Party building data acquisition and real-time analysis system

Abstract

The invention belongs to the technical field of data processing, and provides a party building data acquisition and real-time analysis system, which comprises the steps of cross-platform data acquisition through RESTfulAPI interface, OCR identification and the like, completion of field semantic alignment and feature extraction by using a lightweight 1D-CNN model, screening of unified and non-unified party building data, characterization of the data through Word2Vec and min-max normalization, and analysis of the data in real time. After a vector group is formed, a party building data association group is screened through weighted cosine similarity calculation, a unique association identity code is generated after cross-platform cross verification, and data deduplication and leak repairing are achieved. According to the method, cross-system data semantic consistency and accurate matching are guaranteed through multi-technology fusion, the accuracy and real-time performance of party building data fusion are improved, and reliable data support is provided for party building decision making.

Description

Technical Field

[0001] This invention belongs to the field of Y technology, specifically a system for collecting and analyzing Party building data in real time. Background Technology

[0002] With the advancement of digital transformation in Party building, Party building data is scattered across multiple independent systems, such as the Party member management information system, the "Study the Great Nation" platform, and the Party dues collection system. These systems are built by different departments and need to be integrated for real-time analysis to support accurate decision-making.

[0003] The core bottleneck in cross-system Party building data integration is the conflict of uniqueness of Party member identity identification: As the core identifier for data association, the coding rules for Party member identity are not uniform across different systems. For example, the Party member management system uses a nationally unified code, the Learning Power platform uses a mobile phone number + ID card number, and the grassroots self-built system uses a local number, which leads to frequent occurrences of one person having multiple codes and one code for multiple people, exacerbating the problem of duplication or omission when integrating Party building data.

[0004] The aforementioned problems severely reduce the accuracy and real-time performance of cross-system data fusion, leading to distorted analysis results such as Party member profiles and organizational health assessments, thus hindering the realization of data-driven decision-making. These issues have become core technical challenges that urgently need to be addressed in the field of Party building data fusion and real-time analysis.

[0005] Therefore, the present invention provides a system for collecting and analyzing Party building data in real time. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is:

[0008] One of the objectives of this invention is to provide a system for collecting and analyzing Party building data in real time, comprising:

[0009] Data collection and filtering module: Completes cross-platform Party building data collection, filters the collected Party building data, and identifies unified Party building data and non-unified Party building data;

[0010] Feature processing and matching module: It performs feature processing on unified Party building data and non-unified Party building data to form feature vectors of unified Party building data and non-unified Party building data, and forms feature vector groups. It calculates the overall cosine similarity of the feature vector groups through cosine similarity, and filters out the Party building data association groups based on the overall cosine similarity.

[0011] Verification and fusion optimization module: Performs cross-platform cross-verification of Party building data association groups, generates associated identity codes based on the verification results, and performs deduplication and omission filling of cross-platform integrated Party building data based on the associated identity codes.

[0012] As a further improvement of the present invention, the specific process of completing cross-platform Party building data collection is as follows:

[0013] For the standardized Party building system, a RESTful API interface combined with a data synchronization middleware is used to collect Party building data. The collected Party building data includes the Party member identity field corresponding to the Party building data, as well as the core time series field corresponding to the Party building data.

[0014] For non-standard systems built by grassroots organizations, a combination of structured data import and OCR text recognition is used to collect Party building data, supplement the core time-series fields corresponding to the Party building data, and ensure that the collected Party building data covers the corresponding Party member identity fields.

[0015] As a further improvement of the present invention, the specific process of identifying unified Party building data and non-unified Party building data is as follows:

[0016] For the same type of fields with heterogeneous representations in different systems, the semantic understanding capability of a lightweight 1D-CNN model is used to complete the semantic normalization mapping of the fields. The lightweight 1D-CNN model is used to extract features from the Party building data after the field semantic alignment process, including: extracting the Party member identity field corresponding to the Party building data and extracting the core time series field corresponding to the Party building data.

[0017] If the Party building data meets the requirement that the Party member identity field has a globally unique identifier that overlaps, and the matching degree of the core time series field business association feature corresponding to the Party building data is ≥95%, then it is determined to be unified Party building data.

