Multi-source heterogeneous data fusion management method for scientific and technological innovation big data platform

Abstract

The application discloses a multi-source heterogeneous data fusion management method for a scientific and technological innovation big data platform, and particularly relates to the technical field of data management, obtains enterprise subject, intellectual property, scientific research project, policy realization and personnel service achievement data, and generates a scientific and technological innovation event unit; determines data lag characteristics according to event occurrence time and data warehousing time and generates a source offset sequence; identifies event breakpoints of subject number continuity but achievement attribution mutation or subject name continuity but subject number mutation; judges whether the event belongs to subject natural continuation, subject achievement undertaking or error connection based on a bearing counterargument chain; when the event is subject achievement undertaking, an isolated fusion node is generated, retrieval correlation display and scientific and technological portrait score separation calculation are realized; and the application can reduce the risk of subject mismerger and achievement misattribution caused by multi-source data update mispositioning.

Description

Technical Field

[0001] This invention relates to the field of data governance technology, specifically to a method for the fusion and governance of multi-source heterogeneous data for a science and technology innovation big data platform. Background Technology

[0002] In existing technologies, science and technology innovation big data platforms typically integrate and manage enterprise entity data, intellectual property data, scientific research project data, policy implementation data, and personnel achievement data through methods such as data collection, field mapping, name normalization, entity deduplication, and tag calculation, in order to form enterprise technology profiles, innovation capability scores, industrial chain maps, and policy matching results.

[0003] However, in the actual process of governing science and technology innovation data, the data update times of different data sources are not consistent. For example, business entity data may have been updated with information on company name changes or equity changes, while patent databases may still retain the original applicant's name, research project databases may still show the original undertaking unit, policy implementation databases may record data according to the company name at the time of implementation, and personnel employment data may lag behind changes in corporate organizational relationships. As a result, the same science and technology innovation entity may appear with the same name but different entity numbers in different data sources, or with consecutive entity numbers but abrupt changes in the attribution of achievements.

[0004] Existing merging methods mostly use static fields such as unified social credit code, enterprise name similarity, patent applicant name, and project undertaking unit name as the merging basis. When there are partial or highly similar fields among these, the system usually directly performs entity merging; when the fields are inconsistent, they are split into different entities. This method is difficult to identify event breakpoints caused by data entry delays, achievement transfers, enterprise name changes, project transfers, patent rights changes, or overlapping personnel positions.

[0005] Especially in scenarios such as evaluating technology companies in industrial parks, verifying eligibility for policy implementation, and recommending technology transfer achievements, directly merging the patents, projects, or policy records of historical entities into the current entity can lead to an inflated score for the company's innovation capability. Conversely, completely separating data with actual inheritance relationships can result in the omission of the company's genuine scientific and technological innovation resources. These errors typically do not manifest as single data anomalies but rather as misalignments in attribution across multiple data sources within different time windows, making them difficult to detect promptly through conventional deduplication, fuzzy matching, or manual rules.

[0006] Therefore, there is an urgent need for a multi-source heterogeneous data fusion and governance method for science and technology innovation big data platforms. This method can identify data lag characteristics caused by the inconsistency between the time of event occurrence and the time of data entry during the multi-source data fusion process, discover abnormal breakpoints between the continuity of the subject and the attribution of the results, and avoid direct merging or splitting before confirming the inheritance relationship, thereby improving the accuracy of the profile of science and technology innovation subjects, the judgment of the attribution of results, and the data governance results. Summary of the Invention

[0007] The purpose of this invention is to provide a method for the fusion and governance of multi-source heterogeneous data for a science and technology innovation big data platform, so as to solve the shortcomings of the background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform, comprising:

[0009] Acquire enterprise entity data, intellectual property data, scientific research project data, policy implementation data, and personnel appointment achievement data from the science and technology innovation big data platform, and extract the entity name, entity number, achievement attribution object, event occurrence time, data entry time, and data source type from each data to generate science and technology innovation event units;

[0010] Based on the difference between the event occurrence time and the data entry time, the data lag characteristics of each science and technology innovation event unit are determined, and a source offset sequence is generated based on the data lag characteristics.

[0011] Arrange science and technology innovation event units with the same or similar subject names according to the source offset sequence, and identify event breakpoints where the subject numbers are continuous but the result attribution object changes abruptly, or the subject names are continuous but the subject numbers change abruptly.

[0012] When the event breakpoint is identified, the entity merger is not performed directly. Instead, the equity relationship, patent right change, project undertaking relationship, and personnel appointment overlap relationship before and after the event breakpoint are extracted to generate a chain of evidence to support the continuation of the entity. Based on the chain of evidence to support the continuation of the entity, it is determined whether the data corresponding to the event breakpoint belongs to the natural continuation of the entity, the continuation of the entity's results, or the incorrect attachment.

[0013] When it is determined that the main achievement is inherited, an isolated fusion node is generated so that the data before and after the event breakpoint are displayed together during retrieval and calculated separately during the science and technology profile scoring.

[0014] When new data subsequently accesses the system to fill in the missing relationships in the supporting counter-evidence chain, the isolated fusion node is released, and the corresponding entity's attribution of results and data governance status are updated.

[0015] Preferably, the unit for generating scientific and technological innovation events includes:

[0016] Multiple subject names or multiple result attribution objects that exist simultaneously in the same original data are split into multiple intermediate records according to the relationship of one subject corresponding to one result attribution;

[0017] The subject name, subject number, and event occurrence time in the intermediate record are standardized, and a temporary subject number is generated when the subject number is missing;

[0018] The standardized subject name, subject number, result attribution object, event occurrence time, data entry time, and data source type are encapsulated into a science and technology innovation event unit.

[0019] Preferably, the data lag characteristics of each science and technology innovation event unit are determined, including:

[0020] Science and technology innovation event units with adjacent event occurrence times under the same data source type are grouped into a reference group, and the in-source lag baseline is determined based on the event occurrence time and data entry time of the science and technology innovation event units within the reference group.

[0021] The difference between the event occurrence time and the data entry time corresponding to each science and technology innovation event unit is subtracted from the source lag baseline of its data source type to obtain the data lag characteristics after source self-correction.

[0022] Preferably, generating the source offset sequence includes:

[0023] Science and technology innovation event units with the same or similar subject names are grouped into the same subject name set. Similar subject names are determined by the consistency of core font size and the consistency of at least two types of fields among administrative division, industry description and organizational form.

[0024] Within each set of subject names, the occurrence sequence and entry sequence of the science and technology innovation event unit are determined according to the event occurrence time and data entry time, and the offset status of first occurrence and then entry, later occurrence and first entry, or continuous following is marked accordingly.

[0025] Based on the offset state and the data lag characteristics, the science and technology innovation event units are rearranged to obtain the source offset sequence.

[0026] Preferably, identifying event breakpoints includes:

[0027] Within the same set of subject names, adjacent preceding and following event units are compared sequentially according to the source offset sequence;

[0028] When the preceding event unit and the following event unit satisfy the condition that the subject number is continuous and the result belongs to a sudden change, the position between them is marked as the first type of event breakpoint;

[0029] When the preceding event unit and the following event unit satisfy the condition that the subject name is consecutive and the subject number changes abruptly, the position between them is marked as a second type of event breakpoint.

