Enterprise user portrait data processing method based on big data analysis

Abstract

The invention discloses an enterprise user portrait data processing method based on big data analysis, and the method comprises the steps: collecting information data of a target enterprise from a plurality of different data sources, carrying out the independent preprocessing of the information data collected by each data source, and generating a plurality of structured data sets with source identifiers; constructing a data credibility evaluation model, and obtaining a credibility score value of each data source; according to the credibility score value of each data source, missing indexes existing in the numerical structured data are calculated in an unused mode, and a first result or a second result is obtained; and forming an enterprise information data set based on the first result and the second result, and constructing an enterprise portrait model based on the enterprise information data set.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to an enterprise user portrait data processing method based on big data analysis. BACKGROUND

[0002] With the rapid development of big data technology, enterprise user portrait construction has become an important basic technology in the fields of enterprise credit investigation, risk assessment, and precision marketing. In the practice of enterprise big data application, data sources are increasingly diversified, including government open databases, commercial credit investigation platforms, enterprise websites, and various third-party data sources. This multi-source and heterogeneous characteristic provides rich information for enterprise portraits, but also brings major challenges in data quality control.

[0003] In the prior art, the enterprise data processing method usually adopts a processing mode of "fusion first, cleaning later", that is, first, the enterprise information from different data sources is integrated to form a unified data set, and then the missing value processing and abnormal value correction are performed. Some existing technologies use mean filling or median filling to process missing values after aggregating multi-source data. Some existing technologies use a single K-nearest neighbor algorithm to uniformly fill all missing fields after data fusion. These methods ignore the credibility differences of different data sources, resulting in the quality advantage of high-credibility data sources being diluted by the noise of low-credibility data sources, ultimately affecting the accuracy and reliability of enterprise portraits. SUMMARY

[0004] In view of the above prior art, the present application provides an enterprise user portrait data processing method based on big data analysis, which mainly solves the technical problems in the background art.

[0005] To achieve the above purpose, the technical solution of the embodiment of the present application is as follows: an enterprise user portrait data processing method based on big data analysis, the processing method comprising: Collecting information data of a target enterprise from multiple different data sources, independently preprocessing the information data collected from each data source, and generating multiple structured data sets with source identifiers; Constructing a data credibility evaluation model to obtain the credibility score value of each data source; According to the credibility score value of each data source, calculating the missing indicators in the numerical structured data in different ways to obtain a first result or a second result; Forming an enterprise information data set based on the first result and the second result, and constructing an enterprise portrait model based on the enterprise information data set.

[0006] Optionally, the information data collected from each data source is independently preprocessed to generate multiple structured data sets with source identifiers, specifically including: assigning a unique source identifier to each data source, the source identifier comprising a data source category code, a data source name code, and a data collection timestamp; independently performing a data cleaning operation on the information data of each data source, and establishing a metadata record for each data item in the cleaning process, the metadata record comprising an original field name of the data item, a data source system identifier, a data collection time, and a data conversion history; performing a structured conversion on the cleaned data according to a preset enterprise data model, and constructing a unified field mapping relationship, the enterprise data model comprising business information, procurement information, production information, sales information, financial information, technical information, and corporate governance information; adding a data quality label to the structured converted data, the data quality label comprising an integrity score, a consistency score, and a field confidence, and simultaneously storing the metadata record and the corresponding field data in association to form a structured data set with a source identifier.

[0007] Optionally, a data credibility evaluation model is constructed to obtain a credibility evaluation result of different data sources, and the construction of the data credibility evaluation model specifically comprises: setting different initial weights for different data sources of different categories, and simultaneously calculating a plurality of verification indicators of the different category data sources, and performing a weighted summation on the initial weights and the verification indicators to obtain a data source credibility score value.

[0008] Optionally, for a data source with a data source credibility score value higher than a threshold value, an interpolation method is used to calculate missing indicators existing in numerical structured data to obtain a first result; and for a data source with a data source credibility score value lower than the threshold value, a missing correction calculation is performed on the missing indicators existing in the numerical structured data to obtain a second result.