[0018] If any of the following conditions are met: the Party building data has a missing Party member identity field and no globally unique identifier, or the matching degree of the core time-series field of the Party building data is less than 95%, then it is determined to be non-unified Party building data.

[0019] As a further improvement of the present invention, the core architecture of the lightweight 1D-CNN model is as follows:

[0020] The input layer receives text sequence encoding vectors for the fields of Party building data, covering the text / numerical encoding results of the Party member identity field and core time series fields;

[0021] The convolutional layer consists of three one-dimensional convolutional layers with 64, 32, and 16 kernels respectively. The kernel size is 3 for all layers, the stride is 1 for all layers, the padding method is the same, and the activation function is ReLU, which is used to extract local semantic features and sequence association features of the field.

[0022] A max pooling layer is set after every two convolutional layers, with a pooling kernel size of 2 and a stride of 2, to reduce dimensionality and retain key features.

[0023] The fully connected layer flattens the feature vector output from the pooling layer and then connects it to two fully connected layers with 64 and 32 neurons respectively, and the activation function is ReLU.

[0024] The output layer uses the Sigmoid activation function to output the semantic matching probability values ​​of the fields.

[0025] As a further improvement of the present invention, the specific process of forming a unified Party building data feature vector and a non-unified Party building data feature vector is as follows:

[0026] Obtain the Party member identity field and core time series field from both unified and non-unified Party building data. Perform feature processing on the Party member identity field and core time series field of both data to form feature vectors for unified and non-unified Party building data, respectively.

[0027] The unified Party building data feature vector includes: a unified feature vector for the Party member identity field and a unified feature vector for the core time series field; the non-unified Party building data feature vector includes: a non-unified feature vector for the Party member identity field and a non-unified feature vector for the core time series field.

[0028] As a further improvement of the present invention, the specific process of performing feature processing on the Party member identity field and the core time series field is as follows:

[0029] For the Party member identity field, the Word2Vec encoding algorithm is used to map each text field into a 128-dimensional text encoding vector;

[0030] For the core time series field, the time information is converted into timestamp values, and then processed by min-max normalization into a numerical vector in the range [0,1], and finally concatenated into a 128-dimensional time series encoding vector.

[0031] As a further improvement of the present invention, the specific process of forming the feature vector group is as follows:

[0032] The feature vectors of unified Party building data and non-unified Party building data are combined in pairs to form unique feature vector groups.

[0033] As a further improvement of the present invention, the specific process of obtaining the party building data association group based on the overall cosine similarity is as follows:

[0034] For any set of feature vectors, calculate the cosine similarity of the identity dimension between the feature vectors of the unified party member identity field and the feature vectors of the non-unified party member identity field, and the cosine similarity of the time dimension between the feature vectors of the non-unified core time-series field.

[0035] The overall cosine similarity is obtained by weighted fusion calculation of the cosine similarity of the identity dimension and the cosine similarity of the time sequence dimension.

[0036] If the overall cosine similarity of any feature vector group is greater than or equal to the overall cosine similarity threshold, then the party building data corresponding to the two sets of vectors are determined to have a valid association and are marked as a party building data association group.

[0037] If the overall cosine similarity is less than the overall cosine similarity threshold, then there is no valid association, and the feature vector group is directly removed and not included in subsequent processing.

[0038] 9. The Party building data collection and real-time analysis system according to claim 1, characterized in that: the specific process of cross-platform cross-validation of the Party building data association group is as follows:

[0039] Verify that the field dimensions and data formats of similar Party building data from different source platforms are consistent, and ensure that the data source attributes of similar Party building business data of the same Party member are consistent across platforms.

[0040] The core information in the Party member identity field is consistent and without conflict, and the business behavior time values ​​in the core time sequence field are consistent and the time dimension information is complete.

[0041] If any group of Party building data associations passes the cross-verification of data sources and the cross-verification of core fields, then the Party building data association group is deemed to have passed the verification.

[0042] If any verification item fails, the verification of the Party building data association group is deemed unsuccessful, and it is removed.