[0030] Preferably, generating a chain of counter-evidence includes:

[0031] Using the preceding event unit and the following event unit corresponding to the event breakpoint as the data retrieval boundary, read data records that have a direct recording relationship with the preceding subject number, the following subject number, the preceding result belonging object, and the following result belonging object;

[0032] Extract equity relationships, patent right changes, project undertaking relationships, and personnel appointment overlap relationships from the data records, and record the occurrence time, start point, end point, and corresponding result attribution object of each relationship;

[0033] The extracted relationships are encapsulated according to relationship type, occurrence time, relationship start point, relationship end point, and corresponding result belonging object to obtain the chain of counter-evidence.

[0034] Preferably, the method for determining whether the data corresponding to the event breakpoint belongs to the natural continuation of the main body, the inheritance of the main body's results, or an erroneous attachment, based on the chain of counter-evidence, includes:

[0035] Using the preceding event unit and the following event unit as boundaries, three candidate explanations are constructed: natural continuation of the main body, inheritance of the main body's results, and incorrect attachment. The relationships in the chain of evidence of refusal are written into the three candidate explanations respectively.

[0036] Based on the occurrence time and the pointer of each relation, time gating is performed on the relations after they are written. Relations whose occurrence time is later than the time of change they are interpreted or whose pointers are separated from the subjects on both sides of the event breakpoint are marked as unusable relations.

[0037] In each candidate interpretation, counterfactual replays are performed on relations not marked as unusable, and the corresponding break in the chain of succession is recorded based on whether the changes in subject number, subject name continuation, and changes in the object to which the result belongs can be interpreted.

[0038] Preferably, the determination result includes:

[0039] The number of broken succession chains in the three candidate explanations—natural continuation of the main body, inheritance of the main body's achievements, and erroneous attachment—was counted separately.

[0040] The category with the fewest broken links and which did not use unusable relationships in the explanation is selected as the judgment result of the event breakpoint;

[0041] When the number of broken links in the two candidate interpretations is the same, the one that can explain the change in the object to which the result belongs is selected first. If they are still the same, the pending confirmation status of the event breakpoint is retained and the subject merging is not performed.

[0042] Preferably, generating isolated fusion nodes includes:

[0043] When the judgment result is that the main result is inherited, the inheritance path used to explain the change of the result's ownership object is read, and the inheritance relationship corresponding to the inheritance path is written into the isolated fusion node;

[0044] Record the preceding subject number, the following subject number, the object to which the preceding result belongs, the object to which the following result belongs, the event breakpoint location, the type of succession relationship, the time of occurrence of the succession relationship, and the isolation status in the isolated fusion node;

[0045] Historical results data associated with preceding event units are retained under the preceding entity name, and current results data associated with subsequent event units are retained under the subsequent entity name, while keeping the entity number of the original result belonging to the object unaltered.

[0046] Preferably, removing the isolated fusion node and updating the corresponding entity's attribution of results and data governance status includes:

[0047] Write a list of missing relationships into the isolated fusion node, and when accessing new data in the future, compare the newly added science and technology innovation event unit with the subject to be confirmed, the object to be confirmed, and the direction of the relationship to be confirmed in the list of missing relationships;

[0048] When the supplementary relationship in the newly added science and technology innovation event unit satisfies the corresponding relationship item in the missing relationship list, the relationship item is marked as complete, and the supplementary relationship is written into the receiving counter-evidence chain of the isolated fusion node;

[0049] When the completed relationship items form a complete acceptance path from the preceding subject to the following subject and covering the corresponding result belonging to the object, the isolation status of the isolated fusion node is updated to unisolated, and the corresponding result is marked as accepted and transferred out or accepted and transferred in.

[0050] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0051] This invention differs from existing technologies that rely on name similarity, unified social credit codes, or static field mapping to directly merge data. Its core lies in first converting multi-source science and technology innovation data into science and technology innovation event units with unified temporal meaning, and then forming data lag characteristics and source offset sequences by comparing event occurrence time with data entry time. This invention can make explicit the time misalignment caused by inconsistent update cycles of different data sources, and identify event breakpoints when the subject number is continuous but the attribution object of the achievement changes abruptly, or when the subject name is continuous but the subject number changes abruptly. This avoids the platform mistakenly merging abnormal data directly in scenarios such as enterprise name changes, achievement transfers, project relocations, or personnel mobility.

[0052] Furthermore, after identifying the event breakpoint, this invention does not immediately perform subject merging. Instead, it generates a chain of counter-evidence and uses time gating and counterfactual replay through three candidate interpretations: natural continuation of the subject, inheritance of subject achievements, and erroneous attachment. The interpretation with the fewest breaks in the chain is selected. This allows for the differentiation between "continuation of the same subject," "inheritance of achievements between different subjects," and "unfounded erroneous attachment." When an achievement inheritance is determined, the retrieval association display and profile scoring are separated and calculated by isolating and merging nodes. This preserves the inheritance relationship between data while preventing historical patents, projects, or policy records from being incorrectly included in the current subject's score, thus improving the accuracy of the science and technology innovation profile matching results. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0054] Figure 1 This is a flowchart of the multi-source heterogeneous data fusion and governance method for a science and technology innovation big data platform according to the present invention.

[0055] Figure 2 This is a flowchart of the method for determining whether the data corresponding to an event breakpoint belongs to the natural continuation of the main body, the inheritance of the main body's results, or the incorrect attachment, according to the present invention. Detailed Implementation

[0056] 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. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Example 1, please refer to Figure 1 and Figure 2 As shown in this embodiment, the multi-source heterogeneous data fusion and governance method for a science and technology innovation big data platform includes:

[0058] Acquire enterprise entity data, intellectual property data, scientific research project data, policy implementation data, and personnel appointment achievement data from the science and technology innovation big data platform, and extract the entity name, entity number, achievement attribution object, event occurrence time, data entry time, and data source type from each data to generate science and technology innovation event units.

[0059] In one possible implementation, the science and technology innovation big data platform is pre-configured with a data access directory, which includes enterprise entity data sources, intellectual property data sources, scientific research project data sources, policy implementation data sources, and personnel appointment achievement data sources. Among them, the enterprise entity data source can be a business registration database or a park enterprise database; the intellectual property data source can be a patent, trademark, or software copyright database; the scientific research project data source can be a science and technology plan project database; the policy implementation data source can be a policy application, public announcement, and disbursement database; and the personnel appointment achievement data source can be a talent pool, a paper database, an achievement database, or an expert resume database.

[0060] In this embodiment, the platform acquires the aforementioned data through interface calls, database synchronization, file import, or structured web page data collection, and first writes the acquired raw data into the raw data area. During writing, each raw data record includes the data source identifier, data batch number, and platform receiving time, with the platform receiving time serving as the data entry time for that data. For interface data, the data entry time is the time when the interface-returned data is successfully written by the platform; for batch files, the data entry time is the time when a single record is written to the database after file parsing; for manually entered data, the data entry time is the time when the entered record is submitted and approved.

