and simplifies the matching process.

[0078] S45: Get the details of the matching rules.

[0079] The matching rule details include the preset weight values ​​of each preset matching field corresponding to the task type.

[0080] In this step, the matching rule details are read from the preset matching rule set to obtain the preset weight values ​​of each preset matching field.

[0081] S46: Provide the user with detailed matching rules.

[0082] In this step, the matching rule details are displayed to the user in a visual form through the user interface. For example, the field names of each preset matching field and their corresponding preset weight values ​​are presented in a list, so that the user can intuitively view the rule configuration of the current matching task and determine whether the weight values ​​need to be manually adjusted.

[0083] S47: In response to a user's instruction to modify any preset weight value, obtain the modified weight value included in the modification instruction, and modify the matching rules based on the modified weight value to obtain the target weight value of each preset matching field.

[0084] In this step, users can adjust the weight value of any preset matching field by entering a numerical value or dragging a slider on the interactive interface provided by the system. The system responds to the user's modification and extracts the modified weight value, replacing the original preset weight value to form the modified target weight value.

[0085] S48: In response to the user's confirmation instruction, use the preset weight values ​​of each preset matching field as the target weight values.

[0086] In this step, when the user reviews the matching rule details and deems no changes necessary, or has completed the modification and clicked confirm, the system receives the user's confirmation instruction and determines the weight values ​​of each preset matching field in the current matching rule as the target weight values ​​for subsequent matching and weighted calculation steps.

[0087] The above approach provides both preset rule configurations to improve efficiency and the ability for users to flexibly adjust weight configurations according to actual business needs, thus fully balancing automation and flexibility.

[0088] S50: The matching degree of each preset matching field and the target weight value are weighted and summed to generate the comprehensive matching degree of each candidate option.

[0089] In this step, for each candidate option, the matching degree of each preset matching field is multiplied by its corresponding target weight value to obtain a weighted score for each preset matching field. Then, the weighted scores of all preset matching fields are summed to obtain the overall matching degree of the candidate option, which measures the overall similarity between each candidate option and the user input data. For example, in the scholar matching task, if the matching degree of the research direction field is 90% and the weight is 50%, the matching degree of the published journal field is 80% and the weight is 30%, and the matching degree of the affiliated institution field is 60% and the weight is 20%, then the overall matching degree of the candidate option is 90%×50% + 80%×30% + 60%×20% = 45% + 24% + 12% = 81%. By weighting and summing the matching degree of each field with its corresponding weight, the matching result no longer relies on a binary conclusion based on a single field, but rather on a quantitative score that comprehensively considers the similarity of each field. This makes different candidates comparable and provides an objective and quantifiable basis for subsequent threshold screening and ranking.

[0090] In one embodiment of this application, a specific data matching result generation scheme is provided. In S50, the matching degree and weight value of each preset matching field are weighted and summed to generate the comprehensive matching degree of each candidate option. Specifically, it includes the following steps S51-S53: S51: Based on the preset output fields corresponding to the task type, extract the target field value from the candidate data of each target matching item.

[0091] In this step, different task types correspond to different preset output field configurations. For example, in the achievement registration task, the preset output fields are ISSN number and inclusion type. The corresponding ISSN number field value and inclusion type field value are extracted from the candidate data of the target matching item as the target field value. In the scholar matching task, the preset output fields are scholar name and affiliated institution. The corresponding scholar name field value and affiliated institution field value are extracted from the candidate data of the target matching item as the target field value.

[0092] By using the above method, it is possible to automatically determine which information needs to be extracted from the candidate data as the final matching answer based on the task type, ensuring the relevance and effectiveness of the output content.

[0093] S52: Based on the matching degree and matching method of each target matching item in each preset matching field, generate a matching description for each target matching item.

[0094] In this step, for each preset matching field of each target match, a corresponding natural language description is automatically generated based on the matching degree value of the field and the matching method used. For example, when the matching method is exact matching, a description of "field values ​​are completely identical" is generated; when the matching method is fuzzy matching, a description of "field values ​​achieve xx% similarity through fuzzy matching" is generated; and when the matching method is knowledge graph extended matching, a description of "neural networks and deep learning are semantically related concepts identified through knowledge graph" is generated. The matching descriptions of each field are integrated to form a complete matching description detail for the target match, which is used to explain to the user the specific basis for the scores of each field of the match.

[0095] S53: Based on the target field values ​​and matching descriptions of each target matching item, the preset template is called through the preset language model to generate data matching results.

