Cross-platform electronic invoice data integration and processing method and system

Abstract

The embodiment of the present application provides a cross-platform electronic invoice data integration and processing method and system, which first obtains sample data including cross-platform electronic invoice integration data, order path data and focus label data. Then, the three unit modules of the cross-platform focus mining network are used to perform feature selection on the electronic invoice semantic vector and the order path graph vector, respectively, and generate corresponding prediction data. Next, a differential fusion operation is performed on the vectors after feature selection to generate a graph knowledge structure encoding result as a focus logical feature. Finally, the network prediction error is determined based on the prediction data and the focus label data, and the cross-platform focus mining network is trained and optimized until the network converges to generate a target cross-platform focus mining network, which can effectively improve the accuracy and efficiency of cross-platform electronic invoice data processing, deeply mine data features, and provide enterprises with more comprehensive and accurate electronic invoice data analysis results.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a cross-platform electronic invoice data integration and processing method and system. Background Art

[0002] With the rapid development of e-commerce, electronic invoices have become an indispensable part of corporate transaction settlement. However, due to the differences in data formats, standards and processing methods between different platforms, the integration and processing of cross-platform electronic invoice data faces huge challenges. Traditional electronic invoice data processing methods are often limited to a single platform, making it difficult to achieve cross-platform data integration and analysis, which greatly limits the efficiency of enterprises in using electronic invoice data and value mining. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the embodiments of the present application is to provide a cross-platform electronic invoice data integration and processing method and system.

[0004] According to one aspect of the present application, a cross-platform electronic invoice data integration and processing method is provided, the method comprising:

[0005] Obtain sample cross-platform electronic invoice integration data, each of the sample cross-platform electronic invoice integration data includes cross-platform electronic invoice integration data, order path data of the cross-platform electronic invoice integration data, and cross-platform focus tag data;

[0006] Using the first unit module of the cross-platform focus mining network, feature selection is performed on the basic electronic invoice semantic vector of the cross-platform electronic invoice integration data to generate a translated electronic invoice semantic vector, and first cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the translated electronic invoice semantic vector;

[0007] Using the second unit module of the cross-platform focus mining network, feature selection is performed on the basic order path graph vector of the order path data to generate a translated order path graph vector, and second cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the translated order path graph vector;

[0008] Using the third unit module of the cross-platform focus mining network, a differential fusion operation is performed on the translated electronic invoice semantic vector and the translated order path graph vector to generate differential fusion features, and the differential fusion features are encoded in a graph knowledge structure to generate a graph knowledge structure encoding result, and third cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the graph knowledge structure encoding result; the graph knowledge structure encoding result is a focus logic feature of the sample cross-platform electronic invoice integration data;

[0009] Based on the first cross-platform focus point prediction data, the second cross-platform focus point prediction data, the third cross-platform focus point prediction data and the cross-platform focus point label data, a network prediction error is determined, and the cross-platform focus point mining network is trained and optimized based on the network prediction error until the cross-platform focus point mining network meets the network convergence requirements, thereby generating a target cross-platform focus point mining network.

[0010] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various optional implementations of the above three aspects.

[0011] In the technical solutions provided in some embodiments of the present application, the embodiments of the present application perform deep feature selection and fusion operations on cross-platform electronic invoice integration data, order path data and focus label data through three unit modules of the cross-platform focus mining network, generate a graph knowledge structure encoding result with differential fusion features, and then accurately predict cross-platform focus points, which can effectively integrate and process cross-platform electronic invoice data, improve the accuracy and efficiency of data processing, and generate a target cross-platform focus mining network that meets the network convergence requirements through training optimization based on network prediction errors, thereby further improving the intelligence level of cross-platform electronic invoice data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be extracted in combination with these drawings without creative work.

[0013] Figure 1A flowchart of a cross-platform electronic invoice data integration and processing method provided in an embodiment of the present application;

[0014] Figure 2 This is a schematic block diagram of the architecture of the cross-platform electronic invoice data integration and processing system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The following description is intended to enable one of ordinary skill in the art to implement and combine the present application, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present application may be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0016] Figure 1 This is a flow chart of a cross-platform electronic invoice data integration and processing method provided by an embodiment of the present application. The cross-platform electronic invoice data integration and processing method is introduced in detail below.

[0017] Step S110, obtaining sample cross-platform electronic invoice integration data, each of the sample cross-platform electronic invoice integration data includes cross-platform electronic invoice integration data, order path data of the cross-platform electronic invoice integration data, and cross-platform focus tag data.

[0018] In detail, the sample cross-platform electronic invoice integration data refers to electronic invoice data collected from multiple different e-commerce platforms, which are integrated together for analysis. Each sample cross-platform electronic invoice integration data contains three parts: cross-platform electronic invoice integration data (i.e. detailed information of the invoice, such as product name, price, purchase time, etc.), order path data of cross-platform electronic invoice integration data (recording the consumer's behavior path from browsing to purchasing), and cross-platform focus label data (obtained through manual annotation or preliminary analysis, indicating hot data that customers may be interested in in this consumption scenario).

[0019] For example, in this embodiment, sample cross-platform electronic invoice integration data can be obtained from a database or other data source. For example, a large e-commerce service provider owns multiple e-commerce platforms, including platform A that specializes in selling electronic products, platform B that sells fashion clothing, and platform C that sells household goods. The electronic invoice data of these platforms constitute the source of cross-platform electronic invoice integration data.