[0061] After the original data is written, the platform uses field parsing rules to extract the subject name, subject number, result attribution object, event occurrence time, and data source type from different data types. For enterprise subject data, the subject name is taken from the current name of the enterprise or the name in the change record; the subject number is preferentially taken from the unified social credit code; the event occurrence time is taken from the establishment date, change approval date, or cancellation date; the result attribution object is recorded as the enterprise subject itself; and the data source type is marked as enterprise subject data. For intellectual property data, the subject name is taken from the name of the applicant, patentee, or assignee after the change of rights; the subject number is directly extracted if the original data contains the unified social credit code; otherwise, the record is empty and a matching identifier is generated; the result attribution object is taken from the name, publication number, or registration number of the corresponding patent, trademark, or software copyright; the event occurrence time is determined according to the application date, authorization date, registration date, or effective date of the change of rights; and the data source type is marked as intellectual property data.

[0062] For research project data, the subject name is the name of the project undertaking unit, lead unit, or participating unit; the subject number is the unit number or unified social credit code recorded in the project database; the attribute of the achievement is the project name and project number; the event occurrence time is the project initiation time, contract signing time, acceptance time, or project status change time; and the data source type is marked as research project data. For policy implementation data, the subject name is the name of the applicant unit, support recipient, or funding receiving unit; the subject number is the applicant subject number or unified social credit code; the attribute of the achievement is the policy item name, application batch, or implementation record number; the event occurrence time is the application time, public announcement time, approval time, or funding disbursement time; and the data source type is marked as policy implementation data. For personnel appointment achievement data, the subject name is the name of the personnel's employing unit, achievement completion unit, or authoring unit; the subject number is the employing unit number or the subject number matched by the platform's subject database; the attribute of the achievement is the personnel's name and the names of their papers, patents, projects, awards, or technological achievements; the event occurrence time is the start time of employment, achievement publication time, achievement registration time, or achievement completion time; and the data source type is marked as personnel appointment achievement data.

[0063] When multiple entity names or multiple attribution objects exist in the same original data, the platform splits it into multiple intermediate records according to the principle of "one entity corresponding to one attribution relationship". For example, when a research project includes both a lead unit and participating units, separate records are generated for the lead unit and participating units, retaining their respective unit roles in the records. When a patent has multiple applicants, corresponding records are generated for each applicant, retaining the joint application relationship. This avoids the erroneous compression of multi-entity joint achievements into a single entity's achievement during subsequent integration.

[0064] To ensure that data from different sources can enter the same processing flow, the platform further standardizes the format of extracted fields. Spaces, invalid administrative suffixes in parentheses, and differences between full-width and half-width characters are removed from the subject name; connectors and invisible characters are removed from the subject number; and event occurrence times are uniformly converted to year-month-day format. When the original data only contains year, month, or year, the time is marked as incomplete, while retaining the original time precision. For data from which a subject number cannot be parsed, it is not discarded directly, but rather the subject name, the object to which the result belongs, the event occurrence time, and the data source type are retained, and a temporary subject number is generated. This temporary subject number is used for subsequent supplementary matching with the enterprise subject database or other source data.

[0065] After the above processing, the platform encapsulates each intermediate record into a science and technology innovation event unit. The science and technology innovation event unit includes at least: subject name, subject number, object to which the achievement belongs, event occurrence time, data entry time, and data source type; it may further include original data identifier, event type, subject role, source batch number, original field summary, and time precision identifier. For example, a patent data entry can be encapsulated as follows: subject name is "Jia Intelligent Technology Co., Ltd.", subject number is its unified social credit code or temporary subject number, object to which the achievement belongs is the name and publication number of a certain invention patent, event occurrence time is the patent application date or rights change date, data entry time is the time the platform receives the patent record, and data source type is intellectual property data.

[0066] In this embodiment, through the above method, science and technology innovation data from different sources, with different structures and different field definitions are converted into science and technology innovation event units with unified field meanings. This enables subsequent steps to calculate data lag characteristics based on the difference between the event occurrence time and the data entry time, and to continue to generate source offset sequences and identify event breakpoints.

[0067] Based on the difference between the event occurrence time and the data entry time, the data lag characteristics of each science and technology innovation event unit are determined, and a source offset sequence is generated based on the data lag characteristics.

[0068] In one possible implementation, after generating the science and technology innovation event unit, the event occurrence time and data entry time in each unit are uniformly converted. Specifically, the event occurrence time is converted into an event occurrence day number, and the data entry time is converted into a data entry day number; the day number is the number of calendar days between the corresponding date and the start date of the same calendar. If the event occurrence time is only recorded to the month, the first day of that month is used as the event occurrence day number, and a "month precision" identifier is written into the science and technology innovation event unit; if the event occurrence time is only recorded to the year, the first day of that year is used as the event occurrence day number, and a "year precision" identifier is written; if the event occurrence time is recorded to the day, a "day precision" identifier is written. The above time precision identifiers are only used for further differentiation when sorting in parallel, and do not change the original recorded content of the event occurrence time.

[0069] For the i-th science and technology innovation event unit, calculate its unit lag day Di: Di ​​= Tiin - Tiout; where Di represents the unit lag day of the i-th science and technology innovation event unit, in days; Tiin represents the data entry date sequence number of the i-th science and technology innovation event unit; Tiout represents the event occurrence date sequence number of the i-th science and technology innovation event unit. Through this calculation, the time distance between the occurrence of each science and technology innovation event unit and its entry into the science and technology innovation big data platform can be obtained.

[0070] For science and technology innovation event units under the same data source type, they are arranged from earliest to latest according to the event occurrence time; when the event occurrence time is the same, they are arranged from earliest to latest according to the data entry time; when both the event occurrence time and the data entry time are the same, they are arranged according to the dictionary order after the subject name is normalized. After the arrangement is completed, two adjacent science and technology innovation event units are used as one reference group. For the j-th reference group, the reference lag value Rj is calculated: Rj=(Dj1+Dj2) / 2; where Rj represents the reference lag value of the j-th reference group; Dj1 represents the unit lag days of the science and technology innovation event unit ranked first in the j-th reference group; Dj2 represents the unit lag days of the science and technology innovation event unit ranked last in the j-th reference group. The reference lag value is generated by using two adjacent science and technology innovation event units to ensure that the entry follow-up relationship comes from continuous event changes within the same data source type, rather than from the update time difference between different data source types.

[0071] All reference lag values ​​obtained under the same data source type are arranged in ascending order of value, and the in-source lag baseline Bk for that data source type is calculated. When the number of reference lag values ​​n is odd: Bk = R((n+1) / 2); when the number of reference lag values ​​n is even: Bk = (R(n / 2) + R(n / 2+1)) / 2; where Bk represents the in-source lag baseline corresponding to the k-th data source type; n represents the number of reference lag values ​​under the k-th data source type; R((n+1) / 2) represents the reference lag value located at the (n+1) / 2 position after sorting; R(n / 2) and R(n / 2+1) represent the reference lag values ​​located at the n / 2 and n / 2+1 positions after sorting, respectively.

