An intelligent data analysis system and method for a digital economic platform

Abstract

The application provides a kind of intelligent data analysis system and method for digital economy platform, it is related to data analysis technical field, by real-time access the multi-source business data stream of digital economy platform;Multi-source business data stream is carried out parallel modal coding, and the feature vector of different modal is mapped to shared semantic hidden space, generates uniform representation vector;The dynamic business knowledge graph of digital economy platform is constructed, and then the dynamic business knowledge graph is updated and reasoned, and the context perception vector is output;Receive the analysis task description issued under the upper layer business scene, and construct the data view matched with different business scenes by the analysis task description and context perception vector;According to the semantic analysis of the business scene demand of digital economy platform according to data view, generates the semantic analysis report of business scene demand.The application can realize the semantic level deep fusion of multi-modal business data of digital economy platform, to adapt to the analysis needs of multi-scene of digital economy platform.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and more specifically, to an intelligent data analysis system and method for digital economy platforms. Background Technology

[0002] With the rapid development of the digital economy industry, data analytics technology has become a core support for digital economy platforms to mine data value and empower business decisions. It can efficiently process and deeply analyze the massive amounts of multi-source business data gathered by the platform, and uncover the business patterns and potential value behind the data. This technology can not only provide data support for the refined operation, resource scheduling and scenario-based services of digital economy platforms, but also promote the efficient circulation and utilization of platform data elements, help digital economy platforms upgrade from data accumulation to value transformation, and improve the overall operational efficiency and core competitiveness of the platform.

[0003] In existing technologies, the data generated by digital economy platforms exhibits significant characteristics of massive volume and diversity. There are clear semantic barriers between different types of data, making it difficult for traditional data analysis techniques to achieve deep integration of various data types. They can only complete superficial data organization and simple analysis, failing to uncover deep relationships between data. Furthermore, existing technologies lack the ability to adapt to the diverse business scenarios of digital economy platforms, making it difficult to provide targeted analytical support based on the actual needs of different business scenarios. This results in insufficient practicality of data analysis results, failing to fully meet the multi-scenario and refined analytical needs of digital economy platforms, thus hindering their further development. Therefore, how to achieve semantic-level deep integration of multimodal business data from digital economy platforms to adapt to their multi-scenario analytical needs has become a challenge for the industry. Summary of the Invention

[0004] This application provides an intelligent data analysis system and method for digital economy platforms, which can achieve semantic-level deep fusion of multimodal business data of digital economy platforms, thereby adapting to the analysis needs of multiple scenarios of digital economy platforms.

[0005] Firstly, this application provides an intelligent data analysis method for digital economy platforms, the intelligent data analysis method comprising the following steps:

[0006] Real-time access to multi-source business data streams from the digital economy platform;

[0007] Parallel modal encoding is performed on the multi-source business data streams, and the feature vectors of different modalities are mapped to a shared semantic latent space to generate a unified representation vector;

[0008] A dynamic business knowledge graph of a digital economy platform is constructed, and the unified representation vector is used as the incremental information of entity nodes in the dynamic business knowledge graph. The dynamic business knowledge graph is then updated and reasoned, and a context-aware vector containing rich semantic association is output.

[0009] Receive the analysis task description issued by the upper-layer business scenario, and construct a data view matching different business scenarios through the analysis task description and the context-aware vector;

[0010] Based on the data view, semantic analysis is performed on the business scenario requirements of the digital economy platform to generate a semantic analysis report on the business scenario requirements.

[0011] In this embodiment, parallel modal coding of the multi-source service data stream specifically includes:

[0012] The multi-source business data stream is subjected to adaptive modal segmentation based on business characteristics to obtain different types of modal data streams;

[0013] Heterogeneous preprocessing is performed on various modal data streams;

[0014] Configure independent heterogeneous coding branches for each type of modal data stream and set up coding synchronization scheduling nodes;

[0015] Synchronous heterogeneous encoding is performed on the preprocessed data streams of various modalities to output feature vectors that retain the core business characteristics of each modality.

[0016] In this embodiment, mapping feature vectors from different modalities to a shared semantic latent space to generate a unified representation vector specifically includes:

[0017] Constructing a semantic implicit space for dynamic sharing of digital economy business;

[0018] Perform cross-domain semantic alignment on feature vectors of different modalities;

[0019] Determine the importance and assign weights of each modality of data flow in the business of the digital economy platform;

[0020] By assigning weights based on the importance of each modal data stream in the digital economy platform's business, the feature vectors after cross-domain semantic alignment are mapped to the semantic latent space;

[0021] All feature vectors mapped to the semantic latent space are subjected to feature fusion with business semantic attention to generate a unified representation vector adapted to digital economy business.

[0022] In this embodiment, constructing a dynamic business knowledge graph for the digital economy platform specifically includes:

[0023] Extract entity nodes from various business areas within the digital economy platform;

[0024] Define the semantic constraints and association weight assignment rules for each type of entity node;

[0025] A dynamic business knowledge graph of the digital economy platform is constructed based on the semantic constraints and association weight assignment rules of various types of entity nodes.

[0026] In this embodiment, the unified representation vector is used as incremental information for entity nodes in the dynamic business knowledge graph, thereby updating and reasoning about the dynamic business knowledge graph, and outputting a context-aware vector containing rich semantic associations. Specifically, this includes:

[0027] Incremental information layering is performed on entity nodes in the dynamic business knowledge graph using the unified representation vector.

[0028] The attribute features of entity nodes are updated based on the embedded incremental information, and the semantic rules of the dynamic business knowledge graph are used for association reasoning to obtain explicit semantic association relationships between entity nodes.

[0029] The implicit semantic relationships between entity nodes are determined based on the business association logic of the digital economy platform;

[0030] Cross-node global semantic fusion is performed on the updated entity node attribute features, explicit semantic relationships between entity nodes, and implicit semantic relationships between entity nodes to output a context-aware vector containing rich semantic relationships.

[0031] In this embodiment, constructing a data view matching different business scenarios using the analysis task description and the context-aware vector specifically includes:

[0032] Fine-grained semantic decoupling is performed on the analysis task description to obtain semantic decoupling features;

[0033] The context-aware vector is reconstructed using the semantic decoupling features to obtain a set of context-aware feature vectors.

[0034] A data view matching different business scenarios is constructed based on the scenario-based feature vector set.

[0035] In this embodiment, the semantic analysis of the business scenario requirements of the digital economy platform based on the data view, and the generation of a semantic analysis report of the business scenario requirements, specifically includes:

[0036] Extract semantic information from the data view and label each type of semantic information with the corresponding business scenario requirement tags;

[0037] Perform bidirectional semantic alignment verification on the annotated semantic information;

[0038] Business analysis is performed on the business scenario requirements of the digital economy platform based on the bidirectional semantic alignment verification results.