[0020] For platform A, the electronic invoice integration data obtained by the server may contain invoice information of consumers purchasing electronic products such as mobile phones and computers, such as product name, price, purchase time, etc. The invoice shows that a consumer purchased a laptop worth 5,000 yuan on May 1, 2023, and also contains the seller's information, such as the store name, location, etc. The order path data records in detail the entire process of the consumer from browsing products, adding to shopping carts, placing orders, paying to finally confirming receipt. For example, the consumer first searched for "thin and light laptops" on platform A, then compared several products, spent 30 minutes to add a certain computer to the shopping cart, and completed the payment process after another 10 minutes. The cross-platform focus label data may be previously manually labeled or obtained through preliminary analysis. For example, for the purchase of this electronic product, the focus label data may include "cost-effectiveness", "brand awareness", "after-sales service", etc. These labels represent hot data that customers may be interested in in this consumption scenario.

[0021] Similarly, for fashion apparel sales on platform B, the electronic invoice integration data includes invoice information for consumers who purchase clothing, such as the purchase details of a dress worth 300 yuan. The order path data records the process of consumers browsing new clothing, checking sizes, color selection, and placing orders. For example, consumers first check the popular dress styles of the season, and then select their favorite white and small sizes before placing an order. Cross-platform focus tag data may include "style popularity", "material quality", "color matching", etc.

[0022] The same is true for the home furnishings sales of platform C. The electronic invoice integration data contains the invoice information of consumers who purchased sofas, the order path data records the decision-making process of purchasing sofas, and the cross-platform focus tag data may include "comfort", "material environmental protection", "matching with home style", etc. In this way, the server obtains rich sample cross-platform electronic invoice integration data, which provides a basis for subsequent mining work.

[0023] Step S120, using the first unit module of the cross-platform focus mining network, perform feature selection on the basic electronic invoice semantic vector of the cross-platform electronic invoice integration data, generate a translated electronic invoice semantic vector, and generate the first cross-platform focus prediction data for the sample cross-platform electronic invoice integration data based on the translated electronic invoice semantic vector.

[0024] In detail, the first unit module of the cross-platform focus mining network is a component of the cross-platform focus mining network, which is responsible for processing electronic invoice data. First, the basic electronic invoice semantic vector (i.e., the numerical representation of invoice data) is extracted, and then the vector is feature selected to generate the translated electronic invoice semantic vector (i.e., the optimized and screened vector), and finally the first cross-platform focus prediction data (i.e., the predicted hot data that customers may pay attention to) is generated based on the vector.

[0025] The basic electronic invoice semantic vector refers to a vector formed by digitizing each field in the electronic invoice data (such as product price, brand, purchase time, etc.), which contains all the information of the electronic invoice data.

[0026] For example, in this embodiment, the first unit module of the cross-platform focus mining network can be used for processing. For example, the initialization electronic invoice semantic extraction unit in the first unit module starts working.

[0027] Taking the laptop sales data of platform A as an example, the server will decompose the cross-platform electronic invoice integration data obtained, and will treat the data related to each electronic invoice as an independent unit to form an electronic invoice data set. For the invoice data of one of the laptops, the server extracts the basic electronic invoice semantic vector. This vector may contain information of multiple dimensions. For example, the dimension corresponding to the product price may be expressed as [5000, 0, 0] (here it is assumed that the price range is expressed, 5000 yuan is in a certain price range, and other ranges are 0), and the dimension corresponding to the product brand may be [1, 0, 0] (assuming that there are three types of brands, here it is the first brand), etc. After extracting all the electronic invoice data in the entire electronic invoice data set, a basic electronic invoice semantic vector set is generated.

[0028] Then the electronic invoice semantic translation unit performs feature domain mapping on this basic electronic invoice semantic vector set. For example, the price representation is mapped from a simple numerical range to a more complex feature domain related to market price fluctuations. The original [5000, 0, 0] may be mapped to [0.8, 0.1, 0.1] (indicating information such as relative position under current market price fluctuations). Then feature selection may be performed, which may remove some dimensions that are not important in this scenario, such as the dimensions corresponding to some redundant numbering information in the invoice. After such operations, the translated electronic invoice semantic vector is generated.

[0029] Based on the translated electronic invoice semantic vector, the server generates the first cross-platform focus prediction data for the sample cross-platform electronic invoice integration data. For example, if the dimension value related to brand awareness in the translated vector is high, then the first cross-platform focus prediction data may tend to indicate that "brand awareness" is an important focus, and may predict that in the entire cross-platform consumer market, brand awareness will be a hot data that customers pay more attention to when purchasing electronic products.

[0030] Step S130, using the second unit module of the cross-platform focus mining network, perform feature selection on the basic order path graph vector of the order path data, generate a translated order path graph vector, and generate second cross-platform focus prediction data for the sample cross-platform electronic invoice integration data based on the translated order path graph vector.

[0031] In detail, the basic order path graph vector refers to the vector formed by digitizing each link in the order path data (such as browsing, adding to shopping cart, payment, etc.) and the relationship between them. This vector contains all the information of the order path data. The translated order path graph vector refers to the order path graph vector processed by the second unit module. This vector removes irrelevant or redundant dimensions and only retains features related to customer concerns.