[0072] It should be noted that the above method of taking the median position is adopted because there may be supplementary entries, batch corrections, or delayed release records within the same data source type. Taking the median position can reduce the impact of a small number of extreme inbound records on the lagging baseline within the source.

[0073] After obtaining the source-adjusted data lag baseline, the source-adjusted data lag feature Li is calculated for each science and technology innovation event unit: Li = Di - Bk; where Li represents the data lag feature of the i-th science and technology innovation event unit; Di represents the unit lag days of the i-th science and technology innovation event unit; and Bk represents the source-adjusted data lag baseline corresponding to the data source type to which the science and technology innovation event unit belongs.

[0074] If Li is greater than 0, it indicates that the entry rhythm of this science and technology innovation event unit is lagging behind the usual data entry rhythm of the same data source type;

[0075] If Li is less than 0, it means that the entry rhythm of this science and technology innovation event unit is ahead of the usual entry rhythm of the same data source type;

[0076] If Li equals 0, it means that the science and technology innovation event unit is consistent with the usual data entry rhythm of the same data source type.

[0077] It is understandable that 0 here is the positive and negative boundary of the calculation result, and is not an artificially set threshold.

[0078] Subsequently, science and technology innovation event units with the same or similar subject names are grouped into the same subject name set. The so-called same subject name refers to the subject name text being completely identical after removing spaces, unifying full-width and half-width characters, unifying bracket format, and removing organizational format suffixes.

[0079] The similar subject names refer to those whose core font sizes are completely identical after the above standardization, and at least two types of fields in the administrative division field, industry description field, and organizational form field are identical;

[0080] The core font size refers to the main body name text remaining after deleting the administrative division field, industry description field, and organizational form field. This aggregation method uses field consistency relationships and does not use text similarity thresholds.

[0081] Within each set of subject names, the science and technology innovation event units are sorted by occurrence time to obtain occurrence sequence Pi, and by data entry time to obtain entry sequence Qi. When there are ties, they are first sorted by time precision, with daily precision earlier than monthly precision, and monthly precision earlier than yearly precision; if still ties, they are arranged in lexicographical order of data source type names. For the i-th science and technology innovation event unit, the sequence difference Ai is calculated: Ai = Qi - Pi; where Ai represents the sequence difference of the i-th science and technology innovation event unit; Qi represents its entry sequence; and Pi represents its occurrence sequence.

[0082] If Ai is greater than 0, then mark the science and technology innovation event unit as a first-occurrence-later-entry offset state;

[0083] If Ai is less than 0, then mark the science and technology innovation event unit as a "later-occurred, first-in" offset state;

[0084] If Ai equals 0, then the science and technology innovation event unit is marked as a continuous following offset state.

[0085] It is understandable that the 0 here also comes from the order difference calculation and is not a manually set threshold.

[0086] Finally, within each set of subject names, a source offset sequence is generated based on the offset state and data lag characteristics. Specifically, first, science and technology innovation event units with the "later-occurring, earlier-entered" offset state are selected and arranged in ascending order of data lag characteristic Li; then, science and technology innovation event units that continuously follow the offset state are selected and arranged in ascending order of event occurrence time; finally, science and technology innovation event units with the "first-occurring, later-entered" offset state are selected and arranged in descending order of data lag characteristic Li. When two science and technology innovation event units have the same data lag characteristic, they are arranged in ascending order of data entry time. After arrangement, the sequence positions are written sequentially to obtain the source offset sequence. Thus, subsequent steps can identify event breakpoints along the source offset sequence where subject numbers are continuous but the result attribution object changes abruptly, or where subject names are continuous but subject numbers change abruptly.

[0087] Scientific and technological innovation event units with the same or similar subject names are arranged according to the source offset sequence, and event breakpoints are identified where the subject number is continuous but the object to which the achievement belongs changes abruptly, or the subject name is continuous but the subject number changes abruptly.

[0088] In one possible implementation, after obtaining the source offset sequence, the science and technology innovation event units belonging to the same subject name set are arranged in ascending order according to their sequence position in the source offset sequence. The subject name set is the set obtained by aggregating the aforementioned identical or similar subject names. Each science and technology innovation event unit carries the subject name, subject number, achievement attribution object, event occurrence time, data entry time, data source type, data lag characteristics, and offset status.

[0089] After the sequence is arranged, each pair of adjacent science and technology innovation event units in the sequence is compared one by one. The science and technology innovation event unit that appears earlier in the sequence is designated as the preceding event unit, and the one that appears later is designated as the following event unit. During the comparison, the judgment does not cross the set of subject names to avoid including data from companies with unrelated names in the same breakpoint identification process.

[0090] First, the continuity of the subject numbers of the preceding and following event units is determined. If both the subject number of the preceding and following event units are not empty and are completely identical, then they are determined to satisfy the subject number continuity requirement. If one subject number is a temporary subject number and the other is a formal subject number, then the subject name and data source type retained when generating the temporary subject number are read. If the subject name corresponding to the temporary subject number belongs to the same subject name set as the subject name corresponding to the formal subject number, then they are determined to satisfy the subject number continuity requirement. If both are temporary subject numbers and their corresponding standardized subject names are completely identical, then they are determined to satisfy the subject number continuity requirement. The above determination is based only on the subject number and subject name set already written into the science and technology innovation event unit and does not introduce an additional manual similarity threshold.

[0091] A mutation determination is performed on the attribution of results for both preceding and subsequent event units. Specifically, the attribution of results is first broken down into the result object name, result object number, and result object type. Specifically, the result object number for intellectual property data is the publication number, authorization number, or registration number; the result object number for research project data is the project number; the result object number for policy implementation data is the implementation record number or application batch number; and the result object number for personnel appointment result data is the result registration number, paper identifier, or personnel result record number. If the result object numbers for both preceding and subsequent event units exist and are inconsistent, and the result object types are different, or if the result object numbers exist and are inconsistent but the corresponding subject role changes from one of the rights holder, undertaking unit, funding receiving unit, or employing unit to another, then a mutation in the attribution of results is determined. If the result object number is missing, the standardized text of the result object name and the result object type are used for joint determination; when the result object names and result object types are different, a mutation in the attribution of results is determined.

[0092] When adjacent preceding and following event units satisfy the conditions of continuous subject numbering and abrupt change in the attribution of the achievement, the position between them is marked as a first-type event breakpoint. This first-type event breakpoint indicates that, from the perspective of subject numbering, the data seems to belong to the same scientific and technological innovation subject; however, from the perspective of the attribution of the achievement, the achievement type, achievement number, or subject role has undergone discontinuous changes, and there is a risk that historical achievements may be directly attached to the current subject profile.

[0093] The system assesses cases where subject names are consecutive but subject numbers abruptly change. If the subject names of the preceding and following event units belong to the same subject name set and are adjacent in the source offset sequence, then they are considered to satisfy the subject name continuity condition. Furthermore, if the subject numbers of the preceding and following event units are both non-empty and inconsistent, then a subject number abrupt change is determined. If one is a formal subject number and the other a temporary subject number, and the temporary subject number appears for the first time in the current subject name set, a subject number abrupt change is also determined. If the two temporary subject numbers originate from different data source types and their corresponding result attribution objects are different, a subject number abrupt change is also determined.