[0096] In this step, pre-built templates are constructed for different task types. These templates define fixed sentence structures and information organization methods for the output results, such as "Based on the matching rules, we recommend [Scholar Name], with a comprehensive matching degree of [Comprehensive Matching Degree]%; where the research direction matching degree is [Matching Degree]%: [Matching Description]". The target field values ​​and matching descriptions of each target matching item are used as input parameters and filled according to the pre-built template structure to form structured data matching results.

[0097] By using the above methods, data matching results in natural language form are generated, so that the output content not only includes structured matching data, but also highly readable matching explanations, effectively improving the interpretability of the matching results and user experience.

[0098] S60: Based on the preset matching degree threshold corresponding to the task type and the comprehensive matching degree of each candidate option, determine at least one target matching item among multiple candidate options.

[0099] In this step, the preset matching threshold is a pre-configured screening standard for different task types, used to determine whether candidate items meet the matching requirements. Different task types can be configured with different thresholds; for example, the threshold can be set to 60% for scholar matching tasks and 90% for achievement registration tasks. The overall matching degree of each candidate item is compared with the preset matching threshold one by one. Candidate items with an overall matching degree that reaches or exceeds the threshold are identified as target matching items, and the target matching items are sorted in descending order of overall matching degree so that users can view the results with the highest matching degree first.

[0100] By using the above methods, candidates with low matching scores can be automatically filtered out, avoiding presenting a large number of irrelevant results to users and effectively improving the validity of the results.

[0101] Optionally, preset matching thresholds can be set in advance for each task type based on the matching accuracy requirements of the task type and the precision and recall requirements of the business scenario. For example, a higher threshold can be used when the accuracy requirements for achievement registration are high, while a lower threshold can be used when the candidate range needs to be expanded for scholar recommendations, making the screening mechanism sufficiently flexible.

[0102] S70: Generate data matching results based on the comprehensive matching degree of each target matching item, candidate data, and matching degree details of each preset matching field.

[0103] In this step, for each candidate identified as a target match, the target field value is extracted from the candidate data as the matching answer based on the preset output fields corresponding to the task type. This, combined with the overall matching degree of the target match and its detailed matching degree across each preset matching field, generates the data matching result for that target match. Subsequently, if there are multiple target matches, they are arranged in descending order of overall matching degree, and the arranged data matching results are output to the user interface for display.

[0104] In practical application scenarios, the results display interface uses progress bars or color blocks to visually present the overall matching score level, and displays the weighted score, field matching value and scoring reasons for each matching field in detail through the expanded panel. If the matching method includes knowledge graph extended matching, the scoring reasons include the semantic relationship of concepts identified based on the knowledge graph.

[0105] Through the above methods, users can not only know the final matching result, but also trace the specific calculation basis of each score. This effectively overcomes the shortcomings of existing technologies, such as opaque matching results, difficulty for users to understand and trust matching decisions, and significantly improves the interpretability of matching results and users' trust in the system.

[0106] The following section uses the entry of research results into a university research management system as an example to illustrate a data matching method proposed in this application, which includes the following steps: S1: Obtain the preset knowledge graph.

[0107] Specifically, the pre-defined knowledge graph contains journal nodes. Each journal node stores attribute information such as journal name, ISSN, and inclusion type, and establishes "synonym" relationships between the journal name and its various abbreviations and common spelling errors.

[0108] S2: Get the set of preset matching rules.

[0109] Specifically, the matching rules are detailed as follows: Rule 1, Journal Name Matching, with a weight of 60%, can use fuzzy matching and knowledge graph extended matching methods. It can not only match the full name of the journal, but also identify its abbreviation through synonym relationships; Rule 2, Keyword Matching, with a weight of 40%, can use fuzzy matching to calculate the text similarity between the paper's keywords and the journal's common subject terms.

[0110] S3: Smart Matching.

[0111] Specifically, the user's input data was obtained. The input paper information included the journal name "J. Mach. Learn. Res." and the keyword "support vector machine." Through knowledge graph analysis, it was discovered that "J. Mach. Learn. Res." is an abbreviation of "Journal of Machine Learning Research," and the journal's ISSN number (1532-4435) and indexing type (SCI) were successfully located. Simultaneously, a similarity score between the paper's keywords and the journal's subject terms was calculated using fuzzy matching. After weighted calculation, the journal's overall matching score was 95 points.

[0112] S4: Display of data matching results.