[0009] Optionally, the data source comprises one or more of a government information database, an enterprise database, and a network information database.

[0010] Optionally, the missing correction calculation is performed on the missing indicators to obtain the second result, and specifically comprises: constructing a basic feature vector based on existing complete data of the target enterprise in a high credibility data source; extracting industry reference data related to the missing indicators from the high credibility data source; fusing the basic feature vector and the industry reference data into an enhanced feature vector of the target enterprise; calculating a similarity score of the target enterprise and similar enterprises in the same industry based on the enhanced feature vector, selecting a preset number of similar enterprises with the highest similarity score, and extracting numerical values of the missing indicators of the similar enterprises in the high credibility data source; Take the arithmetic mean value of the numerical values corresponding to the missing indicators of multiple similar enterprises in the high credibility data source as the basic filling value, and take the basic filling value as the second result.

[0011] Optionally, after generating the basic filling value, it further comprises: Based on the enterprise core feature attribute parameter, the enterprises in the same industry are divided into multiple correction levels, and the corresponding layered correction coefficient of each correction level is determined. Based on the basic filling value and the layered correction coefficient, the correction value is calculated, and the correction value is taken as the second result.

[0012] Optionally, the enterprise core feature attribute parameter is extracted, and the enterprises are divided into multiple correction levels through the K-means clustering algorithm, and based on the correction levels obtained by clustering, the correction coefficient of each correction level is set.

[0013] The beneficial effects of the present application are that by binding the data source credibility evaluation and the missing indicator filling depth, combining the differentiated processing strategy and the layered correction mechanism, the technical defects of ignoring the data source quality difference and the individual characteristics of enterprises in the traditional data processing are effectively solved, the accuracy of data filling and the precision of enterprise portrait are significantly improved, the present application takes into account the generality and scalability, can adapt to the needs of different industries and scenes, and can adjust the parameters according to the industry dynamics, ensuring long-term effectiveness. The finally constructed multi-dimensional enterprise portrait can provide reliable decision basis for finance, government, investment and other fields, has wide market application prospect and important practical value. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 It is a flowchart of an enterprise user portrait data processing method based on big data analysis in the embodiments of the present application. DETAILED DESCRIPTION

[0015] The technical solutions of the present application are further described in detail below in combination with the drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application. In the following description, the expression "some embodiments" describes a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0016] In the following description, numerous specific details are set forth to provide a more thorough understanding of the present application. However, it will be apparent to one of skill in the art upon

[0017] It should be understood that the present application can be practiced with the elements in different order, and that none is necessary before another in order for the application to be practiced. Furthermore, some of the steps can be practiced simultaneously, and the order of the steps can be rearranged without necessarily departing from the application. It is therefore intended that the disclosure not be limited by the

[0018] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting, unless the opposite intention is clear. It is still further to be understood that all references to a or an shall be construed in the context to mean at least one and / or one or more, unless otherwise indicated to the contrary. Also, the use of terminology, such as terms like "comprising", "having", "containing" or "including" to describe the elements of a figure or method step are not meant to refer to there being any

[0019] For a more complete understanding of the present application, reference is made to the following description taken in conjunction with the accompanying drawings in which:

[0020] Reference is made to the following descriptions taken in conjunction with the accompanying drawings in which: Figure 1 The present application provides a kind of enterprise user portrait data processing method based on big data analysis, the processing method includes: S1, the information data of target enterprise is collected from multiple different data sources, respectively the information data of each data source is independently preprocessed, generates multiple structured data sets with source mark; In some optional examples, the data sources include one or more of the following: the national enterprise credit information public system database, the official record enterprise credit investigation agency database under the national small and medium-sized enterprise development fund, the database of relevant cooperative units, the enterprise official website database, the Tianyancha database, and the Qichacha database.