[0094] When adjacent preceding and following event units satisfy the conditions of continuous subject names and abrupt change in subject number, the position between them is marked as a second type of event breakpoint. This second type of event breakpoint indicates that: from the perspective of subject name, the data may point to the same or closely related science and technology innovation subject; however, from the perspective of subject number, different identities have appeared in the platform, and there is a risk of companies with the same name, changes in subject, parent and subsidiary companies taking over, or incorrect mergers.

[0095] For the identified first-type and second-type event breakpoints, breakpoint records are written respectively. Each breakpoint record includes at least the preceding event unit identifier, the following event unit identifier, the breakpoint type, the preceding subject number, the following subject number, the object to which the preceding result belongs, the object to which the following result belongs, the preceding data source type, the following data source type, and the position of the breakpoint in the source offset sequence. If multiple event breakpoints exist within the same subject name set, they are arranged in ascending order of their position in the source offset sequence to form an event breakpoint list.

[0096] In this embodiment, through the above processing, event breakpoints are not directly generated based on single name similarity or single entity number difference, but rather generated within the adjacent event relationships defined by the source offset sequence. Therefore, subsequent steps can extract the equity relationships, patent right changes, project undertaking relationships, and personnel appointment overlaps before and after each event breakpoint, further generating a chain of counter-evidence and determining whether the corresponding data belongs to a natural continuation of the entity, inheritance of the entity's achievements, or an erroneous attachment.

[0097] When the event breakpoint is identified, the entity merger is not performed directly. Instead, the equity relationship, patent right change, project undertaking relationship, and personnel appointment overlap relationship before and after the event breakpoint are extracted to generate a chain of evidence to support the continuation of the entity. Based on the chain of evidence to support the continuation of the entity, it is determined whether the data corresponding to the event breakpoint belongs to the natural continuation of the entity, the continuation of the entity's results, or the incorrect attachment.

[0098] In one possible implementation, after obtaining the event breakpoint, the subject records on both sides of the breakpoint are not immediately merged. Instead, the preceding and subsequent event units in the event breakpoint record are used as data retrieval boundaries. The preceding event unit includes at least the preceding subject name, preceding subject number, the object to which the preceding result belongs, and the occurrence time of the preceding event. The subsequent event unit includes at least the following: the subsequent subject name, subsequent subject number, the object to which the subsequent result belongs, and the occurrence time of the subsequent event. During data retrieval, data records that have a direct relationship with the preceding subject number, subsequent subject number, preceding subject name, subsequent subject name, object to which the preceding result belongs, and object to which the subsequent result belongs are read, and equity relationships, patent right changes, project undertaking relationships, and personnel appointment overlap relationships are extracted from them.

[0099] For equity relationships, record the time of occurrence of the equity relationship, the direction of the equity relationship, the shareholding entity, and the shareholding entity; where the direction of the equity relationship is from the shareholding entity to the shareholding entity. For patent right changes, record the time of occurrence of the right change, the original right holder, the new right holder, and the corresponding patent result owner; where the direction of the patent right change is from the original right holder to the new right holder. For project undertaking relationships, record the time of occurrence of the undertaking relationship, the original undertaking entity, the new undertaking entity, and the corresponding project result owner; where the direction of the project undertaking relationship is from the original undertaking entity to the new undertaking entity. For personnel appointment overlap relationships, read the start and end times of the same person's appointment in the preceding and following entities. If the preceding and following appointment periods overlap in natural days, or the end date of the preceding appointment equals the start date of the following appointment, then a personnel appointment overlap relationship is generated from the preceding entity to the following entity; if the appointment time is only recorded to the month or year, then the day sequence number after the aforementioned time precision processing is used for judgment, and the original time precision identifier is retained. The above relationships are encapsulated according to relationship type, occurrence time, relationship start point, relationship end point, and corresponding result belonging object, forming a chain of rebuttal evidence.

[0100] After generating the chain of rebuttal evidence, three types of candidate interpretations are constructed using the preceding and following event units as boundaries. The first type is the candidate interpretation of natural continuation of the subject, meaning that the preceding and following subjects belong to the same scientific and technological innovation subject and are continuously expressed in different data source types. The second type is the candidate interpretation of subject achievement inheritance, meaning that the preceding and following subjects are not subjects pointed to by the same subject number, but the object to which the preceding achievement belongs can be transferred, inherited, or associated with the following subject through the chain of rebuttal evidence. The third type is the candidate interpretation of erroneous attachment, meaning that there is no valid relationship between the subjects on both sides of the event breakpoint that can explain the change in subject number and the change in the object to which the achievement belongs. During construction, each relationship in the chain of rebuttal evidence is copied and written into the three types of candidate interpretations, and the relationship type, occurrence time, relationship start point, relationship end point, and corresponding object to which the achievement belongs are retained in each type of candidate interpretation. This ensures that subsequent judgments are made within the same range of evidence, avoiding deviations in judgment results due to different evidence selections.

[0101] In this embodiment, time gating is required for the relationships within the three types of candidate interpretations. The time gating process uses the occurrence time of the explained change as the cutoff point. For relationships explaining changes in subject number, the cutoff point is the event occurrence time of the subsequent event unit; for relationships explaining changes in the result's attribution object, the cutoff point is the event occurrence time corresponding to the subsequent result's attribution object; when the subsequent result's attribution object does not have a separately recorded event occurrence time, the cutoff point is the event occurrence time of the subsequent event unit. If the occurrence time of a relationship is later than the cutoff point of the explained change, the relationship is marked as unusable. If neither the start nor end point of a relationship contains the preceding subject number, following subject number, preceding subject name, or following subject name, and the corresponding result's attribution object is not equal to the preceding or following result's attribution object, the relationship is marked as unusable. Records marked as unavailable relationships are still retained in the corresponding candidate interpretations, but cannot be used as the basis for changes in the subject number, the continuation of the subject name, or the change of the object to which the result belongs; if a candidate interpretation must rely on an unavailable relationship to form an interpretation, then the candidate interpretation will have one broken link in the chain.

[0102] After time-gating processing is completed, counterfactual replay is performed on each type of candidate interpretation. Counterfactual replay means assuming the candidate interpretation is true, and checking whether the relational interpretations of changes in subject number, continuation of subject name, and changes in the object to which the result belongs can be retained in sequence according to the event breakpoint from the preceding event unit to the following event unit.

[0103] For candidate interpretations of natural continuation of a subject name, if the preceding subject number and the following subject number are the same, then the change in subject number is interpreted; if they are different, but one of them is a temporary subject number, and the subject name corresponding to the temporary subject number and the other subject number belongs to the same set of subject names, then the change in subject number is interpreted; otherwise, one subject number break is recorded. If the preceding subject name and the following subject name belong to the same set of subject names, then the continuation of the subject name is interpreted; otherwise, one subject name break is recorded.