[0039] Generate a semantic analysis report of business scenario requirements based on the business parsing results.

[0040] In this embodiment, the unified representation vector represents the semantic representation vector of multimodal business data of the digital economy platform.

[0041] In this embodiment, the context-aware vector represents the feature vector that perceives the semantic association of the entire business context of the digital economy platform.

[0042] Secondly, this application provides an intelligent data analysis system for digital economy platforms, used to execute an intelligent data analysis method for digital economy platforms, the intelligent data analysis system comprising:

[0043] The data stream receiving module is used to access multi-source business data streams from the digital economy platform in real time.

[0044] The feature vector mapping module is used to perform parallel modal encoding on the multi-source business data stream and map the feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector.

[0045] The semantic context association module is used to construct a dynamic business knowledge graph of the digital economy platform. It uses the unified representation vector as incremental information of entity nodes in the dynamic business knowledge graph, and then updates and infers the dynamic business knowledge graph to output a context-aware vector containing rich semantic association.

[0046] The business scenario matching module is used to receive the analysis task description issued by the upper-layer business scenario, and construct a data view matching different business scenarios through the analysis task description and the context-aware vector.

[0047] The analysis report generation module is used to perform semantic analysis on the business scenario requirements of the digital economy platform based on the data view, and generate a semantic analysis report on the business scenario requirements.

[0048] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0049] The system integrates multi-source business data streams from a digital economy platform in real time; performs parallel modal encoding on the multi-source business data streams and maps feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector; constructs a dynamic business knowledge graph of the digital economy platform, using the unified representation vector as incremental information for entity nodes in the dynamic business knowledge graph, thereby updating and reasoning about the dynamic business knowledge graph and outputting a context-aware vector containing rich semantic associations; receives analysis task descriptions from upper-layer business scenarios, constructs data views matching different business scenarios using the analysis task descriptions and the context-aware vectors; performs semantic analysis on the business scenario requirements of the digital economy platform based on the data views, and generates a semantic analysis report on the business scenario requirements.

[0050] Therefore, this application demonstrates that semantic analysis of the business scenario requirements of the digital economy platform can be performed based on the data view, generating a semantic analysis report of the business scenario requirements. Firstly, by accessing the multi-source business data streams of the digital economy platform in real time, it ensures that the aggregated data comprehensively covers all types of business scenarios on the platform and is timely, solving the problems of incomplete or untimely data sources leading to incomplete fusion results and insufficient scenario adaptability in traditional data analysis techniques. Secondly, parallel modal encoding of the multi-source business data streams and mapping the feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector effectively breaks down semantic barriers between different types of data, achieving a leap from surface-level splicing to underlying semantic-level feature fusion of multi-modal data, solving the core pain point of traditional techniques' difficulty in achieving deep fusion of various types of data. Furthermore, a dynamic business knowledge graph of the digital economy platform is constructed, and the unified representation vector is used as incremental information for updating entity nodes. By reasoning and outputting context-aware vectors rich in semantic associations, the system can deeply mine the profound business relationships between multimodal data, giving the fused data clear business semantic attributes and inference value, thus further enhancing the depth and practicality of multimodal data semantic fusion. Furthermore, by receiving analysis task descriptions from upper-level business scenarios and constructing data views matching different business scenarios through analysis task descriptions and context-aware vectors, the system can achieve precise alignment between semantically fused data and specific business scenario requirements, breaking the predicament of data fusion being disconnected from business scenarios and improving the adaptability of data analysis to different business scenarios. Finally, by performing semantic analysis on business scenario requirements based on the data views and generating semantic analysis reports, the system can transform the fused deep semantic data into practical analysis results tailored to specific business scenario needs, effectively improving the practicality and relevance of data analysis results and fully meeting the multi-scenario, refined analysis needs of digital economy platforms.

[0051] In summary, the technical solution adopted in this application can achieve semantic-level deep fusion of multimodal business data of digital economy platforms, thereby adapting to the analysis needs of digital economy platforms in multiple scenarios. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is an exemplary flowchart of an intelligent data analysis method for digital economy platforms provided in this application;

[0054] Figure 2 This is a flowchart illustrating the process of generating a unified representation vector according to the present application;

[0055] Figure 3 This is a flowchart illustrating the process of generating a semantic analysis report based on the information provided in this application;

[0056] Figure 4 This is a module structure diagram of an intelligent data analysis system for a digital economy platform provided in this application. Detailed Implementation

[0057] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] This application provides an intelligent data analysis system and method for digital economy platforms. Its core is real-time access to multi-source business data streams from the digital economy platform; parallel modal encoding of the multi-source business data streams, mapping feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector; constructing a dynamic business knowledge graph of the digital economy platform, using the unified representation vector as incremental information for entity nodes in the dynamic business knowledge graph, thereby updating and reasoning about the dynamic business knowledge graph, and outputting a context-aware vector containing rich semantic associations; receiving analysis task descriptions from upper-layer business scenarios, constructing data views matching different business scenarios through the analysis task descriptions and the context-aware vectors; performing semantic analysis on the business scenario requirements of the digital economy platform based on the data views, and generating a semantic analysis report of the business scenario requirements.

[0059] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of an intelligent data analysis method for a digital economy platform according to this embodiment of the present application. The intelligent data analysis method includes the following steps:

[0060] In step S1, the multi-source business data stream of the digital economy platform is accessed in real time.

[0061] It should be noted that the multi-source business data streams mentioned in this application refer to various real-time data streams with different data formats and business attributes from different business systems during the operation of the digital economy platform. Specifically, these include structured transaction data streams from the transaction system within the digital economy platform, time-series behavior data streams from the user behavior system, semi-structured qualification data streams from the merchant management system, and unstructured announcement data streams from the platform operation system.