[0021] In the data collection process, the collection methods include one or more of the following: database export, crawler crawling, and manual collection input. For example, for government public databases such as the national enterprise credit information public system, the structured data such as enterprise registration information, shareholder information, and change records are obtained through standardized API interfaces and batch export functions; for commercial databases such as Tianyancha and Qichacha, the deep information such as enterprise association, intellectual property, and judicial risk is obtained through authorized data interfaces and customized crawler technology under the premise of compliance; for unstructured data sources such as enterprise official websites, intelligent crawlers combined with natural language processing technology are used to extract text information such as enterprise introduction, product services, and news updates; for specific information that is difficult to automatically obtain, such as enterprise internal management structure or special qualifications, professional data collection personnel are used for manual verification and input to ensure the completeness and accuracy of key data.

[0022] In some optional examples, the preprocessing method includes: data cleaning, data aggregation, and data reduction on the multi-source information data of the target enterprise, converting the information data after data reduction from text data to structured data, and eliminating data duplication and homonymy problems in the structured data.

[0023] In the data cleaning process, for missing values in key fields in government databases and commercial credit investigation platforms (such as incomplete registered capital information not disclosed by some enterprise official websites), intelligent marking is performed first for subsequent targeted filling based on credibility; for missing non-key fields, reasonable elimination is performed according to pre-set business rules; abnormal values are verified and corrected intelligently through 3σ criteria combined with industry experience thresholds; noise data is removed through stop word filtering and semantic disambiguation in natural language processing technology; The data aggregation link adopts an intelligent association strategy based on a core identifier to integrate the same type of enterprise information scattered in different data sources, such as the registration information of the national enterprise credit information public system, the equity structure information of Tianyancha / Qichacha, and the business correspondence information provided by the cooperation unit about the "enterprise basic attributes" to avoid analysis deviation caused by information fragmentation. The data reduction link implements accurate filtering based on information value evaluation, removes redundant information irrelevant to the portrait tag according to the core needs of enterprise portrait construction, such as marketing advertisement content in the official website of the enterprise and repeated non-core public opinion in network data, and only keeps the key information related to the enterprise operating status, core competence, credit record and the like, thereby significantly improving the subsequent data processing efficiency and analysis accuracy. After completing the basic preprocessing, the text data (such as unstructured content such as product introduction of the official website of the enterprise, industry media reports, cooperation agreement, etc.) after data reduction is converted into structured data. For three types of core named entities contained in the text, name entity (company name, personal name, place name), time expression (date, time) and numerical expression (currency value, percentage, etc.), the present application adopts a multi-method fusion recognition strategy. For time expression and numerical expression with strong regularity, accurate matching extraction is performed through a pre-set regular expression template. For complex and variable name entities, a joint recognition is performed in combination with a rule-based method, such as using the characteristics of enterprise name suffixes such as Limited Company and Joint Stock Company, a statistical machine learning method such as Hidden Markov Model and Conditional Random Field, and a deep learning method such as BERT pre-training model. Meanwhile, a domestic open-source Chinese language processing tool NLPIR is integrated for auxiliary optimization, so as to efficiently and accurately extract key entity information such as Shenzhen Huawei Technology Co., Ltd., December 1, 2023, and 50 million registered capital from the text, and map it to the standard field of structured data.

[0024] For the data duplication and homonymy problem in structured data, further optimization is needed. For the data duplication problem, by comparing the core key fields of structured data (such as enterprise unified social credit code, legal representative and company name combination), identify and delete completely duplicated records; for the case of partial field duplication but other fields with complementary information, keep the record with the highest completeness as the reference, and integrate the complementary information of other records into the reference record to ensure the uniqueness and completeness of the data. For the homonymy problem (such as there may be multiple "XX Technology Co., Ltd." in different regions, or there may be "abbreviation" and "full name" mixed use of the same enterprise), clustering-based disambiguation is used, by calculating the similarity of entity attributes (such as registered address, business scope, registered capital), similar entities are clustered into a class, and different expressions of the same entity are clarified; or entity linking-based disambiguation is used, the entities in structured data are linked and matched with the authoritative entity library (such as the enterprise directory of the national enterprise credit information public system), to ensure that each entity corresponds to a unique real enterprise.