[0104] If the object to which the preceding result belongs is the same as the object to which the subsequent result belongs, or if they belong to the same result object number, then the change in the result object is explained; otherwise, one result ownership break is recorded.

[0105] For the candidate interpretation of the main achievement inheritance, first find the inheritance path from the previous entity to the subsequent entity from the retained relationships. The inheritance path is any of the following defined paths: the previous entity points to the subsequent entity through patent right change and the corresponding achievement ownership object is equal to the previous achievement ownership object or the subsequent achievement ownership object; the previous entity points to the subsequent entity through project undertaking relationship and the corresponding achievement ownership object is equal to the previous achievement ownership object or the subsequent achievement ownership object; there is an equity relationship between the previous entity and the subsequent entity, and within the same inheritance counter-evidence chain, there is also any of the following pointing to the subsequent entity: patent right change, project undertaking relationship, or personnel appointment overlap relationship; the same person has an overlapping personnel appointment relationship between the previous entity and the subsequent entity, and the achievement ownership object corresponding to this person is equal to the previous achievement ownership object or the subsequent achievement ownership object.

[0106] If any of the above paths exist, the change in the attribution of the result is explained;

[0107] If it does not exist, record 1 result continuity failure.

[0108] For candidate interpretations of the main results, if the preceding subject number and the following subject number are different, it is not recorded as a break, because this type of interpretation itself allows subject numbers to be discontinuous; however, if the preceding subject name and the following subject name do not belong to the same subject name set, and there is no inheritance path from the preceding subject to the following subject, then one subject association break is recorded.

[0109] For candidate interpretations of incorrect attribution, it is assumed that there is no valid connection between the subjects and results on both sides of the event breakpoint. If the preceding subject number is different from the following subject number, and there is no connection path that can explain the change in the result's ownership object, then both the change in subject number and the change in the result's ownership object meet the interpretation of incorrect attribution, and the breakpoint is not recorded.

[0110] If the retained relationship contains a path from the preceding subject to the following subject, and the corresponding result belongs to either the preceding or following result, then this path conflicts with the incorrect connection interpretation, and one incorrect interpretation break is recorded.

[0111] If the preceding entity number is the same and the preceding entity name and the following entity name belong to the same entity name set, then this state conflicts with the interpretation of incorrect attachment, and one identity conflict break is recorded.

[0112] Finally, the continuity chain breaks in the three types of candidate interpretations are compared. Let the number of breaks in the candidate interpretation of natural continuation of the main body be the number of natural breaks, the number of breaks in the candidate interpretation of inheritance of the main body's output be the number of inheritance breaks, and the number of breaks in the candidate interpretation of incorrect attachment be the number of erroneous breaks; each break number is the sum of the number of breaks recorded within the corresponding candidate interpretation. The type with the fewest breaks and which does not use unusable relationships in the interpretation is selected as the judgment result.

[0113] If the number of natural breaks is the minimum, and both the change in subject number and the continuation of subject name are explained, then the data corresponding to the event breakpoint is determined to be the natural continuation of the subject.

[0114] If the number of broken segments is minimized and the change in the object of achievement is explained by the achievement path, then the data corresponding to the event breakpoint is determined as the main achievement achievement.

[0115] If the number of erroneous breaks is the fewest, and the relationship between changes in the subject number and changes in the attribute of the result cannot be preserved, then the data corresponding to the event breakpoint is determined to be an erroneous connection.

[0116] If two candidate interpretations have the same number of breaks, the one that can explain the change in the attribution of the results shall be given priority.

[0117] If they are still the same, the pending confirmation status of the event breakpoint is retained, and the main body merging is not performed.

[0118] In this embodiment, the processing ensures that the judgment result is derived from the ability of the chain of counter-evidence to explain the changes on both sides of the event breakpoint, rather than simply merging based on the same name or different number.

[0119] When it is determined that the main achievement is inherited, an isolated fusion node is generated so that the data before and after the event breakpoint can be displayed together during retrieval and calculated separately during the science and technology profile scoring.

[0120] In one possible implementation, when the data corresponding to an event breakpoint is determined to be a main achievement successor, the platform does not directly merge the preceding entity corresponding to the preceding event unit and the subsequent entity corresponding to the subsequent event unit into the same entity. Instead, it generates an isolated fusion node between the two. The isolated fusion node is used to indicate that there is a succession relationship between the preceding entity and the subsequent entity at the level of achievement, project, rights, or personnel, but the two are still calculated separately as different entities in the science and technology profile scoring.

[0121] Specifically, the platform reads the acceptance paths already confirmed during the main achievement acceptance process and writes the corresponding relationships into the isolation and fusion node. The isolation and fusion node includes at least the isolation and fusion node number, the preceding entity number, the following entity number, the preceding entity name, the following entity name, the object to which the preceding achievement belongs, the object to which the following achievement belongs, the event breakpoint location, the type of acceptance relationship, the time when the acceptance relationship occurred, the source of the acceptance path, and the isolation status. The acceptance relationship type can be at least one of equity relationship, patent right change, project undertaking relationship, or overlapping personnel appointment relationship; the initial isolation status is written as "isolated and fused".

[0122] When generating isolated fusion nodes, the platform retains the preceding event units and their associated historical results data under the preceding entity's name, and the subsequent event units and their associated current results data under the subsequent entity's name, without changing the entity number of the original results' owner. If the preceding results' owner is transferred to the subsequent entity through patent rights changes or project undertaking relationships, only the transfer fact is recorded in the isolated fusion node, without directly rewriting the preceding results' owner as the subsequent entity's own results. This avoids the erroneous inclusion of historical patents, historical projects, or historical policy implementation records in the subsequent entity before the entity's results have been transferred and transformed into the entity's natural continuation.

[0123] For search and display, when the platform receives a query request using the name of the preceding entity, the name of the following entity, the number of the preceding entity, the number of the following entity, or the name of the inherited result as search criteria, it first returns the data of the entity being searched, and then reads the isolated and merged nodes related to that entity. If the searched entity is a preceding entity, the following entity and the object to which the following result belongs in the isolated and merged nodes are displayed as "inherited related data"; if the searched entity is a following entity, the preceding entity and the object to which the preceding result belongs in the isolated and merged nodes are displayed as "source related data". During the display, the platform simultaneously displays the type of inheritance relationship, the time when the inheritance relationship occurred, and the location of the event breakpoint, allowing users to see the data connection before and after the breakpoint, but this display result does not change the actual attribution field of any result object.

[0124] For technology profile scoring, when calculating indicators such as enterprise innovation capability, number of intellectual property rights, number of research projects, number of policy implementations, and number of talent achievements, the platform first reads the entity number to be scored, and then excludes achievement data that is only associated through isolated fusion nodes but does not belong to that entity number. In other words, achievement data of the preceding entity only enters the scoring scope of the preceding entity, and achievement data of the following entity only enters the scoring scope of the following entity; the data related to the following entity in the isolated fusion nodes only serves as profile description information, connection relationship prompts, or risk verification information, and is not used as the basis for quantitative scoring of the following entity. If the attribute of an achievement is already recorded as a following entity in the original data, or if subsequent data proves that the achievement has completed a valid transfer of rights, then the achievement can enter the scoring scope of the following entity based on the updated governance status.