[0113] Specifically, the output recommendation result is "Journal of Machine Learning Research (ISSN: 1532-4435, SCI)" and the score details are displayed: "Journal name matching: 60 points (matching abbreviations through the synonym relationship of knowledge graph); Keyword matching: 35 points (based on fuzzy matching method)".

[0114] As can be seen, the above scheme utilizes three pre-defined matching methods—precise matching, fuzzy matching, and knowledge graph-extended matching—to calculate the matching degree between candidate data and user input data for each candidate option under each preset matching field. Then, a weighted sum is performed based on the preset weight values ​​corresponding to each preset matching field to obtain the comprehensive matching degree of each candidate option. Finally, at least one target matching item is selected from multiple candidate options using a preset matching degree threshold, generating data matching results containing detailed matching degree information. By introducing the knowledge graph-extended matching mechanism, the data matching process gains semantic understanding capabilities, enabling it to identify semantically related concepts that differ in literal meaning, effectively avoiding missed judgments due to literal differences, thus significantly improving the recall rate of the matching results. Simultaneously, the weighted summation mechanism replaces the traditional sequential rule chain execution method, overcoming the unreasonable problem of rejecting the entire result due to a single field matching failure, making the judgment of matching results more scientific and objective, and comprehensively improving the accuracy and rationality of the data matching results.

[0115] In one embodiment, a data matching device is provided, which corresponds one-to-one with the data matching method in the above embodiments. For example... Figure 2 As shown, the data matching device 100 includes: a first acquisition module 101, a second acquisition module 102, a calculation module 103, a first determination module 104, a first generation module 105, a second determination module 106, and a second generation module 107. Detailed descriptions of each functional module are as follows: The first acquisition module 101 is used to respond to a matching task request by acquiring the task type included in the matching task request and the user's input data for at least one preset matching field corresponding to the task type. The second acquisition module 102 is used to acquire candidate data of multiple candidate items from a preset database based on the task type; The calculation module 103 is used to calculate the matching degree between the candidate data of each candidate and the target data in each preset matching field through multiple preset matching methods. The multiple preset matching methods include at least one of the following: exact matching method, fuzzy matching method and knowledge graph extended matching method. The first determining module 104 is used to determine the target weight value of each preset matching field based on the task type and the preset matching rule set, wherein the preset matching rule set includes the preset weight value corresponding to each preset matching field under each task type. The first generation module 105 is used to perform a weighted summation of the matching degree of each preset matching field and the target weight value to generate the comprehensive matching degree of each candidate option. The second determining module 106 is used to determine at least one target matching item among multiple candidate items based on the preset matching degree threshold corresponding to the task type and the comprehensive matching degree of each candidate item; The second generation module 107 is used to generate data matching results based on the comprehensive matching degree of each target matching item, candidate data, and matching degree details of each preset matching field.

[0116] In one embodiment, the calculation module 103 is specifically used for: For any candidate, obtain the first field value of the input data in each preset matching field, and the second field value of the candidate data in each preset matching field; Obtain the field type of the first field value on each preset matching field, wherein the field type includes at least one of the following: an identifier field and a text field; For any preset matching field, if the value of the first field is an identifier field, the matching degree between the value of the first field and the value of the second field is calculated using the exact matching method; When the first field value is a text-based field, the matching degree between the first field value and the second field value is calculated using fuzzy matching and knowledge graph extended matching methods.

[0117] In one embodiment, the calculation module 103 is further configured to: Determine if there are common words between the values ​​of the first field and the second field; When there are common words between the first field value and the second field value, the text similarity between the first field value and the second field value is calculated by fuzzy matching, and the text similarity is used as the matching degree. If there are no common words between the first field value and the second field value, the matching degree between the first field value and the second field value is calculated by the knowledge graph extended matching method.

[0118] In one embodiment, the calculation module 103 is further configured to: Set the concept node corresponding to the first field value in the preset knowledge graph as the center node; By using a graph database traversal algorithm, at least one extended field value is obtained at the associated concept node within a preset number of hops, centered on the central node. Calculate the semantic similarity between the values ​​of each extended field and the value of the second field; The maximum value in the semantic similarity is determined as the matching degree between the first field value and the second field value.

[0119] In one embodiment, the device further includes: The third acquisition module is used to acquire the preset matching threshold of each preset matching field; The third generation module is used to generate a matching error message based on the abnormal matching field and its corresponding matching degree when the matching degree of any preset matching field is less than the preset matching threshold. The sending module is used to send matching error messages to the user terminal; The fourth acquisition module is used to acquire the matching degree description of the abnormal matching fields provided by the user in response to the matching error message; The judgment module is used to determine whether the preset weight value needs to be adjusted based on the matching degree description. The adjustment module is used to determine the weight adjustment parameters based on the matching degree description if it is confirmed that the preset weight value needs to be adjusted, and to adjust the preset weight value of the abnormal preset matching field under the task type in the preset matching rule set through the weight adjustment parameters.