[0062] In practical implementation, firstly, a multi-source data stream real-time access gateway for the digital economy platform is built. A distributed architecture is used to deploy access nodes, covering all business systems of the platform (transaction system, user behavior system, merchant management system, and operation system). Corresponding access protocols are configured for the data stream characteristics of different business systems. Structured transaction data streams (including transaction amount, transaction time, and information of both parties) use JDBC; time-series user behavior data streams (including user click time, page views, and dwell time) use MQTT lightweight messaging protocol; semi-structured merchant qualification data streams (including scanned copies of merchant business licenses and qualification certificates) use HTTP; and unstructured platform announcement data streams use HTTP. Simultaneously, a data receiving buffer (1024MB) is set in the access gateway to temporarily store the real-time received data streams to prevent data loss. Secondly, the data streams from each business system are output... The ports are configured with a data push frequency of 1 second per instance to ensure the real-time performance of the data stream. Access nodes monitor the data stream output ports of each business system in real time. When a new data stream is detected, the corresponding protocol transmission channel is immediately activated to encrypt the data stream (using AES encryption algorithm with a 128-bit key length) to prevent data leakage. Next, a preliminary data format verification module is set up in the access gateway to verify the format of various received data streams, eliminating invalid data streams with format errors or incomplete data (such as transaction data streams missing transaction amounts or behavior data streams without user identifiers). Data streams that pass verification are uniformly formatted (by adding business type tags to each type of data stream, such as "transaction" or "user behavior"). Finally, all verified and marked real-time data streams are aggregated and integrated, and the resulting aggregated data is used as the multi-source business data stream of the digital economy platform, thus completing the real-time access process for multi-source business data streams.

[0063] In step S2, the multi-source service data stream is subjected to parallel modal encoding, and the feature vectors of different modalities are mapped to a shared semantic latent space to generate a unified representation vector.

[0064] In this embodiment, parallel modal coding of the multi-source service data stream can be achieved using the following steps:

[0065] The multi-source business data stream is subjected to adaptive modal segmentation based on business characteristics to obtain different types of modal data streams;

[0066] Heterogeneous preprocessing is performed on various modal data streams;

[0067] Configure independent heterogeneous coding branches for each type of modal data stream and set up coding synchronization scheduling nodes;

[0068] Synchronous heterogeneous encoding is performed on the preprocessed data streams of various modalities to output feature vectors that retain the core business characteristics of each modality.

[0069] It should be noted that, in this application, the adaptive modal segmentation based on business characteristics refers to the process of segmenting multi-source business data streams according to their business attributes, data formats, semantic types, and application scenarios using an adaptive classification method; the modal data stream refers to a class of business data streams with the same or similar business characteristics, data formats, and semantic types; the heterogeneous preprocessing refers to the process of removing data noise, supplementing missing information, and unifying data formats to ensure data quality, based on the heterogeneous characteristics (data format, structure, semantic differences) of different modal data streams; the heterogeneous coding branch refers to a coding module configured separately for each type of modal data stream, adapted to its data characteristics and business requirements, used to extract the core business features of the corresponding modal data stream; the coding synchronization scheduling node refers to a control module used to coordinate the coding timing synchronization of each heterogeneous coding branch, control the coding progress, and ensure the synchronous output of coding results for various modal data streams; the feature vector represents a vector characterizing the core business information of the modal data stream.

[0070] In specific implementation, first, for the multi-source business data streams of the digital economy platform accessed in real time, the K-means clustering algorithm based on business characteristics is used for adaptive modal division. Taking business characteristic tags (transaction attributes, user attributes, merchant attributes, operation attributes) as the clustering basis, the preset number of clusters is 4. The multi-source business data streams are iteratively clustered through the K-means clustering algorithm (the number of iterations is set to 50, and the convergence threshold is set to 0.001). The data streams classified into the same category after clustering are used as the modal data streams of the corresponding categories, and are respectively marked as transaction class modal data streams, user behavior class modal data streams, merchant class modal data streams, and platform operation class modal data streams. Secondly, heterogeneous preprocessing is performed on each type of modal data stream. For the transaction class modal data stream, the 3σ principle is used to eliminate abnormal transaction data (such as abnormally high / low transaction amounts), the mean filling method is used to supplement missing numerical fields (such as missing transaction fees), the mode filling method is used for categorical fields (such as missing transaction types), and all numerical fields are mapped to the [0, 1] interval through the min-max normalization method. After the preprocessing is completed, it is used as the preprocessed transaction class modal data stream. For the user behavior class time series data stream, the time series alignment technology is used to align user click, browse, order placement and other behavior records at a time granularity of 1 second. The adjacent value filling method is used to supplement missing time series data, and duplicate behavior records within the same time granularity of the same user are eliminated. After the preprocessing is completed, it is used as the preprocessed user behavior class modal data stream. For the merchant class semi-structured data stream, the label parsing technology is used to extract core information such as merchant qualifications, business scope, and entry time, convert the semi-structured data into a structured format, and supplement missing merchant information (such as the missing industry of the merchant). After the preprocessing is completed, it is used as the preprocessed merchant class modal data stream. For the platform operation class unstructured data stream (such as operation announcements, fault notifications), the jieba word segmentation algorithm is used to perform word segmentation on the text data, remove meaningless stop words such as "de, le, shi", and convert the text data into a structured feature matrix through the TF-IDF algorithm. After the preprocessing is completed, it is used as the preprocessed platform operation class modal data stream. Then, an independent heterogeneous coding branch is configured for each type of preprocessed modal data stream. For the transaction class modal data stream, a multi-layer perceptron is used to construct a heterogeneous coding branch, and 3 fully connected layers are set (the dimension of the input layer is the same as the dimension of the preprocessed transaction class modal data stream, the dimension of the hidden layer is set to 128, and the dimension of the output layer is set to 64). The preprocessed transaction class modal data stream is input, and the core business characteristics are gradually extracted through the fully connected layers. For the user behavior class time series data stream, a long short-term memory network is used to construct a heterogeneous coding branch, the dimension of the hidden layer is set to 64, and the time step is set to 30. The preprocessed user behavior class time series data stream is input, and the user behavior time series characteristics are extracted through the long short-term memory network;For merchant-related modal data streams, a convolutional neural network is used to construct heterogeneous coding branches, with three convolutional kernels (sizes 3, 4, and 5 respectively). The preprocessed merchant-related modal data stream is input, and core merchant features are extracted through convolutional and pooling layers. For platform operation-related modal data streams, a BERT base model is used to construct heterogeneous coding branches. The preprocessed text feature matrix is ​​input, and the output of the last layer of the model is used as the initial features. The feature dimension is adjusted to 64 dimensions through a fully connected layer. A coding synchronization scheduling node is also set up to collect the coding progress of each heterogeneous coding branch in real time. When all coding branches have completed encoding a single batch of data, a synchronization output command is triggered to ensure that all coding branches output their encoding results synchronously. Finally, synchronous heterogeneous coding processing is initiated. Each heterogeneous coding branch, according to the instructions of the coding synchronization scheduling node, synchronously performs encoding operations on the preprocessed modal data streams. The 64-dimensional feature vectors output by the heterogeneous coding branches corresponding to the transaction, user behavior, merchant, and platform operation modal data streams are used as feature vectors to retain the core business features of each modality, thereby completing the parallel modal coding process for multi-source business data streams.