[0125] For example, the preceding entity "Company A" owns a certain invention patent, and the subsequent entity "Company A Intelligent Technology Co., Ltd." obtains the patent's right change record after its establishment and uses the patent to apply for policy benefits. If the platform determines through the chain of evidence that the two entities have inherited the results, it generates an isolated fusion node, establishing an inheritance relationship between the patent, the right change record, and the policy benefit record. When a user searches for "Company A Intelligent Technology Co., Ltd.", they can see that the patent originates from the inheritance relationship of "Company A"; however, when calculating the number of authorized patents of "Company A Intelligent Technology Co., Ltd.", this patent is not directly included in its own authorized patent count until the subsequent attribution update conditions are met. This achieves the related display of data before and after the event breakpoint at the search level, as well as the separate calculation at the technology profile scoring level.

[0126] When new data subsequently accesses the system to fill in the missing relationships in the supporting counter-evidence chain, the isolated fusion node is released, and the corresponding entity's attribution of results and data governance status are updated.

[0127] In one possible implementation, after the data corresponding to the event breakpoint is determined to be the main result acceptance and an isolated fusion node is generated, a missing relationship list is simultaneously written into the isolated fusion node. The missing relationship list is used to record relationship items in the current acceptance counter-evidence chain that cannot yet support direct de-isolation. Each relationship item includes at least the missing relationship type, the subject to be confirmed, the object to be confirmed as the result, the direction of the relationship to be confirmed, the time of occurrence of the confirmation, and the reason for the missing relationship. The missing relationship type can be one or more of the following: equity relationship, patent right change, project undertaking relationship, or personnel appointment overlap relationship; the direction of the relationship to be confirmed is from the preceding subject to the subsequent subject, or from the subsequent subject to the preceding subject; the reason for the missing relationship is used to record situations in the acceptance counter-evidence chain such as missing right change time, missing project undertaking change subject, missing personnel appointment overlap time, or missing result object number.

[0128] When integrating new data, the process begins by extracting the subject name, subject number, result attribution object, event occurrence time, data entry time, and data source type from the new data, following the aforementioned method for generating science and technology innovation event units. This forms a new science and technology innovation event unit. Subsequently, the isolated and fused nodes in the "isolated and fused" state are read, and the subject number, subject name, and result attribution object in the new science and technology innovation event unit are matched with the preceding subject number, subsequent subject number, preceding subject name, subsequent subject name, preceding result attribution object, and subsequent result attribution object in the isolated and fused nodes. If the subject number in the new science and technology innovation event unit equals the preceding or subsequent subject number, or if the subject name in the new science and technology innovation event unit belongs to the subject name set corresponding to the isolated and fused node, or if the result attribution object in the new science and technology innovation event unit equals the preceding or subsequent result attribution object, then this new science and technology innovation event unit is treated as supplementary data to be verified.

[0129] For the supplementary data to be verified, further extract the supplementary relationships contained therein. If the supplementary data to be verified is intellectual property data, extract the patent right change relationship, which includes the original right holder, the new right holder, the time of the right change, and the corresponding patent achievement attribution object. If the supplementary data to be verified is scientific research project data, extract the project undertaking relationship, which includes the original undertaking entity, the new undertaking entity, the time of the undertaking relationship, and the corresponding project achievement attribution object. If the supplementary data to be verified is corporate entity data, extract the equity relationship, which includes the shareholding entity, the shareholding entity, the time of the equity relationship, and the relationship direction. If the supplementary data to be verified is personnel appointment achievement data, extract the personnel appointment overlap relationship, which includes the personnel identifier, the preceding appointment entity, the subsequent appointment entity, the start and end time of the appointment, and the corresponding achievement attribution object.

[0130] After obtaining the supplementary relationships, each supplementary relationship is compared with the list of missing relationships in the isolated fusion nodes. During the comparison, the supplementary relationship must simultaneously meet the following conditions: the relationship type of the supplementary relationship is consistent with the type of the missing relationship; the start and end points of the supplementary relationship match the direction of the relationship to be confirmed; the subject in the supplementary relationship contains at least one of the preceding and following subjects, and the other subject can be corresponding to a set of subject names or subject numbers; the result attribution object in the supplementary relationship is equal to the result attribution object of the preceding or following result, or has the same result object number as both; the occurrence time of the supplementary relationship is no later than the result attribution change time it explains. All of the above conditions are derived from the recorded fields in the isolated fusion nodes and the list of missing relationships, without using a manual similarity threshold.

[0131] When a supplementary relationship satisfies all the conditions of the corresponding missing relationship item, the missing relationship item is marked as supplemented, and the supplementary relationship is written into the continuation and counter-evidence chain of the isolated fusion node. If all relationship items in the missing relationship list of the isolated fusion node are marked as supplemented, or if the supplemented relationship items can form a complete continuation path from the preceding entity to the following entity and cover the corresponding result attribution object, then the isolation status of the isolated fusion node is updated from "isolated fusion" to "de-isolated". The complete continuation path refers to the existence of a time-gated equity relationship, patent right change, project undertaking relationship, or personnel appointment overlap relationship between the preceding entity and the following entity, and this relationship can explain the change of result attribution object recorded in the isolated fusion node.

[0132] When de-isolated and fused nodes, the isolated and fused nodes are not deleted. Instead, their node number, event breakpoint location, original supporting counter-evidence chain, supplementary relationship, and de-isolation time are retained and recorded as part of the data governance process. Subsequently, the attribution of the corresponding subject's results is updated based on the supplemented supporting counter-evidence chain. If the supplementary relationship is a patent right change relationship, the current attribution subject of the corresponding patent result is updated to the subject corresponding to the changed right holder; if the supplementary relationship is a project undertaking relationship, the current undertaking subject of the corresponding research project is updated to the undertaking subject after the change; if the supplementary relationship is a personnel appointment overlap relationship, only the personnel's results are treated as related results of the subsequent subject, without changing the original result completion subject; if the supplementary relationship is an equity relationship and simultaneously has a patent right change relationship or a project undertaking relationship, the attribution of results is determined according to the patent right change relationship or the project undertaking relationship.