[0043] As a further improvement of the present invention, the specific process of deduplicating and filling gaps in cross-platform integrated Party building data based on associated identity codes is as follows:

[0044] To verify the unique identity code of the Party building data association group, the association identity code is bound one-to-one with the identity of the Party member. The association identity code integrates the globally unique identification information of the Party member, the overall cosine similarity value of the Party building data association group, and the coding of the Party building data source platform, and has uniqueness and traceability.

[0045] The associated identity code serves as a unified identity identifier for the Party member's cross-platform Party building data, and the associated Party building data includes all unified and non-unified Party building data within the associated Party building data group.

[0046] For Party building data under the same associated identity code, duplicate Party building data with identical Party member identity field, core time series field, and Party building business content are filtered out. Only one complete Party building data is retained. Party building data with complete Party member identity field but missing core business field, and complete core time series field but incomplete business content are identified. The missing fields are filled in based on the unified Party building data in the Party building data association group.

[0047] The second objective of this invention is to provide a method for collecting and analyzing Party building data in real time, including:

[0048] Step S10: Complete cross-platform Party building data collection, and filter the collected Party building data to identify unified Party building data and non-unified Party building data;

[0049] Step S20: Perform feature processing on unified Party building data and non-unified Party building data to form feature vectors for unified Party building data and non-unified Party building data, and form feature vector groups. Calculate the overall cosine similarity of the feature vector groups using cosine similarity, and filter out the Party building data association groups based on the overall cosine similarity.

[0050] Step S30: Perform cross-platform cross-validation on the Party building data association group, generate an association identity code based on the validation results, and perform deduplication and omission filling on the cross-platform integrated Party building data based on the association identity code.

[0051] The beneficial effects of this invention are as follows:

[0052] 1. To address the issue of multiple codes for one person and one code for multiple people caused by inconsistent coding rules for Party member identification across different systems, a lightweight 1D-CNN machine learning model is used for field semantic normalization mapping and two-dimensional feature matching. This achieves accurate association of heterogeneous data, completely resolving the problems of duplication and omission during Party building data fusion. Cross-platform cross-validation and association with identification codes ensure no conflicts in Party member identification information and no contradictions in time-series data. Furthermore, deduplication and omission correction operations form complete and non-redundant standardized Party building data, preventing distortion of analysis results such as Party member profiles and organizational health assessments.

[0053] 2. For standardized systems, RESTful APIs and data synchronization middleware are used to achieve real-time incremental data collection. For non-standard systems, structured import and OCR recognition are used to efficiently complete data. The entire technical process is streamlined and efficient, meeting the needs of real-time analysis and decision support for Party building data. It is compatible with multiple platforms, including Party member management systems and self-built non-standard systems at the grassroots level. It supports multi-dimensional data processing, including text and time-series data, and can cover core business data throughout the entire lifecycle of Party members. It also adapts to the flow and integration needs of exclusive identity data for mobile Party members, veteran Party members, etc. Through standardized integration and precise matching, it outputs high-quality cross-platform integrated Party building data, providing reliable data support for the digital transformation of Party building work and facilitating precise policy implementation and efficient management. Attached Figure Description

[0054] The invention will now be further described with reference to the accompanying drawings.

[0055] Figure 1 This is a system module diagram of a Party building data collection and real-time analysis system according to the present invention;

[0056] Figure 2 This is a flowchart of the steps of a method for collecting and analyzing Party building data in real time according to the present invention. Detailed Implementation

[0057] To make the technical means, creative features, objectives, and effects of this invention easily understood, the Party building data in this invention specifically refers to the core business data of Party members throughout their entire life cycle, which is linked across systems. This includes, but is not limited to: the unique national Party member code, ID number, name, political affiliation, affiliated Party organization, organizational relationship status and transfer records, participation records in organizational activities, learning points and learning completion status, Party dues payment base and payment records, key information of the entire process of developing Party members, and exclusive identity transfer information for mobile Party members / veteran Party members / college graduate Party members, hereinafter referred to as Party building data. The invention will be further described below with reference to specific implementation methods.