[0032] For example, in this embodiment, the second unit module of the cross-platform focus mining network is used to process the order path data. Taking the notebook purchase order path data of platform A as an example, the initialization graph autoencoder starts to extract the basic order path graph vector of the order path data.

[0033] Assume that in the order path graph vector, the nodes represent the various links in the order process, such as browsing, adding to the shopping cart, and paying, and the edges represent the transfer relationship between the links. For the purchase of this laptop, the edge from browsing to adding to the shopping cart may have a weight, indicating the probability or importance of this transfer. The server accurately extracts these basic order path graph vectors by initializing the graph autoencoder.

[0034] Then the order path graph translation unit performs feature selection on the basic order path graph vector. For example, it may be found that some intermediate operations from adding to the shopping cart to payment are not very critical for analyzing customer concerns, so the vector dimensions corresponding to these operations will be removed. After such processing, a translated order path graph vector is generated.

[0035] Based on the translated order path graph vector, the server generates the second cross-platform focus prediction data for the sample cross-platform electronic invoice integration data. For example, if the transfer weight from browsing to adding to the shopping cart in the translated order path graph vector is very high, it may mean that the customer's decision-making process at this stage is very critical, then the second cross-platform focus prediction data may predict that "preliminary product screening" is an important focus, that is, the factors involved in the process of customers browsing and initially choosing to add products to the shopping cart, such as the appearance and basic functions of the product, are hot data that customers in the cross-platform consumer market pay attention to.

[0036] Step S140, using the third unit module of the cross-platform focus mining network, performing a differential fusion operation on the translated electronic invoice semantic vector and the translated order path graph vector to generate differential fusion features, performing graph knowledge structure encoding on the differential fusion features to generate graph knowledge structure encoding results, and generating third cross-platform focus prediction data for the sample cross-platform electronic invoice integration data based on the graph knowledge structure encoding results. The graph knowledge structure encoding results are the focus logic features of the sample cross-platform electronic invoice integration data.

[0037] In detail, the difference fusion feature refers to the feature obtained by fusing the translated electronic invoice semantic vector and the translated order path graph vector, which integrates the information of the two vectors and can more comprehensively reflect the customer's concerns. The graph knowledge structure encoding result refers to the result after converting the difference fusion feature into the data form of the graph structure, which can more intuitively show the logical relationship and structural characteristics between the customer's concerns.

[0038] For example, in this embodiment, the server uses the third unit module of the cross-platform focus mining network to operate. Still taking the notebook computer sales data of platform A as an example, first analyze the position and importance of the semantic concept represented by each element in the translated electronic invoice semantic vector in the entire electronic invoice semantic system, and the degree of association between each element in the translated order path graph vector and each link of the order process.

[0039] For the translated e-invoice semantic vector, if one of the elements represents the processor performance of a laptop, in the entire e-invoice semantic system, processor performance is a key factor affecting product performance and has a high position importance. For the translated order path graph vector, for example, one of the elements represents whether installment payment is selected in the payment stage, which is closely related to the payment stage in the order process.

[0040] Then, based on these analysis results, the translated electronic invoice semantic vector and the translated order path map vector are decomposed. For the translated electronic invoice semantic vector, the elements representing the associated semantic concepts or related in business logic are combined, such as the processor performance, memory size, hard disk capacity and other elements of the laptop are combined into the first sub-vector sequence, because these are all related to the performance of the computer. For the translated order path map vector, the relevant elements are combined according to the different stages of the order process, such as combining the elements of the browsing stage and the elements of the add to cart stage respectively to generate the second sub-vector sequence. And the two sub-vector sequences are rearranged to make them logically adaptable, such as arranging them according to the order of purchase decisions.

[0041] Then, the target sub-vector combinations with associated logical meanings in the first sub-vector sequence and the second sub-vector sequence are determined and aligned in position. For example, the computer performance-related elements in the first sub-vector sequence are aligned with the elements in the product comparison and selection stage in the second sub-vector sequence, because at this stage customers will consider the performance of the computer.

[0042] Then, based on the prior knowledge of business logic, the importance of the features in the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment is evaluated. For example, for the purchase of laptops, based on previous sales data and market research, it is known that the processor performance is more important than the appearance color. Then, based on this feature importance evaluation result, the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment are pre-fused and adjusted.

[0043] Then, the corresponding sub-vectors in the pre-fused and adjusted translated electronic invoice semantic vector and the translated order path graph vector are fused according to the determined weights to obtain fused sub-vectors, and all fused sub-vectors are combined to generate differential fusion features.

[0044] Classify the elements in the differential fusion features, and preliminarily build a graph structure based on the element classification results. For example, elements related to product performance are taken as a category to form a node in the graph structure. The connection relationship between nodes is determined according to business logic. For example, a product with good performance may lead customers to enter the payment link faster, so a connection is established between these two nodes. At the same time, attributes are assigned to each node, such as the numerical information and category identification information of the element in the differential fusion features.

[0045] Check whether there are redundant connections or connections that should exist but are not established in the nodes in the initially constructed graph structure, delete the detected redundant connections, supplement the connections that should exist, and then optimize the properties of the nodes. For example, if it is found that the connection between a node related to the appearance color and the payment link is redundant, delete this connection; if it is found that there should be a connection between product performance and customer satisfaction but it is not, supplement this connection and optimize the properties of the relevant nodes.