[0071] Preferably, in this embodiment, feature vectors from different modalities are mapped to a shared semantic latent space to generate a unified representation vector, referencing... Figure 2 As shown in the figure, this is a schematic diagram of the process for generating a unified representation vector in some embodiments of this application. In this embodiment, generating a unified representation vector can be achieved by the following steps:

[0072] In step S21, a semantic latent space for dynamic sharing of digital economy business is constructed;

[0073] In step S22, cross-domain semantic alignment is performed on the feature vectors of different modalities;

[0074] In step S23, the importance allocation weight of each modal data stream in the digital economy platform business is determined;

[0075] In step S24, the feature vectors after cross-domain semantic alignment are mapped to the semantic latent space by assigning weights based on the importance of each modal data stream in the digital economy platform business.

[0076] In step S25, feature fusion with business semantic attention is performed on all feature vectors mapped to the semantic latent space to generate a unified representation vector adapted to digital economy business.

[0077] It should be noted that the semantic latent space representation for dynamic sharing of digital economy business described in this application is used to uniformly carry the shared space of different modal feature vectors to achieve cross-modal semantic interoperability; the cross-domain semantic alignment processing represents the process of eliminating semantic deviations of different modal feature vectors, unifying feature distribution, and enabling various modal features to have the same semantic dimension; the importance allocation weight represents the weight of the importance of different modal data streams in the core business of the digital economy platform; and the unified representation vector represents the semantic representation vector of multimodal business data of the digital economy platform.

[0078] In specific implementation, firstly, the feature distribution of various modal feature vectors (transaction, user behavior, merchant, and platform operation, all 64-dimensional) output by parallel modal encoding is used as the training basis. The spatial dimension is set to 128 dimensions, and the feature similarity metric within the space is defined as cosine similarity. Through adversarial training between the generator and discriminator (training iterations are set to 100, and the learning rate is set to 0.001), the space can adapt to the semantic distribution of various modal features. At the same time, a dynamic update trigger condition is set (when the average similarity between the newly accessed modal feature vector and the existing features in the space is less than 0.7) to ensure the dynamic adaptability of the space. The space that has been trained and has the ability to be dynamically updated is used as the semantic latent space for dynamic sharing of digital economy businesses. Next, using the transaction-type modal feature vector as the baseline vector, the cosine similarity between the user behavior, merchant, and platform operation modal feature vectors and the baseline vector is calculated. Abnormal feature vectors with a cosine similarity lower than 0.6 are removed, and the remaining feature vectors undergo dimensional calibration (the semantic offset of all modal feature vectors is corrected to a consistent range through linear transformation). The calibrated feature vectors of each modality are used as feature vectors after cross-domain semantic alignment. Then, the analytic hierarchy process (AHP) is used to assign importance weights to each type of modal data stream, and a business importance judgment matrix for the digital economy platform is constructed (using transaction, user behavior, merchant, and platform operation as judgment objects and business contribution as the judgment standard). The matrix consistency is then tested (consistency index CI < 0).1) Verify the rationality of the judgment matrix, calculate the weights of the passed judgment matrices to obtain the weights of transaction modalities, user behavior modalities, merchant modalities, and platform operation modalities. Use each weight coefficient as the importance allocation weight for the corresponding modal data stream in the digital economy platform business. Then, perform a weighted mapping on the feature vectors after cross-domain semantic alignment using the importance allocation weights of each modal data stream in the digital economy platform business. Multiply the aligned feature vector of each modality by the corresponding semantic weight, and expand the dimension of the weighted feature vector (extending the 64-dimensional feature vector to 128 dimensions through linear transformation, consistent with the dimension of the semantic latent space). Input the expanded weighted feature vector into the constructed semantic latent space to complete the feature mapping. The mapping of eigenvectors uses the mapped feature vectors of various modalities as feature vectors mapped to the semantic latent space. Finally, a self-attention mechanism is used to perform feature fusion with business semantic attention on all feature vectors mapped to the semantic latent space. The 128-dimensional feature vectors of the four categories of transaction, user behavior, merchant, and platform operation are concatenated into a 4×128 feature matrix according to their dimensions. This feature matrix is ​​used as the input of the self-attention mechanism. The input 4×128 feature matrix is ​​linearly transformed by three independent 128×128 linear transformation weight matrices (with bias terms set to 0), resulting in the query matrix Q, key matrix K, and value matrix V. The query matrix Q is multiplied by the transpose of the key matrix K, and the result of the multiplication is the square root of the feature dimension. The scale is adjusted, and the scaled result is normalized using the Softmax activation function to obtain a 4×4 basic attention weight matrix. Each element in this matrix represents the basic semantic attention of one type of feature vector to another type of feature vector. Combining the importance weights of each modality data stream in the digital economy platform business, the weight value of each row in the basic attention weight matrix is ​​multiplied by the importance weight of the corresponding modality to complete the business semantic correction of the basic attention weight matrix, resulting in a 4×4 attention weight matrix with business semantics. This attention weight matrix with business semantics is then multiplied by the value matrix V to obtain a 4×128 weighted feature matrix. The weighted feature matrix is ​​then summed row by row to obtain a 1×128 fused feature vector, which is used as the unified representation vector.

[0079] In step S3, a dynamic business knowledge graph of the digital economy platform is constructed. The unified representation vector is used as the incremental information of entity nodes in the dynamic business knowledge graph. The dynamic business knowledge graph is then updated and reasoned to output a context-aware vector containing rich semantic associations.

[0080] In this embodiment, the construction of a dynamic business knowledge graph for a digital economy platform can be achieved through the following steps:

[0081] Extract entity nodes from various business areas within the digital economy platform;

[0082] Define the semantic constraints and association weight assignment rules for each type of entity node;

[0083] A dynamic business knowledge graph of the digital economy platform is constructed based on the semantic constraints and association weight assignment rules of various types of entity nodes.

[0084] It should be noted that, in this application, the entity node refers to the core business object with independent business semantics that can be uniquely identified within each business area of ​​the digital economy platform; the semantic constraint condition refers to the rule that constrains the semantic consistency of entity nodes; the association weight assignment rule refers to the specific criteria used to determine the degree of business association between different entity nodes and to assign weight coefficients to the association relationship; and the dynamic business knowledge graph refers to the knowledge graph that synchronizes the business operation status of the digital economy platform in real time.