[0025] S2, constructing a data credibility evaluation model to obtain the credibility score of each data source; In some optional examples, the credibility evaluation results of different data sources are obtained by the data credibility evaluation model. When constructing the data credibility evaluation model, first set the initial weights of different data sources. For government information databases (such as the national enterprise credit information public system and the official record enterprise credit investigation agency database under the national small and medium-sized enterprise development fund), since they are operated by government departments or national authorized agencies, the data collection process is subject to laws and regulations, and the data quality is guaranteed by the system, so they are given a higher initial weight, which is set in the range of 0.7 to 0.9. For enterprise databases (such as Tianyan, Qichaicha and other commercial databases), their data comes from public channels and some authorized cooperation, although there is a certain quality guarantee, but there is a commercial interest influence, the initial weight is set in the range of 0.4 to 0.6. For network information databases (such as enterprise website, social media, industry forum, etc.), due to the high openness and easy tampering of data, the initial weight is relatively low, and the initial weight is set in the range of 0.1 to 0.3. These initial weights are not fixed, but are used as a reference value for subsequent dynamic evaluation, establishing a differentiated credibility evaluation starting point for different data sources.

[0026] After obtaining the initial weights of different data sources, a plurality of verification indicators of different categories of data sources are calculated, including but not limited to a historical accuracy indicator, calculated by comparing the historical records of the data source with the authoritative verification results (such as audit reports, official announcements); a timeliness indicator, evaluated based on the matching degree of data update frequency and industry change speed; a consistency indicator, measuring the internal logical consistency of the same enterprise at different time points or different business dimension data; a cross-validation indicator, calculated by horizontal comparison with the same data of other high-trust data sources. For example, for enterprise financial data, the system will analyze the historical consistency of the financial indicators provided by the data source with the listed company announcements or audit reports; for enterprise basic information, the time difference between data update and business change registration will be evaluated; for public opinion data, the proportion of positive, negative and neutral evaluations will be checked to see if it conforms to the industry benchmark distribution.

[0027] Further, the initial weights and the verification indicators are weighted and summed to obtain a data source credibility score value, and the calculation formula is: credibility score value = a x initial weight + b1 x historical accuracy indicator + b2 x timeliness indicator + b3 x consistency indicator + b4 x cross-validation indicator, wherein a + b1 + b2 + b3 + b4 = 1, and each coefficient is dynamically adjusted according to the application scenario.

[0028] S3, according to the credibility score value of each data source, a missing indicator existing in the numerical structured data is calculated in different ways to obtain a first result or a second result. In some examples, when the data source credibility score value is higher than a preset threshold, for example, set to 0.65, it indicates that the overall quality of the data source is reliable, and the missing indicators are more likely to be random missing. At this time, the missing indicators of the numerical structured data are calculated by using the interpolation method to obtain the first result, and the interpolation method includes linear interpolation, polynomial interpolation or spline interpolation, and the specific selection is determined according to the time series characteristics and business rules of the data.

[0029] When the data source credibility score value is lower than the threshold, it indicates that the data source has systematic quality problems, and the missing indicators may be non-random missing. Simple interpolation may cause deviation accumulation, so the missing indicators existing in the numerical structured data are calculated to obtain the second result. Through the differential processing strategy based on the credibility score, the technical defects of using a single processing method for all missing data in the traditional method are avoided, and the accuracy of data filling and the reliability of enterprise portrait are significantly improved.