[0046] Mining knowledge structure data in the optimized and adjusted graph structure, for example, discovering a substructure consisting of product performance, customer satisfaction, and brand awareness, which represents a business logic unit, and discovering the path pattern from product browsing to adding to the shopping cart to payment and finally to customer evaluation. Classifying and sorting the mined knowledge structure data, determining the coding rules and scope for each category of knowledge structure data, and then encoding the knowledge structure data in the graph structure according to the determined coding rules and scope. For example, for the knowledge structure data of product performance, a coding value is assigned according to the category to which it belongs and the position in the category, and a graph knowledge structure coding result is generated. This result is the logical feature of the focus of the sample cross-platform electronic invoice integration data. Based on this graph knowledge structure coding result, the server generates the third cross-platform focus prediction data for the sample cross-platform electronic invoice integration data. For example, if the proportion of coding related to product performance in the graph knowledge structure coding result is large, then the third cross-platform focus prediction data may predict that product performance is an important cross-platform focus.

[0047] Step S150, determining a network prediction error based on the first cross-platform focus point prediction data, the second cross-platform focus point prediction data, the third cross-platform focus point prediction data and the cross-platform focus point label data, training and optimizing the cross-platform focus point mining network based on the network prediction error, until the cross-platform focus point mining network meets the network convergence requirements, and generating a target cross-platform focus point mining network.

[0048] In this embodiment, the server determines the network prediction error based on the first cross-platform focus point prediction data, the second cross-platform focus point prediction data, the third cross-platform focus point prediction data and the cross-platform focus point label data obtained previously.

[0049] For example, for the sample data of laptop computer sales on platform A, the first cross-platform focus prediction data predicts that brand awareness is an important focus, the second cross-platform focus prediction data predicts that product preliminary screening is an important focus, and the third cross-platform focus prediction data predicts that product performance is an important focus, while the actual cross-platform focus label data shows that brand awareness, cost-effectiveness and after-sales service are important concerns.

[0050] Based on the first cross-platform focus prediction data and the cross-platform focus label data, a first prediction error is determined. If the predicted probability of brand awareness in the first cross-platform focus prediction data is 0.8, and the actual weight of brand awareness in the label data is 0.6, then the first prediction error is a measure of the difference between the two, such as an absolute difference of 0.2.

[0051] Likewise, a second prediction error is determined based on the second cross-platform focus point prediction data and the cross-platform focus point label data, and a third prediction error is determined based on the third cross-platform focus point prediction data and the cross-platform focus point label data.

[0052] Then, these three prediction errors are weighted to generate the network prediction error. Assuming that the weight of the first prediction error is 0.3, the weight of the second prediction error is 0.3, and the weight of the third prediction error is 0.4, the network prediction error is calculated according to the weighted formula.

[0053] Based on this network prediction error, the cross-platform focus mining network is trained and optimized. If the network prediction error is large, it means that the network's prediction ability is not accurate enough. The server will adjust the parameters in the network, such as adjusting the feature selection criteria in the first unit module, or adjusting the weights of the fusion operation in the third unit module. This process is repeated until the cross-platform focus mining network meets the network convergence requirements, that is, the network prediction error is reduced to an acceptable range, and the target cross-platform focus mining network is generated. This target cross-platform focus mining network can be used for subsequent accurate mining and analysis of other cross-platform electronic invoice integration data to accurately predict the hot data that customers in the cross-platform consumer market are interested in.

[0054] Based on the above steps, the embodiment of the present application performs deep feature selection and fusion operations on the cross-platform electronic invoice integration data, order path data and focus label data through the three unit modules of the cross-platform focus mining network, generates a graph knowledge structure encoding result with differential fusion features, and then accurately predicts cross-platform focus points, which can effectively integrate and process cross-platform electronic invoice data, improve the accuracy and efficiency of data processing, and generate a target cross-platform focus mining network that meets the network convergence requirements through training optimization based on network prediction errors, further improving the intelligence level of cross-platform electronic invoice data processing.

[0055] In a possible implementation, the first unit module includes an initialization electronic invoice semantic extraction unit and an electronic invoice semantic translation unit. Step S120 includes:

[0056] Step S121, using the initialized electronic invoice semantic extraction unit, decompose the cross-platform electronic invoice integration data to generate an electronic invoice data set, extract the basic electronic invoice semantic vector of each electronic invoice data in the electronic invoice data set, and generate a basic electronic invoice semantic vector set.

[0057] Step S122, using the electronic invoice semantic translation unit, performing feature domain mapping on the basic electronic invoice semantic vector set to generate a feature domain mapping result, performing feature selection on the feature domain mapping result to generate a translated electronic invoice semantic vector.