[0085] In specific implementation, firstly, the core business areas of the digital economy platform are transaction, user, merchant, and platform operation areas. Pre-defined entity keyword rules are established for each area (transaction area keywords include transaction number, transaction amount, transaction status, etc.; user area keywords include user ID, user name, registered mobile phone number, etc.; merchant area keywords include merchant ID, merchant name, business scope, etc.; platform operation area keywords include operation announcement ID, operation activity name, fault type, etc.). Candidate entities containing the pre-defined keywords are extracted using rule matching. Then, the candidate entities and corresponding business text are input into the BERT base model (input text length set to 128 characters, training iterations set to 50). Candidate entities with a semantic relevance higher than 0.8 are selected. The selected entities are deduplicated (using unique entity ID identification to remove duplicate entities), categorized by business area, and the deduplicated entities are designated as entity nodes within each business area of ​​the digital economy platform. Each entity node is assigned a unique entity ID and business area label. Secondly, semantic constraints and association weight assignment rules are defined for each type of entity node, and specific rules are set for different types of entity nodes. The constraints for user-type entity nodes are as follows: User ID is an 18-digit number combination; registered mobile phone number is an 11-digit valid mobile phone number; registration time format is YYYY-MM-DD. The constraints for merchant-type entity nodes are as follows: Merchant ID is a 12-digit alphanumeric combination; business scope is selected from a preset industry list; qualification certificates are complete and valid. The constraints for transaction-type entity nodes are as follows: Transaction order number is a 20-digit alphanumeric combination; transaction amount is greater than 0; transaction status is a preset enumeration value (pending payment, paid, canceled, completed). The constraints for platform operation-type entity nodes are as follows: The announcement ID is a 10-digit number combination, and the operational activity name cannot exceed 50 characters. Regular expressions are used to convert these constraints into verifiable code rules. The association weight assignment rule uses a combination of business association frequency and business contribution, setting the weight coefficient range to 0-1. The frequency of business interactions between different entity nodes within the past three months is statistically analyzed; the higher the interaction frequency, the higher the weight coefficient. Simultaneously, adjustments are made based on business contribution (the contribution coefficient for core business associations such as user-transaction and merchant-transaction is set to 1.2, while the contribution coefficient for non-core business associations such as user-operational announcements is set to 0).8) Calculate the association weight coefficient = (interaction frequency / highest interaction frequency) × contribution coefficient. Use this calculation method and weight range as the association weight assignment rule between entity nodes. Finally, using the Neo4j graph database as the graph storage medium, first import the extracted entity nodes and their corresponding attributes into the database. Then, perform semantic constraint verification on the imported entity nodes using preset regular expression validation rules, removing entity nodes that do not meet the semantic constraints and re-extracting and supplementing them. Next, sort out the business relationships between each entity node (user-consumption-transaction, transaction-ownership-merchant, merchant-participation-operation activities, etc.), calculate the weight coefficient of each relationship according to the association weight assignment rule, create the relationship between entity nodes in the Neo4j graph database and assign corresponding weight coefficients. Simultaneously, set up a graph dynamic update interface in the database to receive subsequent unified representation vectors as incremental information for entity nodes, completing the graph initialization construction. The initialized graph, with a dynamic update interface and conforming to semantic constraints and weight rules, serves as the dynamic business knowledge graph of the digital economy platform.

[0086] In this embodiment, the unified representation vector is used as incremental information for entity nodes in the dynamic business knowledge graph, thereby updating and reasoning about the dynamic business knowledge graph and outputting a context-aware vector containing rich semantic associations. This can be achieved through the following steps:

[0087] Incremental information layering is performed on entity nodes in the dynamic business knowledge graph using the unified representation vector.

[0088] The attribute features of entity nodes are updated based on the embedded incremental information, and the semantic rules of the dynamic business knowledge graph are used for association reasoning to obtain explicit semantic association relationships between entity nodes.

[0089] The implicit semantic relationships between entity nodes are determined based on the business association logic of the digital economy platform;

[0090] Cross-node global semantic fusion is performed on the updated entity node attribute features, explicit semantic relationships between entity nodes, and implicit semantic relationships between entity nodes to output a context-aware vector containing rich semantic relationships.

[0091] It should be noted that the explicit semantic associations described in this application represent intuitively identifiable business associations between entity nodes obtained through semantic rule reasoning of dynamic business knowledge graphs; the implicit semantic associations represent potential business associations between entity nodes obtained through mining business association logic of the digital economy platform; and the context-aware vector represents a feature vector that perceives the semantic associations of the entire business context of the digital economy platform.

[0092] In specific implementation, firstly, for the four types of entity nodes in the dynamic business knowledge graph—users, merchants, transactions, and platform operations—a graph attention network is used to achieve the hierarchical embedding of incremental information in a unified representation vector. The 1×128 unified representation vector is divided into three feature vectors according to the business dimension: entity attribute layer, business association layer, and temporal incremental layer. Two hidden layers are set for the graph attention network, each with a dimension of 128. The original feature vectors of the four types of entity nodes and the incremental feature vectors of the corresponding dimensions are used as inputs to the graph attention network. An attention mechanism is used to assign embedding weights to the incremental features of each layer (0 for the attribute layer). 5. The embedding operation of the three-layer incremental features is completed node by node (association layer 0.3, temporal layer 0.2). The entity node feature vector after embedding incremental information is used as the entity node feature vector. Next, the attribute features of the entity node are updated based on the embedded incremental information. The entity node feature vector after hierarchical embedding is weighted and fused with the original feature vector (fusion weight set to 0.7 for incremental features and 0.3 for original features). The fused feature vector is compressed to 128 dimensions and replaces the original attribute feature vector of the corresponding entity node in the dynamic business knowledge graph, completing the update of the entity node attribute features. Using the SPARQL query language combined with the semantic constraints and association weight assignment rules preset in the dynamic business knowledge graph, association reasoning query statements are written to traverse and reason through the updated graph, extracting the direct business relationships between entity nodes (e.g., user ID corresponding to transaction number, transaction number belonging to merchant ID, merchant ID participating in operational activities, etc.). Corresponding association weight coefficients are matched to the reasoned relationships, and the direct business relationships between entities with weight coefficients are used as explicit semantic relationships between entity nodes. Then, a graph neural network is used in conjunction with the business association logic of the digital economy platform. The implicit semantic relationships between entity nodes are mined. The updated dynamic business knowledge graph is used as input to a graph neural network. The graph neural network is set to have 3 propagation layers, 128 hidden layer dimensions, 50 training iterations, and a learning rate of 0.001. The graph neural network extracts potential feature relationships between entity nodes. Combined with the platform's preset business relationship logic (such as the relationship between user browsing behavior and merchant product recommendations, the relationship between overlapping customer groups of different merchants, and the potential impact relationship between operational activities and transaction data), the potential relationships extracted by the graph neural network are filtered, retaining those with a semantic relationship degree higher than 0.The potential relationships in section 7 are analyzed by calculating association weight coefficients for the filtered potential relationships (the calculation method is consistent with that of explicit relationships). These weighted potential business relationships between entities are then used as implicit semantic relationships between entity nodes. Finally, a cross-domain attention fusion mechanism is employed to perform cross-node global semantic fusion on the updated entity node attribute features, explicit semantic relationships between entity nodes, and implicit semantic relationships. First, the weight coefficients of explicit and implicit semantic relationships are converted into 128-dimensional association feature vectors. Then, the 128-dimensional attribute feature vectors, explicit 128-dimensional association feature vectors, and implicit 128-dimensional association feature vectors of all updated entity nodes are concatenated into a global feature matrix. This global feature matrix is ​​input into the cross-domain attention fusion mechanism, which assigns fusion weights to the features of different entities and different relationships. The weighted global feature matrix is ​​then summed and normalized dimensionally to obtain a 256-dimensional fusion feature vector. This 256-dimensional fusion feature vector serves as a context-aware vector containing rich association semantics.