[0030] In some optional examples, the missing indicators are calculated for missing correction to obtain the second result, specifically including: The missing indicators are calculated for missing correction to obtain the second result, specifically including: construct a basic feature vector based on the complete data of the target enterprise in the high-credibility data source; extract industry reference data related to the missing indicators from the high-credibility data source; fuse the basic feature vector and the industry reference data into an enhanced feature vector of the target enterprise; calculate the similarity scores of the target enterprise and other enterprises in the same industry based on the enhanced feature vector, select a preset number of similar enterprises with the highest similarity scores, and extract the numerical values of the missing indicators of the similar enterprises in the high-credibility data source; take the arithmetic mean of the numerical values of the missing indicators of the similar enterprises in the high-credibility data source as the basic filling value, and take the basic filling value as the second result.

[0031] Specifically, relying on the complete data items of the target enterprise obtained in the high-credibility data source, a basic feature vector that can comprehensively reflect the core attributes of the target enterprise is constructed. The basic feature vector needs to cover the effective data of the target enterprise in key dimensions such as business information, financial status, production and operation, technical research and development, and corporate governance, to ensure that the vector can accurately depict the individual attributes and operating status of the target enterprise, and provide a reliable individual feature basis for subsequent data processing.

[0032] Further, from the high-credibility data source with a credibility score higher than a preset threshold, industry reference data related to the missing indicators is extracted. The industry reference data should include key information such as the distribution characteristics, average level, and fluctuation range of indicators in the same industry, to ensure the authority and reliability of the data source and the relevance to the missing indicators, and to provide industry-level common benchmark basis for filling the missing indicators.

[0033] Then, the basic feature vector constructed above and the extracted industry reference data are organically fused to form an enhanced feature vector of the target enterprise. The fusion process needs to use a preset data fusion algorithm to ensure that the individual uniqueness of the target enterprise embodied in the basic feature vector and the industry common rules contained in the industry reference data are effectively combined, so that the enhanced feature vector has more comprehensive and accurate feature expression capability.

[0034] Based on the enhanced feature vector, a preset similarity calculation algorithm (such as cosine similarity algorithm, Euclidean distance algorithm, etc.) is used to calculate the similarity scores between the target enterprise and other enterprises in the same industry. According to the order of similarity scores from high to low, a preset number (such as 5-10) of enterprises are selected as the most similar enterprises to the target enterprise, and further the complete numerical data of the missing indicators of these similar enterprises are extracted from the high-credibility data source to ensure the reliability and effectiveness of the extracted data and avoid interference of low-quality data on the filling result.

[0035] The statistical analysis is performed on the values of the missing indicators of the extracted similar enterprises in the high-credibility data source, and the arithmetic mean of the values is calculated to determine the basic filling value of the missing indicators as the second result. Using the high-credibility data mean of the most similar enterprises in the same industry for filling can make full use of the industry commonality and the individual characteristics of similar enterprises, effectively avoid the possible systematic bias of low-credibility data sources, and ensure the objectivity, rationality and accuracy of the filling result of the missing indicators, providing high-quality data support for the subsequent construction of enterprise portrait model.

[0036] In some optional examples, after generating the basic filling value, it further includes: Based on the enterprise core characteristic attribute parameters, the enterprises in the same industry are divided into multiple correction levels, and a layered correction coefficient corresponding to each correction level is determined; The correction value is calculated based on the basic filling value and the layered correction coefficient, and the correction value is taken as the second result.

[0037] Further, the enterprise core characteristic attribute parameters are extracted, and the enterprises are divided into multiple correction levels by a K-means clustering algorithm, and the correction coefficient of each correction level is set based on the correction levels obtained by clustering.

[0038] To further improve the accuracy and business adaptability of the filling result of the missing indicators, the present application introduces a layered correction mechanism to realize the secondary optimization of the filling result through the fine analysis of the core characteristic attributes of the enterprises. Specifically, first, the enterprises in the same industry are scientifically stratified based on the core characteristic attribute parameters, which include but are not limited to enterprise size grade (such as divided into micro, small, medium and large according to revenue), technology intensity index (such as R&D personnel proportion, R&D input proportion), market coverage breadth (such as regional, national and transnational) and development stage identifier (such as start-up, growth, mature and decline), which not only conforms to the industry general cognition, but also ensures objectivity through data-driven and avoids the deviation caused by subjective division.