[0058] In this embodiment, the server uses the first unit module of the cross-platform focus mining network to process the cross-platform electronic invoice integration data. Taking the electronic invoice integration data of the laptop computer sold by platform A as an example, the electronic invoice semantic extraction unit is first initialized to start working. The server transmits the electronic invoice integration data about the laptop computer sales in platform A to the initialized electronic invoice semantic extraction unit, and the initialized electronic invoice semantic unit will decompose the data. For example, an electronic invoice contains information such as the time, price, model, and sales store of the laptop computer purchased. It will split this information into individual electronic invoice data, and these electronic invoice data together constitute an electronic invoice data set. Then, for each electronic invoice data in this electronic invoice data set, the basic electronic invoice semantic vector is extracted. For example, for the electronic invoice data of the price of the laptop computer, its basic electronic invoice semantic vector may contain information such as the price value, the price range, and the comparison relationship with the market average price; for the model of the laptop computer, its semantic vector may contain information such as the model series and the configuration level. By combining the semantic vectors of all these individual electronic invoice data, a basic electronic invoice semantic vector set is generated.

[0059] Then the electronic invoice semantic translation unit operates on the basic electronic invoice semantic vector set. For the elements in the basic electronic invoice semantic vector set, the electronic invoice semantic translation unit performs feature domain mapping. For example, for the element of price, the original simple price value may be mapped to the feature domain related to market price fluctuations, cost performance, etc. The vector that originally only represents the specific price will be converted into a new vector form containing information such as price competitiveness and price fluctuation trend, which forms the feature domain mapping result. Then, feature selection is performed on this basis. Since it is a hot data mining for customers in the cross-platform consumer market, features related to customer decisions may be selected, such as price competitiveness, model configuration level, etc., while some features that are not closely related to customer interests are removed, such as some internal numbers in the invoice. After such operations, a translated electronic invoice semantic vector is finally generated. This translated electronic invoice semantic vector can better reflect the electronic invoice information related to customer concerns, and provide a basis for the subsequent generation of cross-platform concern prediction data.

[0060] In a possible implementation, the second unit module includes an initialization graph autoencoder and an order path graph translation unit. Step S130 includes:

[0061] Step S131, using the initialized graph autoencoder, extracting the basic order path graph vector of the order path data.

[0062] Step S132, utilizing the order path graph translation unit to perform feature selection on the basic order path graph vector to generate a translated order path graph vector.

[0063] In this embodiment, the server uses the second unit module of the cross-platform focus mining network to process the order path data. Taking the order path data of the laptop computer on platform A as an example, the initialization graph autoencoder takes the lead. The server provides the order path data of the entire process of the laptop computer in platform A from the customer placing an order to the final receipt to the initialization graph autoencoder. The order path data contains information about each link from the customer browsing the laptop page to adding to the shopping cart, selecting the payment method, the merchant processing the order, shipping, logistics transportation, and the customer confirming the receipt. The initialization graph autoencoder will analyze these links in detail, and represent each link and its association in the form of a vector, thereby extracting the basic order path graph vector of the order path data. For example, for the browsing link, the basic order path graph vector may include information such as the browsing time and the number of product pages browsed; for the adding to the shopping cart link, it may include information such as the time of adding to the shopping cart, whether there are preferential activities that affect the decision to add to the shopping cart, and the vector representations of these links are combined to form the basic order path graph vector.

[0064] Then the order path graph translation unit performs feature selection on the basic order path graph vector. From the perspective of mining customer interest hotspots in the cross-platform consumer market, the order path graph translation unit will focus on selecting features related to customer decisions and interests. For example, in the entire order path, features such as the number of times a customer compares different laptop models before adding them to the shopping cart and whether they prefer a specific payment discount when choosing a payment method may be very important for understanding customer interests, while some features such as internal operation records that are not closely related to customer interests will be ignored. Through such a screening process, the order path graph translation unit generates a translated order path graph vector, which focuses more on order path features related to customer interest hotspots, laying the foundation for the subsequent generation of the second cross-platform focus prediction data for the sample cross-platform electronic invoice integration data.

[0065] In a possible implementation, step S140 includes:

[0066] Step S141, analyze the position and importance of the semantic concept represented by each element in the translated electronic invoice semantic vector in the entire electronic invoice semantic system, as well as the degree of association between each element in the translated order path graph vector and each link of the order process, and generate a first feature analysis result of the translated electronic invoice semantic vector and a second feature analysis result of the translated order path graph vector.

[0067] Step S142, according to the first feature analysis result and the second feature analysis result, the translated electronic invoice semantic vector and the translated order path graph vector are decomposed, for the translated electronic invoice semantic vector, the elements representing related semantic concepts or related in business logic are combined to generate a first sub-vector sequence, for the translated order path graph vector, the related elements are combined according to different stages of the order process to generate a second sub-vector sequence, and the first sub-vector sequence and the second sub-vector sequence are rearranged so that the first sub-vector sequence and the second sub-vector sequence are logically adapted.

[0068] Step S143, determining a target sub-vector combination having associated logical meaning in the first sub-vector sequence and the second sub-vector sequence, and aligning the target sub-vector combination in position to generate a translated electronic invoice semantic vector and a translated order path graph vector after feature alignment.

[0069] Step S144, based on the prior knowledge of business logic, the importance of the features in the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment is evaluated to generate corresponding feature importance evaluation results, and the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment are pre-fused and adjusted according to the feature importance evaluation results to generate the translated electronic invoice semantic vector and the translated order path graph vector after pre-fusion and adjustment.

[0070] Step S145, the corresponding sub-vectors in the pre-fusion adjusted translated electronic invoice semantic vector and the translated order path graph vector are fused according to the determined weights to obtain fused sub-vectors, and all fused sub-vectors are combined to generate differential fusion features.