[0058] Example 1

[0059] like Figure 1 As shown in the embodiment of the present invention, a Party building data collection and real-time analysis system includes:

[0060] Data collection and filtering module: Completes cross-platform Party building data collection, filters the collected Party building data, and identifies unified Party building data and non-unified Party building data;

[0061] The specific process for completing cross-platform Party building data collection in the data collection and filtering module is as follows:

[0062] For standardized Party building systems such as the Party Member Management System, the "Study the Great Nation" platform, and the Party Dues Collection System, a RESTful API interface combined with data synchronization middleware is used to collect core business data related to the entire lifecycle of Party members across systems, namely Party building data. The collected Party building data includes the Party member identity fields corresponding to the Party building data (such as the unique national Party member code, ID number, name, mobile phone number, and affiliated Party organization), as well as the core time-series fields corresponding to the Party building data (the time of Party building data generation and the time of occurrence of the business behavior to which the Party building data belongs), realizing real-time incremental collection and scheduled full collection of Party building data.

[0063] For non-standard systems built by grassroots organizations, a combination of structured data import (Excel / CSV format) and OCR text recognition is used to collect Party building data; core time-series fields corresponding to the Party building data are supplemented simultaneously (such as supplementing the business occurrence time by manual filling), while ensuring that the collected Party building data covers the corresponding Party member identity fields;

[0064] In the data collection and filtering module, the specific process of filtering the collected Party building data and identifying unified Party building data and non-unified Party building data is as follows:

[0065] Design a lightweight 1D-CNN model to perform semantic alignment processing on the Party member identity field in the collected Party building data:

[0066] For the same type of field with heterogeneous expressions in different systems (such as "Party organization to which you belong" being expressed as "branch affiliation" in some systems), the semantic understanding capability of a lightweight 1D-CNN model is used to complete the semantic normalization mapping of the field, ensuring the semantic consistency of the Party member identity field and the core time series field.

[0067] A lightweight 1D-CNN model is used to extract features from the Party building data after the field semantic alignment process, including: extracting the Party member identity field corresponding to the Party building data (to ensure the uniqueness of the Party member identity) and extracting the core time series field corresponding to the Party building data;

[0068] The feature extraction results are categorized and filtered:

[0069] If the Party building data satisfies the condition that the Party member identity field has a globally unique identifier that overlaps, and the matching degree of the core time series field business association feature of the Party building data is ≥95%, then it is determined to be unified Party building data (i.e., Party building data that can be directly identified as the same Party member on different platforms).

[0070] If any of the following conditions are met: the Party building data has a missing Party member identity field and no globally unique identifier, or the matching degree of the core time series field business association feature of the Party building data is less than 95%, then it is determined to be non-unified Party building data (i.e., Party building data to be matched and associated later).

[0071] The core architecture of the lightweight 1D-CNN model is as follows:

[0072] The input layer receives text sequence encoding vectors (128-dimensional) of field Party building data, covering the text / numerical encoding results of the Party member identity field and core time series fields;

[0073] The convolutional layer consists of three one-dimensional convolutional layers with 64, 32, and 16 kernels respectively. The kernel size is 3 for all layers, the stride is 1 for all layers, the padding method is the same, and the activation function is ReLU, which is used to extract local semantic features and sequence association features of the field.

[0074] A max pooling layer is set after every two convolutional layers (two layers in total). The pooling kernel size is 2 and the stride is 2. This is used to reduce dimensionality and retain key features.

[0075] The fully connected layer flattens the feature vector output from the pooling layer and then connects it to two fully connected layers with 64 and 32 neurons respectively, and the activation function is ReLU.

[0076] The output layer uses the Sigmoid activation function to output the semantic matching probability value of the field (ranging from 0 to 1), which is used to determine the semantic consistency of heterogeneous fields.

[0077] Feature processing and matching module: It performs feature processing on unified Party building data and non-unified Party building data to form feature vectors of unified Party building data and non-unified Party building data, and forms feature vector groups. It calculates the overall cosine similarity of the feature vector groups through cosine similarity, and filters out the Party building data association groups based on the overall cosine similarity.

[0078] In the feature processing and matching module, the specific process of performing feature processing on unified Party building data and non-unified Party building data to form feature vectors for unified Party building data and non-unified Party building data is as follows:

[0079] Based on the unified and non-unified Party building data obtained through screening, the Party member identity field and core time series field of the unified and non-unified Party building data are obtained through the extraction of the lightweight 1D-CNN model. The Party member identity field and core time series field of the two are characterized respectively to form the feature vector of unified Party building data and the feature vector of non-unified Party building data.