[0093] In step S4, an analysis task description issued by the upper-layer business scenario is received, and a data view matching different business scenarios is constructed through the analysis task description and the context-aware vector.

[0094] It should be noted that the analysis task description mentioned in this application refers to the task instructions issued by the upper-layer business scenario according to business needs. The upper-layer business scenario refers to various business application scenarios built by the digital economy platform to meet actual operational needs, specifically including user operation, merchant analysis, transaction risk control, and platform activity optimization.

[0095] In practice, a task interaction gateway based on RPC+HTTP / HTTPS dual protocols is first built as a dedicated task interaction channel between upper-layer business scenarios and lower-layer data analysis modules. The gateway is configured with an independent task receiving port and a data transmission encryption mechanism (using SSL / TLS encryption). At the same time, a task cache queue (with a queue capacity of 1000 entries) is set up in the gateway to temporarily store the received analysis task descriptions to avoid task loss. The real-time monitoring function of the gateway is enabled to receive analysis task descriptions issued by upper-layer business scenarios (user operation, merchant analysis, transaction risk control, platform activity optimization, etc.) in real time and temporarily store the received task descriptions in the task cache queue as analysis task descriptions to be processed.

[0096] In this embodiment, constructing a data view matching different business scenarios using the analysis task description and the context-aware vector can be achieved through the following steps:

[0097] Fine-grained semantic decoupling is performed on the analysis task description to obtain semantic decoupling features;

[0098] The context-aware vector is reconstructed using the semantic decoupling features to obtain a set of context-aware feature vectors.

[0099] A data view matching different business scenarios is constructed based on the scenario-based feature vector set.

[0100] It should be noted that the semantic decoupling feature representation described in this application represents a set of features that characterize the core analytical needs of a business scenario; the scenario-based feature vector set represents a combination of feature vectors that adapt to the analytical needs of different business scenarios; and the data view represents a structured data carrier that fits the analytical needs of a business scenario.

[0101] In specific implementation, firstly, the jieba word segmentation tool is used to segment the description of the analysis task. Then, a digital economy platform business domain dictionary (containing core vocabulary from areas such as user operation, merchant analysis, transaction risk control, and platform operation) is loaded to complete the segmented words, removing stop words without business semantics. The processed word sequence is then input into a BiLSTM model for core semantic element extraction. The BiLSTM model is configured with two hidden layers, each with a dimension of 256. The training iterations are set to 50, and the learning rate is set to 0.001. Core semantic elements such as business scenario type, analysis dimension, data granularity, and core analysis indicators are extracted from the analysis task. Semantic elements are transformed into 256-dimensional feature vectors of the same dimension as the context-aware vector through linear transformation. This set of 256-dimensional feature vectors is then combined as semantic decoupling features. Next, the semantic decoupling features are used as attention masks to calculate the attention weights of each dimension of the semantic decoupling features on each dimension of the context-aware vector. Then, each dimension's attention weight is multiplied dimension-by-dimensionally with the corresponding dimension of the context-aware vector to obtain a 256-dimensional feature vector adapted to the analysis needs of a single scenario. For different business scenarios such as user operation, merchant analysis, transaction risk control, and platform operation of digital economy platforms, the above feature reconstruction operations are performed respectively, and the adapted feature vectors for each business scenario are combined. As a scenario-based feature vector set; finally, pre-set structured data view templates for various business scenarios of the digital economy platform, including user operation view templates containing core modules such as user profiles, behavior analysis, and retention rate indicators; merchant analysis view templates containing core modules such as merchant operating data, transaction trends, and customer characteristics; transaction risk control view templates containing core modules such as transaction flow, abnormal transaction detection, and risk level assessment; and platform operation view templates containing core modules such as operational activity effects, platform traffic, and transaction data statistics. Pre-set structured data fields and indicator dimensions for each template, and input the 256-dimensional feature vectors corresponding to each business scenario in the scenario-based feature vector set into a fully connected neural network, for... The network has two hidden layers, each with a dimension of 256. The output layer dimension is set according to the structured data field dimension of the corresponding view template. The number of training iterations is set to 50, and the learning rate is set to 0.001. The feature vectors are transformed into structured data indicator values ​​of the corresponding view template through a fully connected neural network. The output structured data indicator values ​​are dimensionally normalized and integrated with business associations. Invalid indicator data is removed and missing related business data is added. Dynamic update interfaces are configured for each structured data view to connect to dynamic business knowledge graphs. The integrated structured data is encapsulated according to a preset template, and the encapsulated structured data carrier is used as a data view to match different business scenarios.

[0102] In step S5, semantic analysis is performed on the business scenario requirements of the digital economy platform based on the data view, and a semantic analysis report of the business scenario requirements is generated.

[0103] Preferably, in this embodiment, semantic analysis is performed on the business scenario requirements of the digital economy platform based on the data view to generate a semantic analysis report of the business scenario requirements, for reference. Figure 3 As shown in the figure, this is a schematic diagram of the process of generating a semantic analysis report in some embodiments of this application. In this embodiment, generating a semantic analysis report can be achieved by the following steps:

[0104] In step S51, semantic information is extracted from the data view, and each type of semantic information is labeled with a requirement tag corresponding to the business scenario.

[0105] In step S52, bidirectional semantic alignment verification is performed on the annotated semantic information;

[0106] In step S53, the business scenario requirements of the digital economy platform are analyzed based on the bidirectional semantic alignment verification results.

[0107] In step S54, a semantic analysis report of business scenario requirements is generated based on the business parsing results.