[0071] Step S146, classify the elements in the differential fusion feature, and preliminarily construct a graph structure based on the element classification results, wherein the elements in each category are used as nodes in the graph structure, the connection relationship between the nodes is determined according to the business logic, and attributes are assigned to each node, and the attributes include the numerical information and category identification information of the element in the differential fusion feature. The preliminarily constructed graph structure is used to reflect the data relationship in the differential fusion feature.

[0072] Step S147, detect whether there are redundant connections or connections that should exist but are not established among the nodes in the initially constructed graph structure, delete the detected redundant connections accordingly, supplement the connections that should exist accordingly, and then optimize the attributes of the nodes to generate an optimized and adjusted graph structure.

[0073] Step S148, mining knowledge structure data in the optimized and adjusted graph structure, the knowledge structure data including substructures and path patterns in the graph structure, the substructure being a structure composed of a group of interrelated nodes, representing a business logic unit, the path pattern being a connection path between nodes in the knowledge structure, reflecting a process pattern in the business process.

[0074] Step S149, classify and organize the mined knowledge structure data, determine the coding rules and scope for each category of knowledge structure data, and encode the knowledge structure data in the graph structure according to the determined coding rules and scope. For each knowledge structure data, a coding value is assigned according to the category it belongs to and its position in the category, and a graph knowledge structure coding result is generated.

[0075] Continuing with the above scenario as an example, the third unit module of the cross-platform attention mining network is used to process the translated electronic invoice semantic vector and the translated order path map vector. Taking the relevant data of the laptop computer on platform A as an example, we first analyze the position and importance of the semantic concept represented by each element in the translated electronic invoice semantic vector in the entire electronic invoice semantic system, as well as the degree of association between each element in the translated order path map vector and each link of the order process.

[0076] For the translated electronic invoice semantic vector, if one of the elements is the processor performance of the laptop, in the entire electronic invoice semantic system, the processor performance is the key factor to measure the performance of the laptop, which directly affects the value of the product and the customer's purchase decision, so its position is very important. If it is the appearance color element of the laptop, its importance in the overall semantic system may be slightly lower than the processor performance, because it has less impact on the computer performance, but it still has a certain impact on some customers who pursue personalized appearance. For the translated order path graph vector, for example, one of the elements is the number of times the customer compares different laptop models before placing an order. This element is closely related to the order decision link in the order process. The more comparisons, the more cautious the customer is when placing an order and the higher the requirements for the product. If it is the transportation origin element in the logistics transportation link, although it is related to the order process, the direct correlation with the customer's order decision is relatively weak. Through such analysis, the first feature analysis result of the translated electronic invoice semantic vector and the second feature analysis result of the translated order path graph vector are generated.

[0077] Then, according to the first feature analysis result and the second feature analysis result, the translated electronic invoice semantic vector and the translated order path map vector are decomposed. For the translated electronic invoice semantic vector, the elements representing the associated semantic concepts or the elements associated in the business logic are combined to generate the first sub-vector sequence. For example, the elements such as the processor performance, memory size, and hard disk capacity of the laptop are combined together, because these elements are directly related to the performance of the computer, and the performance will affect the customer's overall evaluation and purchase intention of the product. For the translated order path map vector, the relevant elements are combined according to the different stages of the order process to generate the second sub-vector sequence. For example, the browsing time when the customer browses the laptop page and the number of product pages browsed are combined together, because these elements are related to the customer's behavior in the browsing stage; then the order time in the ordering link, whether to use coupons and other elements are combined together. Then the first sub-vector sequence and the second sub-vector sequence are rearranged to make them logically adaptable. For example, according to the order of customer purchase decisions, first browse and compare products, then place an order, pay, and finally receive the goods, the elements in the first sub-vector sequence related to browsing are arranged in front, the elements in the second sub-vector sequence related to ordering are arranged in the middle, and so on.

[0078] Then, the target sub-vector combination with associated logical meaning in the first sub-vector sequence and the second sub-vector sequence is determined, and the target sub-vector combination is aligned in position to generate the translated electronic invoice semantic vector and the translated order path map vector after feature alignment. For example, the computer performance-related elements in the first sub-vector sequence are aligned with the elements in the second sub-vector sequence that customers consider performance factors when comparing products, because customers will focus on computer performance during the product comparison stage. If there are elements representing prices in the first sub-vector sequence, they will also be aligned with the elements in the second sub-vector sequence that price affects order decisions during the ordering process, because price is an important consideration for customers when placing orders.

[0079] Based on the prior knowledge of business logic, the features in the translated electronic invoice semantic vector and the translated order path map vector after feature alignment are evaluated for importance, and the corresponding feature importance evaluation results are generated. In the business logic of laptop sales, processor performance and price are usually factors that customers are very concerned about, so the importance evaluation results of these two features will be relatively high. For some features such as computer packaging color, although they are also part of the product, they have a relatively small impact on customer purchase decisions, so the importance evaluation results will be lower. According to this feature importance evaluation result, the translated electronic invoice semantic vector and the translated order path map vector after feature alignment are pre-fused and adjusted. For example, for elements related to processor performance and price with high importance, higher weights are given during fusion, while for elements with low importance, lower weights are given, thereby generating the translated electronic invoice semantic vector and the translated order path map vector after pre-fusion adjustment.