[0120] In one embodiment, the first determining module 104 is specifically used for: Based on the task type, determine whether the matching task includes subject matching; When the matching task includes subject matching, the subject to be matched is determined based on the input data; From the preset set of matching rules, obtain the matching rules corresponding to the task type and the subject to be matched; If the matching task does not include subject matching, retrieve the matching rule corresponding to the task type from the preset matching rule set; Get the matching rule details of the matching rules, which include the preset weight values ​​of each preset matching field corresponding to the task type; The matching rules details will be provided to the user. In response to a user's instruction to modify any preset weight value, the modified weight value included in the modification instruction is obtained, and the matching rules are modified based on the modified weight value to obtain the target weight value of each preset matching field. In response to the user's confirmation instruction, the preset weight values ​​of each preset matching field are used as the target weight values.

[0121] In one embodiment, the second generation module 107 is specifically used for: Based on the preset output fields corresponding to the task type, extract the target field values ​​from the candidate data of each target matching item; Based on the matching degree and matching method of each target matching item in each preset matching field, a matching description for each target matching item is generated; Based on the target field values ​​and matching descriptions of each target matching item, a preset template is invoked through a preset language model to generate data matching results.

[0122] The data matching device 100 provided by this invention utilizes three pre-set matching methods—precise matching, fuzzy matching, and knowledge graph extended matching—to calculate the matching degree between candidate data of each candidate option and user input data under each preset matching field. Then, a weighted sum is performed based on the preset weight values ​​corresponding to each preset matching field to obtain the comprehensive matching degree of each candidate option. Finally, at least one target matching item is selected from multiple candidate options through a preset matching degree threshold, and a data matching result containing matching degree details is generated. By introducing a knowledge graph extended matching mechanism, the data matching process acquires semantic understanding capabilities, enabling it to identify semantically related concepts that differ in literal meaning, effectively avoiding missed judgments due to literal differences, thereby significantly improving the recall rate of the matching results. Simultaneously, the weighted summation mechanism replaces the traditional sequential rule chain execution method, overcoming the unreasonable problem of overall rejection due to a single field matching failure, making the judgment of matching results more scientific and objective, and comprehensively improving the accuracy and rationality of the data matching results.

[0123] For specific limitations regarding the data matching device, please refer to the limitations on the data matching method above, which will not be repeated here. Each module in the aforementioned data matching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware within or independently of the processor in the electronic device, or stored in software within the memory of the electronic device, so that the processor can call and execute the operations corresponding to each module.

[0124] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the data matching method described above.

[0125] In one embodiment, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the data matching method described above.

[0126] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0127] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0129] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A data matching method, characterized in that, include: In response to a matching task request, the task type included in the matching task request and the user's input data for at least one preset matching field corresponding to the task type are obtained. Based on the task type, candidate data for multiple candidate options are obtained from a preset database; The matching degree between the candidate data and the target data of each candidate item is calculated in each preset matching field using multiple preset matching methods. The multiple preset matching methods include at least one of the following: precise matching method, fuzzy matching method, and knowledge graph extended matching method. Based on the task type and the preset matching rule set, the target weight value of each preset matching field is determined, wherein the preset matching rule set includes the preset weight value corresponding to each preset matching field under each task type; The matching degree of each preset matching field and the target weight value are weighted and summed to generate the comprehensive matching degree of each candidate option; Based on the preset matching threshold corresponding to the task type and the comprehensive matching degree of each candidate option, at least one target matching item is determined from multiple candidate options; Based on the overall matching degree of each target matching item, candidate data, and matching degree details of each preset matching field, data matching results are generated.

2. The data matching method according to claim 1, characterized in that, The step of calculating the matching degree between candidate data and target data in each preset matching field using multiple preset matching methods specifically includes: For any candidate, obtain the first field value of the input data in each preset matching field, and the second field value of the candidate data in each preset matching field; Obtain the field type of the first field value on each preset matching field, wherein the field type includes at least one of the following: an identifier field and a text field; For any preset matching field, if the value of the first field is an identifier field, the matching degree between the value of the first field and the value of the second field is calculated by the exact matching method. When the value of the first field is a text-based field, the matching degree between the value of the first field and the value of the second field is calculated using fuzzy matching and knowledge graph extended matching methods.