[0108] It should be noted that the semantic information mentioned in this application refers to the structured semantic content related to business scenario requirements carried in the data view; the requirement tag refers to the structured tag used to identify the business scenario requirement type corresponding to the semantic information; the bidirectional semantic alignment verification refers to the process of bidirectionally verifying the semantic consistency between semantic information and requirement tags, and between semantic information and corresponding business scenarios; and the semantic analysis report refers to the semantic parsing document adapted to the business scenario requirements of the digital economy platform.

[0109] In practice, the BeautifulSoup parsing tool is first used to perform structured semantic extraction on data views for different business scenarios. For data views in scenarios such as user operations, merchant analysis, transaction risk control, and platform operations, structured semantic information of the corresponding core modules is extracted. Specifically, the user operations view extracts semantic information such as user profile parameters, behavioral feature data, and retention rate-related indicators; the merchant analysis view extracts semantic information such as merchant operating data, transaction trend parameters, and customer group characteristic information; the transaction risk control view extracts semantic information such as transaction flow data, anomaly detection parameters, and risk level indicators; and the platform operations view extracts semantic information such as... Semantic information such as campaign performance data, platform traffic parameters, and transaction statistics is extracted and then deduplicated and denoised to remove redundant information. A pre-defined tag library for digital economy platform business scenarios is also used, containing core requirement tags such as user retention improvement, merchant operation optimization, risk control accuracy improvement, and operational efficiency optimization. The KNN algorithm is used to match the processed semantic information with the requirement tags in the tag library. With K set to 5 and cosine similarity as the distance metric, requirement tags with a matching score higher than 0.8 are used as the corresponding semantic information labels, ensuring each type of semantic information is uniquely labeled with a corresponding business. The requirement labels for the scenario are used to combine the annotated semantic information with the requirement labels as the semantic dataset to be verified. Next, a Siamese neural network is used to perform bidirectional semantic alignment verification on the annotated semantic information. Each type of semantic information and its corresponding requirement label in the semantic dataset to be verified is transformed into a 256-dimensional feature vector (consistent with the dimension of the context-aware vector) through linear transformation. These two types of feature vectors are used as input to the Siamese neural network. The Siamese neural network has two hidden layers, each with a dimension of 256. The training iterations are set to 50, and the learning rate is set to 0.001. The Siamese neural network is then used to perform the verification. The cosine similarity calculation unit in the output layer of the ESE neural network first performs L2 normalization on the two types of 256-dimensional feature vectors input separately to eliminate the calculation bias caused by the scaling of feature vector dimensions. Then, it calculates the cosine value of the angle between the two normalized feature vectors. This cosine value is the semantic similarity between the two types of feature vectors. The semantic similarity value is controlled between 0 and 1. The closer the value is to 1, the higher the semantic consistency between the two types of feature vectors. The closer the value is to 0, the lower the semantic consistency. At the same time, it combines the preset semantic rules of various business scenarios of the digital economy platform to verify the semantic consistency between the semantic information and the corresponding business scenario, and retains semantic similarity values ​​higher than 0.7. Semantic information and corresponding requirement tags that are consistent with the semantics of the business scenario are collected. Inconsistent data is removed and re-annotated. The verified semantic information and requirement tags are used as the bidirectional semantic alignment verification result. Then, the bidirectional semantic alignment verification result is input into the LSTM model and combined with the business rule library of the digital economy platform for business parsing. Two hidden layers are set for the LSTM model, and the dimension of each hidden layer is 256. The number of training iterations is set to 50, and the learning rate is set to 0.001. The business rule library of each business area of ​​the digital economy platform (including business logic, requirement judgment standards, indicator interpretation rules, etc. in areas such as user operation, merchant analysis, transaction risk control, and platform operation) is loaded. The LSTM model is used to extract key information such as the core elements of business scenario requirements, requirement priority, core indicator requirements, and potential business pain points from the verification result. This information is then combined with the relationship between entity nodes in the dynamic business knowledge graph. To supplement the business logic supporting the business requirements, the key extracted information is structurally integrated to obtain business analysis results including a requirement overview, core demands, detailed indicators, and pain point analysis. Finally, a structured template for the semantic analysis report is pre-defined, containing five core modules: requirement overview, semantic verification instructions, business analysis details, core conclusions, and business optimization suggestions. Fixed data fields and expression standards are pre-defined for each module. The business analysis results are filled into the corresponding positions in the template according to the module, and the report content is structured and encapsulated using Markdown format. A dynamic update interface is configured for the report, connecting to the data views of various business scenarios to ensure that the report is updated synchronously when the data views are updated. Simultaneously, the report content undergoes format verification, removing non-standard expressions and logical inconsistencies, supplementing missing business-related information, and the encapsulated and approved structured report serves as the semantic analysis report for the business scenario requirements.

[0110] Therefore, this application demonstrates that semantic analysis of the business scenario requirements of the digital economy platform can be performed based on the data view, generating a semantic analysis report of the business scenario requirements. Firstly, by accessing the multi-source business data streams of the digital economy platform in real time, it ensures that the aggregated data comprehensively covers all types of business scenarios on the platform and is timely, solving the problems of incomplete or untimely data sources leading to incomplete fusion results and insufficient scenario adaptability in traditional data analysis techniques. Secondly, parallel modal encoding of the multi-source business data streams and mapping the feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector effectively breaks down semantic barriers between different types of data, achieving a leap from surface-level splicing to underlying semantic-level feature fusion of multi-modal data, solving the core pain point of traditional techniques' difficulty in achieving deep fusion of various types of data. Furthermore, a dynamic business knowledge graph of the digital economy platform is constructed, and the unified representation vector is used as incremental information for updating entity nodes. By reasoning and outputting context-aware vectors rich in semantic associations, the system can deeply mine the profound business relationships between multimodal data, giving the fused data clear business semantic attributes and inference value, thus further enhancing the depth and practicality of multimodal data semantic fusion. Furthermore, by receiving analysis task descriptions from upper-level business scenarios and constructing data views matching different business scenarios through analysis task descriptions and context-aware vectors, the system can achieve precise alignment between semantically fused data and specific business scenario requirements, breaking the predicament of data fusion being disconnected from business scenarios and improving the adaptability of data analysis to different business scenarios. Finally, by performing semantic analysis on business scenario requirements based on the data views and generating semantic analysis reports, the system can transform the fused deep semantic data into practical analysis results tailored to specific business scenario needs, effectively improving the practicality and relevance of data analysis results and fully meeting the multi-scenario, refined analysis needs of digital economy platforms.

[0111] In summary, the technical solution adopted in this application can achieve semantic-level deep fusion of multimodal business data of digital economy platforms, thereby adapting to the analysis needs of digital economy platforms in multiple scenarios.