[0080] The unified Party building data feature vector includes: a unified feature vector for the Party member identity field and a unified feature vector for the core time series field; the non-unified Party building data feature vector includes: a non-unified feature vector for the Party member identity field and a non-unified feature vector for the core time series field.

[0081] The specific process for characterizing the Party member identity field and the core time series field is as follows:

[0082] For the Party member identity field, the Word2Vec encoding algorithm is used to map each text field into a 128-dimensional text encoding vector;

[0083] For the core time series field, the time information is converted into timestamp values, and then normalized by min-max to become a numerical vector in the range [0,1], which is finally concatenated into a 128-dimensional time series encoding vector;

[0084] In the feature processing and matching module, the specific process of forming feature vector groups is as follows:

[0085] The feature vectors of unified Party building data and non-unified Party building data are combined in pairs to form non-repeating feature vector groups.

[0086] In the feature processing and matching module, the overall cosine similarity is obtained by calculating the feature vector group through cosine similarity, and the specific process of filtering the party building data association group based on the overall cosine similarity is as follows:

[0087] For any set of feature vectors, calculate the cosine similarity of the two core dimensions: first, the identity dimension cosine similarity between the unified party member identity field feature vector and the non-unified party member identity field feature vector; second, the time dimension cosine similarity between the unified core time series field feature vector and the non-unified core time series field feature vector.

[0088] Based on the matching characteristics of Party building data (with time-series business relevance as the core and identity uniqueness as the foundation), the cosine similarity of the two dimensions, identity dimension and time-series dimension, is weighted and fused to obtain the overall cosine similarity (example weight allocation: time-series dimension cosine similarity weight 70%, identity dimension cosine similarity weight 30%).

[0089] A pre-set overall cosine similarity threshold (e.g., 85%) is used. If the overall cosine similarity of any feature vector group is greater than or equal to the overall cosine similarity threshold, then the party building data corresponding to the two groups of vectors are determined to have a valid association and are marked as a party building data association group.

[0090] If the overall cosine similarity is less than the overall cosine similarity threshold, then it is determined that there is no effective association, and the feature vector group is directly removed and not included in subsequent processing.

[0091] Verification and fusion optimization module: Performs cross-platform cross-verification of the Party building data association group, generates an association identity code based on the verification results, and performs deduplication and omission filling on the cross-platform integrated Party building data based on the association identity code;

[0092] In the verification and fusion optimization module, the specific process of cross-platform verification of the Party building data association group is as follows:

[0093] For any unified and non-unified Party building data within a Party building data association group, verify that the field dimensions and data formats of the same type of Party building data under different source platforms remain consistent, and ensure that the data source attributes of the same type of Party building business data for the same Party member are consistent across platforms.

[0094] The core information of the Party member identity field in the Party building data association group is verified to be conflict-free and consistent. At the same time, the business behavior time values ​​of the core time series field are verified to be consistent and the time dimension information is complete. Party building data association groups with conflicting Party member identity information and conflicting business time values ​​are removed to ensure the authenticity and validity of the matched data.

[0095] If any group of Party building data associations passes the cross-verification of data sources and the cross-verification of core fields, then the Party building data association group is deemed to have passed the verification.

[0096] If any verification item fails, the verification of the Party building data association group is deemed unsuccessful and it is removed.

[0097] In the verification fusion optimization module, the specific process of generating the associated identity code based on the verification result is as follows:

[0098] Based on the verification of the party building data association group, a unique association identity code is generated for each party building data association group. The association identity code is bound one-to-one with the party member's identity. The association identity code integrates the party member's globally unique identifier information, the overall cosine similarity value of the party building data association group, and the party building data source platform code, and has uniqueness and traceability.

[0099] The associated identity code serves as a unified identity identifier for the Party member's cross-platform Party building data, and the associated Party building data includes all unified and non-unified Party building data within the associated Party building data group.