[0039] By using a clustering analysis algorithm (such as K-means clustering, hierarchical clustering, etc.), the enterprise set in the same industry is divided into multiple correction levels, and the enterprises in each level have significant homogeneity in business model and indicator performance, and the levels have obvious heterogeneity, for example, the enterprises in the information technology industry are divided into three levels of start-up technology enterprises, growth technology enterprises and mature technology enterprises, or the enterprises in the equipment manufacturing industry are divided into labor-intensive manufacturing enterprises, capital-intensive manufacturing enterprises and technology-intensive manufacturing enterprises, to ensure the consistency of the key indicator fluctuation rules of the enterprises in each level.

[0040] After completing the correction level division, based on the historical data accumulated in the high credibility data source, the corresponding hierarchical correction coefficient is calculated and determined for each correction level. The determination of the hierarchical correction coefficient needs to fully consider the systematic deviation degree of the index distribution characteristics of each level enterprise and the industry benchmark level, which is specifically obtained by calculating the deviation of the key indicators (indicators of the same type as the missing indicators) of each level enterprise relative to the industry average level. For example, in the filling scenario of the R&D input proportion index, the R&D input proportion of the growth-type technology enterprise level is usually 15%-25% higher than the industry average level, so the corresponding hierarchical correction coefficient of this level is set to 1.15-1.25; the R&D input proportion of the mature technology enterprise level is close to or slightly lower than the industry average level, so its hierarchical correction coefficient is set to 0.95-1.05. In order to adapt to the dynamic changes of the industry development, the hierarchical correction coefficient needs to be recalculated and updated regularly through newly obtained high credibility data, to ensure that the coefficient always reflects the latest development trend of the industry, and to avoid correction deviation caused by market environment, policy adjustment and other factors.

[0041] In determining the correction level to which the target enterprise belongs, the membership score of the target enterprise in each correction level is calculated, which is calculated based on the distance (such as Euclidean distance, cosine distance) between the core characteristic attribute parameters of the target enterprise and the center point of each level (a vector composed of the mean values of the enterprise characteristics in the level), so that the target enterprise can have partial characteristics of multiple correction levels. For example, an equipment manufacturing enterprise in the business transformation period may have both capital-intensive and technology-intensive characteristics. According to the membership proportion of the enterprise in the two levels, such as capital-intensive level membership 0.6 and technology-intensive level membership 0.4, the weighted mixed hierarchical correction coefficient is calculated, such as capital-intensive coefficient 1.05 and technology-intensive coefficient 1.2, mixed coefficient = 1.05 x 0.6 + 1.2 x 0.4 = 1.11, which effectively avoids the deviation of the traditional hard division method in dealing with boundary enterprises and improves the adaptability of the correction process.

[0042] The basic filling value is multiplied by the hierarchical correction coefficient to obtain the correction value, and the correction value is taken as the second result.

[0043] S4, form the preprocessed information data based on the first filling result and the second filling result, and construct an enterprise portrait model based on the preprocessed information.

[0044] In this embodiment, the enterprise portrait is a multi-dimensional and multi-level user model. According to the type of information data of the target enterprise, a vector space is defined as the enterprise portrait, and the dimensional characteristics constituting the enterprise portrait include enterprise basic attributes, enterprise business scope, enterprise main products, enterprise correlation analysis, and enterprise industry evaluation. The specific model structure is as follows: Enterprise =<BaseInfo, BusiScope, MainProduct, Relation, Comments> Wherein, BaseInfo represents enterprise basic attributes, BusiScope represents enterprise business scope, MainProduct represents enterprise main product, Relation represents enterprise correlation analysis, and Comments represents enterprise industry evaluation, and the five tuple characteristic labels jointly constitute enterprise portrait.