[0080] Then, the corresponding sub-vectors in the pre-fusion adjusted translated electronic invoice semantic vector and the translated order path map vector are fused according to the determined weights to obtain the fused sub-vectors, and all the fused sub-vectors are combined to generate differential fusion features. For example, the processor performance-related sub-vectors that have been adjusted according to the weights and the sub-vectors in which the processor performance affects the order decision in the ordering link are fused to form a new fused sub-vector, which integrates the information about the processor performance in the electronic invoice semantics and the order path map.

[0081] Classify the elements in the differential fusion features, and preliminarily construct a graph structure based on the element classification results. For example, the elements related to product performance are classified into one category, including fused elements such as processor performance and memory size, and these elements are used as nodes in the graph structure. The connection relationship between nodes is determined according to business logic. For example, a laptop with good processor performance may prompt customers to place orders faster, so a connection is established between the node representing processor performance and the node representing the order decision. At the same time, attributes are assigned to each node. The attributes include the numerical information of the element in the differential fusion features (such as the specific value of the processor performance or the performance level) and the category identification information (indicating that it is a product performance category). The preliminarily constructed graph structure is used to reflect the data relationship in the differential fusion features.

[0082] Check whether there are redundant connections or connections that should exist but have not been established in the nodes in the initially constructed graph structure, delete the detected redundant connections accordingly, supplement the connections that should exist accordingly, and then optimize the properties of the nodes to generate an optimized and adjusted graph structure. For example, if it is found that the connection between a node related to the computer's heat dissipation performance in the product performance category and the payment link is redundant, because there is no direct business logic relationship between the heat dissipation performance and the payment link, delete this connection. If it is found that there should be a connection between product performance and customer satisfaction but there is none, because good product performance usually improves customer satisfaction, supplement this connection. And after supplementing the connection, optimize the properties of the relevant nodes, such as adjusting the properties of the customer satisfaction node and adding a description of the influencing factors related to product performance.

[0083] Mining knowledge structure data in the optimized graph structure, for example, found a substructure consisting of product performance, customer satisfaction, and brand awareness. This substructure represents a business logic unit, indicating that product performance and brand awareness will affect customer satisfaction. At the same time, it was found that the path pattern from customers browsing products, comparing product performance, considering brand awareness, to finally placing an order and affecting customer satisfaction, reflects a process pattern in the business process.

[0084] The mined knowledge structure data are classified and sorted, and the coding rules and ranges are determined for each category of knowledge structure data. For example, for knowledge structure data related to product performance, the coding range is 1-100, for those related to customer satisfaction, it is 101-200, for those related to brand awareness, it is 201-300, etc. Then, according to the determined coding rules and ranges, the knowledge structure data in the graph structure are encoded. For each knowledge structure data, a coding value is assigned according to the category it belongs to and the position in the category, and the graph knowledge structure coding result is generated. For example, for the substructure of product performance-customer satisfaction, if it belongs to a specific structure of the product performance category according to the rules and is at the front of this category, it may be assigned a coding value of 10, indicating the specific meaning and order of this knowledge structure data in the product performance category. This graph knowledge structure coding result can be used for subsequent operations such as the generation of the third cross-platform focus prediction data of the sample cross-platform electronic invoice integration data.

[0085] In a possible implementation, step S150 includes:

[0086] Step S151: determining a first prediction error based on the first cross-platform focus point prediction data and the cross-platform focus point label data.

[0087] Step S152: determining a second prediction error based on the second cross-platform focus point prediction data and the cross-platform focus point label data.

[0088] Step S153: determining a third prediction error based on the third cross-platform focus point prediction data and the cross-platform focus point label data.

[0089] Step S154: performing weighted calculation on the first prediction error, the second prediction error and the third prediction error to generate the network prediction error.

[0090] In this embodiment, taking the notebook computer sales related data of platform A as an example, firstly, a first prediction error is determined based on the first cross-platform focus point prediction data and the cross-platform focus point label data.

[0091] Assume that the first cross-platform focus prediction data shows that processor performance is the factor that customers care about most in laptop sales, and its predicted focus is 40%. The cross-platform focus label data is an accurate result obtained by analyzing a large amount of actual sales data and customer feedback, which shows that the actual focus on processor performance is 30%. The first prediction error here is the difference between the predicted value and the actual value. In this example, the first prediction error is 40% -30% = 10%. This error reflects the degree of deviation between the first cross-platform focus prediction data and the actual situation on the focus of processor performance.

[0092] Then, the second prediction error is determined based on the second cross-platform focus prediction data and the cross-platform focus label data. For example, the second cross-platform focus prediction data predicts that the number of times customers compare different laptop models before placing an order has a greater impact on customer decisions, and predicts that its impact is 30%. However, the cross-platform focus label data shows that the actual impact is 20%. Then the second prediction error is 30% - 20% = 10%, which means that there is a certain deviation between the prediction of the second cross-platform focus prediction data on this factor and the actual situation.

[0093] Then, the third prediction error is determined based on the third cross-platform focus prediction data and the cross-platform focus label data. For example, the third cross-platform focus prediction data believes that the relationship between product performance, customer satisfaction, and brand awareness is an important logical relationship that affects sales, and predicts that this logical relationship has an influence of 50% on sales. However, the cross-platform focus label data shows that the actual influence is 40%. Therefore, the third prediction error is 50% - 40% = 10%. This error reflects the difference between the third cross-platform focus prediction data and the actual situation in this logical relationship prediction. By determining these three prediction errors, a basis is provided for the subsequent calculation of network prediction errors and the training and optimization of the cross-platform focus mining network.