3. The data matching method according to claim 2, characterized in that, The step of calculating the matching degree between the first field value and the second field value using fuzzy matching and knowledge graph extended matching methods specifically includes: Determine whether there are common words between the values ​​of the first field and the second field; When there are common words between the first field value and the second field value, the text similarity between the first field value and the second field value is calculated by fuzzy matching, and the text similarity is used as the matching degree. If there are no common words between the first field value and the second field value, the matching degree between the first field value and the second field value is calculated by the knowledge graph extended matching method.

4. The data matching method according to claim 3, characterized in that, The step of calculating the matching degree between the first field value and the second field value using the knowledge graph extended matching method specifically includes: Set the concept node corresponding to the first field value in the preset knowledge graph as the center node; By using a graph database traversal algorithm, at least one extended field value is obtained at the associated concept node within a preset number of hops, centered on the central node. Calculate the semantic similarity between the values ​​of each extended field and the value of the second field; The maximum value in the semantic similarity is determined as the matching degree between the first field value and the second field value.

5. The data matching method according to claim 1, characterized in that, After calculating the matching degree between the candidate data and the target data of each candidate option in each preset matching field using multiple preset matching methods, the method further includes: Get the preset matching threshold for each preset matching field; If the matching degree of any preset matching field is less than the preset matching threshold, a matching error message is generated based on the abnormal matching field and its corresponding matching degree. Send the matching error message to the user terminal; Obtain the matching degree description of the abnormal matching field provided by the user in response to the matching error message; Based on the matching degree description, determine whether the preset weight value needs to be adjusted; If it is confirmed that the preset weight value needs to be adjusted, the weight adjustment parameter is determined based on the matching degree description, and the preset weight value of the abnormal preset matching field under the task type in the preset matching rule set is adjusted through the weight adjustment parameter.

6. The data matching method according to claim 1, characterized in that, The step of determining the target weight value of each preset matching field based on the task type and the preset matching rule set specifically includes: Based on the task type, determine whether the matching task includes subject matching; If the matching task includes subject matching, the subject to be matched is determined based on the input data; From the preset matching rule set, obtain the matching rules corresponding to the task type and the subject to be matched; If the matching task does not include the subject matching, the matching rule corresponding to the task type is obtained from the preset matching rule set; Obtain the matching rule details of the matching rules, wherein the matching rule details include the preset weight values ​​of each preset matching field corresponding to the task type; The matching rules details will be provided to the user. In response to a user's instruction to modify any preset weight value, the modified weight value included in the modification instruction is obtained, and the matching rules are modified based on the modified weight value to obtain the target weight value of each preset matching field. In response to the user's confirmation instruction, the preset weight values ​​of each preset matching field are used as the target weight values.

7. The data matching method according to claim 1, characterized in that, The step of generating data matching results based on the comprehensive matching degree of each target matching item, candidate data, and matching degree details of each preset matching field specifically includes: Based on the preset output fields corresponding to the task type, extract the target field values ​​from the candidate data of each target matching item; Based on the matching degree and matching method of each target matching item in each preset matching field, a matching description for each target matching item is generated; Based on the target field values ​​and matching descriptions of each target matching item, the data matching results are generated by calling a preset template through a preset language model.

8. A data matching device, characterized in that, include: The first acquisition module is used to respond to a matching task request by acquiring the task type included in the matching task request, and the user's input data for at least one preset matching field corresponding to the task type. The second acquisition module is used to acquire candidate data of multiple candidate items from a preset database based on the task type; The calculation module is used to calculate the matching degree between the candidate data and the target data of each candidate item in each preset matching field through multiple preset matching methods, wherein the multiple preset matching methods include at least one of the following: precise matching method, fuzzy matching method and knowledge graph extended matching method; The first determining module is used to determine the target weight value of each preset matching field based on the task type and the preset matching rule set, wherein the preset matching rule set includes the preset weight value corresponding to each preset matching field under each task type; The first generation module is used to perform a weighted summation of the matching degree of each preset matching field and the target weight value to generate the comprehensive matching degree of each candidate option; The second determining module is used to determine at least one target matching item among multiple candidate items based on the preset matching degree threshold corresponding to the task type and the comprehensive matching degree of each candidate item; The second generation module is used to generate data matching results based on the comprehensive matching degree of each target matching item, candidate data, and matching degree details of each preset matching field.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the data matching method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the data matching method as described in any one of claims 1 to 7.

Images