[0100] In the verification and fusion optimization module, the specific process of deduplicating and filling gaps in cross-platform integrated Party building data based on associated identity codes is as follows:

[0101] For Party building data under the same associated identity code, duplicate Party building data with identical Party member identity field, core time series field and Party building business content are filtered out. Only one complete Party building data is retained and the rest of the duplicate data is removed to achieve redundancy-free integration of Party building data.

[0102] For Party building data under the same associated identity code, we identify Party building data with complete Party member identity fields but missing core business fields, and complete core time series fields but incomplete business content. Based on the unified Party building data within the Party building data association group, we complete the missing fields and mark the source and basis of the completed data to ensure the integrity and continuity of Party building data.

[0103] Example 2

[0104] like Figure 2 As shown, based on the specific implementation process of Embodiment 1, the present invention provides a method for Party building data collection and real-time analysis, including:

[0105] Step S10: Complete cross-platform Party building data collection, and filter the collected Party building data to identify unified Party building data and non-unified Party building data;

[0106] Step S20: Perform feature processing on unified Party building data and non-unified Party building data to form feature vectors for unified Party building data and non-unified Party building data, and form feature vector groups. Calculate the overall cosine similarity of the feature vector groups using cosine similarity, and filter out the Party building data association groups based on the overall cosine similarity.

[0107] Step S30: Perform cross-platform cross-validation on the Party building data association group, generate an association identity code based on the validation results, and perform deduplication and omission filling on the cross-platform integrated Party building data based on the association identity code.

[0108] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A system for collecting and analyzing Party building data in real time, characterized in that: include: Data collection and filtering module: Completes cross-platform Party building data collection, filters the collected Party building data, and identifies unified Party building data and non-unified Party building data; Feature processing and matching module: It performs feature processing on unified party building data and non-unified party building data to form feature vectors of unified party building data and non-unified party building data, and forms feature vector groups. It calculates the overall cosine similarity of the feature vector groups through cosine similarity, and filters out party building data association groups based on the overall cosine similarity. Verification and fusion optimization module: Performs cross-platform cross-verification of Party building data association groups, generates associated identity codes based on the verification results, and performs deduplication and omission filling of cross-platform integrated Party building data based on the associated identity codes.

2. The Party building data collection and real-time analysis system according to claim 1, characterized in that: The specific process for completing cross-platform Party building data collection is as follows: For the standardized Party building system, a RESTful API interface combined with a data synchronization middleware is used to collect Party building data. The collected Party building data includes the Party member identity field and the core time series field corresponding to the Party building data. For non-standard systems built by grassroots organizations, a combination of structured data import and OCR text recognition is used to collect Party building data, supplement the core time-series fields corresponding to the Party building data, and ensure that the collected Party building data covers the corresponding Party member identity fields.

3. The Party building data collection and real-time analysis system according to claim 1, characterized in that: The specific process for identifying unified Party building data and non-unified Party building data is as follows: For the same type of fields with heterogeneous representations in different systems, the semantic understanding capability of the lightweight 1D-CNN model is used to complete the semantic normalization mapping of the fields. The lightweight 1D-CNN model is used to extract features from the party building data after the field semantic alignment process, including: extracting the party member identity field corresponding to the party building data and extracting the core time series field corresponding to the party building data. If the Party building data meets the requirement that the Party member identity field has a globally unique identifier that overlaps, and the matching degree of the core time series field business association feature corresponding to the Party building data is ≥95%, then it is determined to be unified Party building data. If any of the following conditions are met: the Party building data has a missing Party member identity field and no globally unique identifier that overlaps, or the matching degree of the core time-series field of the Party building data is less than 95%, then it is determined to be non-unified Party building data.

4. The Party building data acquisition and real-time analysis system according to claim 3, characterized in that: The core architecture of the lightweight 1D-CNN model is as follows: The input layer receives text sequence encoding vectors for the fields of Party building data, covering the text / numerical encoding results of the Party member identity field and core time series fields; The convolutional layer consists of three one-dimensional convolutional layers with 64, 32, and 16 kernels respectively. The kernel size is 3 for all layers, the stride is 1 for all layers, the padding method is the same, and the activation function is ReLU, which is used to extract local semantic features and sequence association features of the field. A max pooling layer is set after every two convolutional layers, with a pooling kernel size of 2 and a stride of 2, to reduce dimensionality and retain key features. The fully connected layer flattens the feature vector output from the pooling layer and then connects it to two fully connected layers with 64 and 32 neurons respectively, and the activation function is ReLU. The output layer uses the Sigmoid activation function to output the semantic matching probability values ​​of the fields.