[0045] From the preprocessed information data formed based on the first filling result and the second filling result, corresponding data is extracted and filled into the model, so as to obtain the enterprise portrait.

[0046] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. The protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for processing enterprise user profile data based on big data analysis, characterized in that, The processing method includes: Information data of the target enterprise is collected from multiple different data sources. The information data collected from each data source is preprocessed independently to generate multiple structured datasets with source labels. Build a data credibility assessment model to obtain credibility scores for each data source; Based on the credibility scores of each data source, the missing indicators in the numerical structured data are calculated using different methods to obtain the first result or the second result. A corporate information dataset is formed based on the first and second results, and a corporate profile model is constructed based on the corporate information dataset.

2. The enterprise user profile data processing method based on big data analysis according to claim 1, characterized in that, The information data collected from each data source is preprocessed independently to generate multiple structured datasets with source labels, specifically including: Each data source is assigned a unique source identifier, which includes the data source category code, the data source name code, and the data collection timestamp. Data cleaning is performed independently for information data from each data source. During the cleaning process, a metadata record is created for each data item. The metadata record includes the original field name of the data item, the data source system identifier, the data collection time, and the data transformation history. The cleaned data is structured and transformed according to a preset enterprise data model to construct a unified field mapping relationship. The enterprise data model includes business registration information, procurement information, production information, sales information, financial information, technical information, and corporate governance information. Data quality tags are added to the structured data, including integrity scores, consistency scores, and field confidence scores. At the same time, metadata records are associated with and stored with the corresponding field data to form a structured dataset with source identification.

3. The enterprise user profile data processing method based on big data analysis according to claim 1, characterized in that, A data credibility assessment model is constructed to obtain credibility assessment results for different data sources. The construction of the data credibility assessment model specifically includes: setting different initial weights for different data sources, calculating multiple verification indicators for different categories of data sources, and weighted summing of the initial weights and verification indicators to obtain the data source credibility score.

4. The enterprise user profile data processing method based on big data analysis according to claim 3, characterized in that, For data sources with a data source credibility score higher than the threshold, interpolation is used to calculate the missing indicators in the numerical structured data to obtain the first result; for data sources with a data source credibility score lower than the threshold, missing indicator correction is performed on the missing indicators in the numerical structured data to obtain the second result.

5. The enterprise user profile data processing method based on big data analysis according to claim 3, characterized in that, The data source includes one or more of the following: government information databases, enterprise databases, and network information databases.

6. The enterprise user profile data processing method based on big data analysis according to claim 3, characterized in that, Missing indicators are corrected for to obtain a second result, which includes: Construct a basic feature vector based on the target company's existing complete data from a high-reliability data source; Extract industry reference data related to missing indicators from highly reliable data sources; The basic feature vector is fused with industry reference data to create an enhanced feature vector for the target company; Calculate the similarity score between the target company and companies in the same industry based on the enhanced feature vector, select a preset number of similar companies with the highest similarity scores, and extract the values ​​corresponding to the missing indicators of multiple similar companies in a high-confidence data source. The arithmetic mean of the missing indicators corresponding to multiple similar enterprises in a high-reliability data source is used as the base imputation value, and the base imputation value is used as the second result.

7. The enterprise user profile data processing method based on big data analysis according to claim 6, characterized in that, After generating the base fill value, the process also includes: Based on the core characteristic parameters of enterprises, enterprises in the same industry are divided into multiple correction levels, and the corresponding hierarchical correction coefficient for each correction level is determined. The correction value is calculated based on the basic filling value and the stratification correction coefficient, and the correction value is used as the second result.

8. The enterprise user profile data processing method based on big data analysis according to claim 7, characterized in that, The core characteristic parameters of enterprises are extracted, and the enterprises are divided into multiple correction levels by K-means clustering algorithm. Based on the correction levels obtained by clustering, the correction coefficient of each correction level is set.

Images