[0112] Example 2: This application provides an intelligent data analysis system for digital economy platforms, referring to... Figure 4 As shown in the figure, this is a module structure diagram of an intelligent data analysis system for a digital economy platform according to this embodiment of the present application. The intelligent data analysis system includes:

[0113] The data stream receiving module 100 is used to access the multi-source business data stream of the digital economy platform in real time;

[0114] The feature vector mapping module 200 is used to perform parallel modal encoding on the multi-source business data stream and map the feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector.

[0115] The semantic context association module 300 is used to construct a dynamic business knowledge graph of the digital economy platform. It uses the unified representation vector as the incremental information of entity nodes in the dynamic business knowledge graph, and then updates and infers the dynamic business knowledge graph to output a context-aware vector containing rich semantic association.

[0116] The business scenario matching module 400 is used to receive the analysis task description issued by the upper-layer business scenario, and construct a data view matching different business scenarios through the analysis task description and the context-aware vector.

[0117] The analysis report generation module 500 is used to perform semantic analysis on the business scenario requirements of the digital economy platform based on the data view, and generate a semantic analysis report on the business scenario requirements.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0120] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A smart data analysis method for digital economy platforms, characterized in that, The intelligent data analysis method includes the following steps: The multi-source business data streams that are accessed in real time to the digital economy platform refer to various real-time data streams with different data formats and business attributes from different business systems during the operation of the digital economy platform. Parallel modal encoding is performed on the multi-source business data streams, and the feature vectors of different modalities are mapped to a shared semantic latent space to generate a unified representation vector; A dynamic business knowledge graph of a digital economy platform is constructed, and the unified representation vector is used as the incremental information of entity nodes in the dynamic business knowledge graph. The dynamic business knowledge graph is then updated and reasoned, and a context-aware vector containing rich semantic association is output. Receive the analysis task description issued by the upper-layer business scenario, and construct a data view matching different business scenarios through the analysis task description and the context-aware vector; Based on the data view, semantic analysis is performed on the business scenario requirements of the digital economy platform to generate a semantic analysis report on the business scenario requirements. Specifically, parallel modal coding of the multi-source service data stream includes: The multi-source business data stream is subjected to adaptive modal segmentation based on business characteristics to obtain different types of modal data streams; Heterogeneous preprocessing is performed on various modal data streams; Configure independent heterogeneous coding branches for each type of modal data stream and set up coding synchronization scheduling nodes; Synchronous heterogeneous encoding is performed on the preprocessed data streams of various modalities to output feature vectors that retain the core business characteristics of each modality. Specifically, mapping feature vectors from different modalities to a shared semantic latent space to generate a unified representation vector includes: Constructing a semantic implicit space for dynamic sharing of digital economy business; Perform cross-domain semantic alignment on feature vectors of different modalities; Determine the importance and assign weights of each modality of data flow in the business of the digital economy platform; By assigning weights based on the importance of each modal data stream in the digital economy platform's business, the feature vectors after cross-domain semantic alignment are mapped to the semantic latent space; All feature vectors mapped to the semantic latent space are subjected to feature fusion with business semantic attention to generate a unified representation vector adapted to digital economy business. Specifically, constructing a data view matching different business scenarios through the analysis task description and the context-aware vector includes: Fine-grained semantic decoupling is performed on the analysis task description to obtain semantic decoupling features; The context-aware vector is reconstructed using the semantic decoupling features to obtain a set of context-aware feature vectors. A data view matching different business scenarios is constructed based on the scenario-based feature vector set.

2. The intelligent data analysis method for digital economy platforms as described in claim 1, characterized in that, The construction of a dynamic business knowledge graph for a digital economy platform specifically includes: Extract entity nodes from various business areas within the digital economy platform; Define the semantic constraints and association weight assignment rules for each type of entity node; A dynamic business knowledge graph of the digital economy platform is constructed based on the semantic constraints and association weight assignment rules of various types of entity nodes.

3. The intelligent data analysis method for digital economy platforms as described in claim 1, characterized in that, Using the unified representation vector as incremental information for entity nodes in the dynamic business knowledge graph, and then updating and reasoning the dynamic business knowledge graph to output a context-aware vector containing rich semantic associations, specifically includes: Incremental information layering is performed on entity nodes in the dynamic business knowledge graph using the unified representation vector. The attribute features of entity nodes are updated based on the embedded incremental information, and the semantic rules of the dynamic business knowledge graph are used for association reasoning to obtain explicit semantic association relationships between entity nodes. The implicit semantic relationships between entity nodes are determined based on the business association logic of the digital economy platform; Cross-node global semantic fusion is performed on the updated entity node attribute features, explicit semantic relationships between entity nodes, and implicit semantic relationships between entity nodes to output a context-aware vector containing rich semantic relationships.

4. The intelligent data analysis method for digital economy platforms as described in claim 1, characterized in that, Based on the data view, semantic analysis is performed on the business scenario requirements of the digital economy platform to generate a semantic analysis report on business scenario requirements, specifically including: Extract semantic information from the data view and label each type of semantic information with the corresponding business scenario requirement tags; Perform bidirectional semantic alignment verification on the annotated semantic information; Business analysis is performed on the business scenario requirements of the digital economy platform based on the bidirectional semantic alignment verification results. Generate a semantic analysis report of business scenario requirements based on the business parsing results.

5. The intelligent data analysis method for digital economy platforms as described in claim 1, characterized in that, The unified representation vector represents the semantic representation vector of multimodal business data on the digital economy platform.

6. The intelligent data analysis method for digital economy platforms as described in claim 1, characterized in that, The context-aware vector represents the feature vector of semantic association of the full-domain business context of the digital economy platform.

7. An intelligent data analysis system for a digital economy platform, used to execute the intelligent data analysis method for a digital economy platform as described in any one of claims 1 to 6, characterized in that, The intelligent data analysis system includes: The data stream receiving module is used to access multi-source business data streams from the digital economy platform in real time. The feature vector mapping module is used to perform parallel modal encoding on the multi-source business data stream and map the feature vectors of different modalities to a shared semantic latent space to generate a unified representation vector. The semantic context association module is used to construct a dynamic business knowledge graph of the digital economy platform. It uses the unified representation vector as incremental information of entity nodes in the dynamic business knowledge graph, and then updates and infers the dynamic business knowledge graph to output a context-aware vector containing rich semantic association. The business scenario matching module is used to receive the analysis task description issued by the upper-layer business scenario, and construct a data view matching different business scenarios through the analysis task description and the context-aware vector. The analysis report generation module is used to perform semantic analysis on the business scenario requirements of the digital economy platform based on the data view, and generate a semantic analysis report on the business scenario requirements.

Images