[0133] After updating the attribution of results, the data governance status of the corresponding entities is further updated. For the preceding entity, results that have been accepted are marked as "accepted and transferred out"; for the following entity, results with the completed acceptance relationship are marked as "accepted and transferred in"; for the original isolated and merged nodes, the governance status is marked as "acceptance relationship completed". In the subsequent technology profile scoring, results that have been accepted and transferred in and have evidence of changes in rights or project undertakings can be included in the scoring scope of the following entity; results that are only related through overlapping personnel positions are still displayed as related data and are not included in the quantitative scoring scope of the following entity. Thus, it is possible to release isolated data relationships after the new data completes the acceptance counter-evidence chain, and to avoid premature changes in result attribution based solely on similar names or personnel associations.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for multi-source heterogeneous data fusion and governance in a science and technology innovation big data platform, characterized in that, include: Acquire enterprise entity data, intellectual property data, scientific research project data, policy implementation data, and personnel appointment achievement data from the science and technology innovation big data platform, and extract the entity name, entity number, achievement attribution object, event occurrence time, data entry time, and data source type from each data to generate science and technology innovation event units; Based on the difference between the event occurrence time and the data entry time, the data lag characteristics of each science and technology innovation event unit are determined, and a source offset sequence is generated based on the data lag characteristics. Arrange science and technology innovation event units with the same or similar subject names according to the source offset sequence, and identify event breakpoints where the subject numbers are continuous but the result attribution object changes abruptly, or the subject names are continuous but the subject numbers change abruptly. When the event breakpoint is identified, the entity merger is not performed directly. Instead, the equity relationship, patent right change, project undertaking relationship, and personnel appointment overlap relationship before and after the event breakpoint are extracted to generate a chain of evidence to support the continuation of the entity. Based on the chain of evidence to support the continuation of the entity, it is determined whether the data corresponding to the event breakpoint belongs to the natural continuation of the entity, the continuation of the entity's results, or the incorrect attachment. When it is determined that the main achievement is inherited, an isolated fusion node is generated so that the data before and after the event breakpoint are displayed together during retrieval and calculated separately during the science and technology profile scoring. When new data subsequently accesses the system to fill in the missing relationships in the supporting counter-evidence chain, the isolated fusion node is released, and the corresponding entity's attribution of results and data governance status are updated.

2. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 1, characterized in that, The generated science and technology innovation event unit includes: Multiple subject names or multiple result attribution objects that exist simultaneously in the same original data are split into multiple intermediate records according to the relationship of one subject corresponding to one result attribution; The subject name, subject number, and event occurrence time in the intermediate record are standardized, and a temporary subject number is generated when the subject number is missing; The standardized subject name, subject number, result attribution object, event occurrence time, data entry time, and data source type are encapsulated into a science and technology innovation event unit.

3. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 2, characterized in that, Determine the data lag characteristics of each science and technology innovation event unit, including: Science and technology innovation event units with adjacent event occurrence times under the same data source type are grouped into a reference group, and the in-source lag baseline is determined based on the event occurrence time and data entry time of the science and technology innovation event units within the reference group. The difference between the event occurrence time and the data entry time corresponding to each science and technology innovation event unit is subtracted from the source lag baseline of its data source type to obtain the data lag characteristics after source self-correction.

4. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 1, characterized in that, Generate the source offset sequence, including: Science and technology innovation event units with the same or similar subject names are grouped into the same subject name set. Similar subject names are determined by the consistency of core font size and the consistency of at least two types of fields among administrative division, industry description and organizational form. Within each set of subject names, the occurrence sequence and entry sequence of the science and technology innovation event unit are determined according to the event occurrence time and data entry time, and the offset status of first occurrence and then entry, later occurrence and first entry, or continuous following is marked accordingly. Based on the offset state and the data lag characteristics, the science and technology innovation event units are rearranged to obtain the source offset sequence.

5. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 4, characterized in that, Identify event breakpoints, including: Within the same set of subject names, adjacent preceding and following event units are compared sequentially according to the source offset sequence; When the preceding event unit and the following event unit satisfy the condition that the subject number is continuous and the result belongs to a sudden change, the position between them is marked as the first type of event breakpoint; When the preceding event unit and the following event unit satisfy the condition that the subject name is consecutive and the subject number changes abruptly, the position between them is marked as a second type of event breakpoint.

6. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 5, characterized in that, Generate a chain of counter-evidence, including: Using the preceding event unit and the following event unit corresponding to the event breakpoint as the data retrieval boundary, read data records that have a direct recording relationship with the preceding subject number, the following subject number, the preceding result belonging object, and the following result belonging object; Extract equity relationships, patent right changes, project undertaking relationships, and personnel appointment overlap relationships from the data records, and record the occurrence time, start point, end point, and corresponding result attribution object of each relationship; The extracted relationships are encapsulated according to relationship type, occurrence time, relationship start point, relationship end point, and corresponding result belonging object to obtain the chain of counter-evidence.

7. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 1, characterized in that, Based on the chain of counter-evidence, determine whether the data corresponding to the event breakpoint belongs to the natural continuation of the main body, the inheritance of the main body's results, or an erroneous attachment, including: Using the preceding event unit and the following event unit as boundaries, three candidate explanations are constructed: natural continuation of the main body, inheritance of the main body's results, and incorrect attachment. The relationships in the chain of evidence of refusal are written into the three candidate explanations respectively. Time-gated processing is applied to the written relations based on their occurrence time and relation pointers. Relationships whose occurrence time is later than their interpreted change time or whose relation pointers are separated from the subjects on both sides of the event breakpoint are marked as unusable relations. In each candidate interpretation, counterfactual replays are performed on relations not marked as unusable, and the corresponding break in the succession chain is recorded based on whether the changes in subject number, subject name continuation, and changes in the object to which the result belongs can be interpreted.

8. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 7, characterized in that, Determine the judgment result, including: The number of broken succession chains in the three candidate explanations—natural continuation of the main body, inheritance of the main body's achievements, and erroneous attachment—was counted separately. The category with the fewest broken links and which did not use unusable relationships in the explanation is selected as the judgment result of the event breakpoint; When the number of broken links in the two candidate interpretations is the same, the one that can explain the change in the object to which the result belongs is selected first. If they are still the same, the pending confirmation status of the event breakpoint is retained and the subject merging is not performed.

9. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 8, characterized in that, Generate isolated fusion nodes, including: When the judgment result is that the main result is inherited, the inheritance path used to explain the change of the result's ownership object is read, and the inheritance relationship corresponding to the inheritance path is written into the isolated fusion node; Record the preceding subject number, the following subject number, the object to which the preceding result belongs, the object to which the following result belongs, the event breakpoint location, the type of succession relationship, the time of occurrence of the succession relationship, and the isolation status in the isolated fusion node; Historical results data associated with preceding event units are retained under the preceding entity name, and current results data associated with subsequent event units are retained under the subsequent entity name, while ensuring that the entity number of the original result belonging to the object is not rewritten.

10. The method for multi-source heterogeneous data fusion and governance for a science and technology innovation big data platform according to claim 9, characterized in that, Remove the isolated fusion node and update the corresponding entity's attribution of results and data governance status, including: Write a list of missing relationships into the isolated fusion node, and when accessing new data in the future, compare the newly added science and technology innovation event unit with the subject to be confirmed, the object to be confirmed, and the direction of the relationship to be confirmed in the list of missing relationships; When the supplementary relationship in the newly added science and technology innovation event unit satisfies the corresponding relationship item in the missing relationship list, the relationship item is marked as complete, and the supplementary relationship is written into the receiving counter-evidence chain of the isolated fusion node; When the completed relationship items form a complete acceptance path from the preceding subject to the following subject and covering the corresponding result belonging to the object, the isolation status of the isolated fusion node is updated to unisolated, and the corresponding result is marked as accepted and transferred out or accepted and transferred in.

Images