5. The Party building data acquisition and real-time analysis system according to claim 1, characterized in that: The specific process for forming a unified feature vector and a non-unified feature vector for Party building data is as follows: Obtain the Party member identity field and core time series field from both unified and non-unified Party building data. Perform feature processing on the Party member identity field and core time series field of both data to form feature vectors for unified and non-unified Party building data, respectively. The unified Party building data feature vector includes: a unified feature vector for the Party member identity field and a unified feature vector for the core time series field; the non-unified Party building data feature vector includes: a non-unified feature vector for the Party member identity field and a non-unified feature vector for the core time series field.

6. The Party building data acquisition and real-time analysis system according to claim 5, characterized in that: The specific process for performing feature processing on the Party member identity field and the core time series field is as follows: For the Party member identity field, the Word2Vec encoding algorithm is used to map each text field into a 128-dimensional text encoding vector; For the core time series field, the time information is converted into timestamp values, and then processed by min-max normalization into a numerical vector in the range [0,1], and finally concatenated into a 128-dimensional time series encoding vector.

7. The Party building data collection and real-time analysis system according to claim 1, characterized in that: The specific process for forming the feature vector set is as follows: The feature vectors of unified Party building data and non-unified Party building data are combined in pairs to form unique feature vector groups.

8. The Party building data collection and real-time analysis system according to claim 1, characterized in that: The specific process for obtaining the association groups of Party building data based on the overall cosine similarity is as follows: For any set of feature vectors, calculate the cosine similarity of the identity dimension between the feature vectors of the unified party member identity field and the feature vectors of the non-unified party member identity field, and the cosine similarity of the time dimension between the feature vectors of the non-unified core time-series field. The overall cosine similarity is obtained by weighted fusion calculation of the cosine similarity of the identity dimension and the cosine similarity of the time sequence dimension. If the overall cosine similarity of any feature vector group is greater than or equal to the overall cosine similarity threshold, then the party building data corresponding to the two sets of vectors are determined to have a valid association and are marked as a party building data association group. If the overall cosine similarity is less than the overall cosine similarity threshold, then there is no valid association, and the feature vector group is directly removed and not included in subsequent processing.

9. The Party building data collection and real-time analysis system according to claim 1, characterized in that: The specific process for cross-platform cross-validation of the Party building data association group is as follows: Verify that the field dimensions and data formats of similar Party building data from different source platforms are consistent, and ensure that the data source attributes of similar Party building business data of the same Party member are consistent across platforms. The core information in the Party member identity field is consistent and without conflict, and the business behavior time values ​​in the core time sequence field are consistent and the time dimension information is complete. If any group of Party building data associations passes the cross-verification of data sources and the cross-verification of core fields, then the Party building data association group is deemed to have passed the verification. If any verification item fails, the verification of the Party building data association group is deemed unsuccessful, and it is removed.

10. A Party building data acquisition and real-time analysis system according to claim 1, characterized in that: The specific process of deduplicating and filling gaps in cross-platform integrated Party building data based on associated identity codes is as follows: To verify the unique identity code of the Party building data association group, the association identity code is bound one-to-one with the identity of the Party member. The association identity code integrates the globally unique identification information of the Party member, the overall cosine similarity value of the Party building data association group, and the coding of the Party building data source platform, and has uniqueness and traceability. The associated identity code serves as a unified identity identifier for the Party member's cross-platform Party building data, and the associated Party building data includes all unified and non-unified Party building data within the associated Party building data group. For Party building data under the same associated identity code, duplicate Party building data with identical Party member identity field, core time series field, and Party building business content are filtered out. Only one complete Party building data is retained. Party building data with complete Party member identity field but missing core business field, and complete core time series field but incomplete business content are identified. The missing fields are filled in based on the unified Party building data in the Party building data association group.

Images