[0094] In a possible implementation, the method further includes:

[0095] Step A110, obtaining target cross-platform electronic invoice integration data.

[0096] Step A120, using the initialized electronic invoice semantic extraction unit, decompose the target cross-platform electronic invoice integration data to generate an electronic invoice data set, extract the electronic invoice semantic vector of each electronic invoice data in the electronic invoice data set, and for each electronic invoice semantic vector, calculate the semantic deviation between the electronic invoice semantic vector and each electronic invoice semantic vector in the electronic invoice semantic vector library of the candidate cross-platform electronic invoice integration data, and determine the electronic invoice semantic vector in the electronic invoice semantic vector library whose semantic deviation from the electronic invoice semantic vector is less than the set deviation.

[0097] Step A130, based on the electronic invoice data corresponding to the determined electronic invoice semantic vector, determine the electricity invoice fragment that has a matching relationship with the target cross-platform electronic invoice integration data, arrange the cross-platform electronic invoice integration data where the electricity invoice fragment is located in descending order according to the data volume of the electricity invoice fragment, and generate cross-platform electronic invoice integration data that has a business order linkage relationship with the target cross-platform electronic invoice integration data.

[0098] Step A140, using the target cross-platform focus mining network, extract corresponding graph knowledge structure encoding results for the cross-platform electronic invoice integration data with business order linkage relationship, and the graph knowledge structure encoding results are the focus point logical features of the cross-platform electronic invoice integration data.

[0099] Step A150, based on the focus point logical features of each cross-platform electronic invoice integration data having a business order linkage relationship, determine the focus point logical features of the target cross-platform electronic invoice integration data.

[0100] Step A160, based on the point of interest logical features of the target cross-platform electronic invoice integration data, searching the point of interest logical features from the point of interest logical feature library to generate a point of interest logical feature sequence.

[0101] Step A170: clustering the focus point logical feature sequence to generate a plurality of cluster members and a cluster center of each cluster member, wherein the cluster center is a focused feature of the focus point logical features in the cluster members.

[0102] Step A180, searching for the concern point logical features with the shortest feature distance from the corresponding cluster center from the multiple cluster members, and using the cross-platform electronic invoice integration data corresponding to the searched multiple concern point logical features as the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data. The concern point logical features in the concern point logical feature library are generated by extracting the corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform concern point mining network.

[0103] In this embodiment, continuing to take the above scenario as an example, the server first obtains the target cross-platform electronic invoice integration data. For example, there are multiple platforms under the e-commerce service provider, including platform A (selling electronic products), platform B (selling fashion apparel) and platform C (selling household goods). The target cross-platform electronic invoice integration data may be the electronic invoice integration data related to the sales of a specific product (such as a laptop) from platform A within a certain period of time. These data include a lot of information such as the sales price, sales quantity, buyer information, sales time, etc. of the laptop, as well as order path data such as the customer's order process and some tag data related to cross-platform focus points (which may be data related to customer interest hotspots obtained from the previous preliminary analysis).

[0104] Next, the target cross-platform electronic invoice integration data is processed using the initialized electronic invoice semantic extraction unit. Taking the laptop computer sales data of platform A as an example, the initialized electronic invoice semantic extraction unit will decompose these data. For example, an electronic invoice contains information such as the model, price, buyer's region, and purchase time of the laptop computer. These information will be split into separate electronic invoice data, and all these individual electronic invoice data constitute the electronic invoice data set. Then, for each electronic invoice data in this electronic invoice data set, the electronic invoice semantic vector is extracted. For example, for the price data of the laptop computer, its electronic invoice semantic vector may contain information such as the price value, the price range, and the comparison relationship with the market average price; for the model of the laptop computer, its semantic vector may contain information such as the model series and configuration level.

[0105] Afterwards, for each electronic invoice semantic vector, the semantic deviation between the electronic invoice semantic vector and each electronic invoice semantic vector in the electronic invoice semantic vector library of the candidate cross-platform electronic invoice integration data is calculated. Assume that the electronic invoice semantic vector library stores a large number of electronic invoice semantic vectors from different platforms and different time periods. The electronic invoice semantic vector of the price of a laptop computer in the target cross-platform electronic invoice integration data will be compared with each vector in the vector library. The method for calculating the semantic deviation may be based on a distance metric between vectors, such as Euclidean distance or cosine similarity. If there is a previously stored electronic invoice semantic vector of the price of a laptop computer in the vector library that is very similar to the target semantic vector in terms of numerical value, price range, and comparison relationship with the market average price, then their semantic deviation will be smaller. Through such a comparison, the electronic invoice semantic vector in the electronic invoice semantic vector library whose semantic deviation from the target electronic invoice semantic vector is less than the set deviation is determined.

[0106] Based on the electronic invoice data corresponding to the determined electronic invoice semantic vectors, determine the power invoice fragments that have a matching relationship with the target cross-platform electronic invoice integration data. The power invoice fragments here may refer to those parts of the electronic invoice data that are highly semantically matched with the target data among the numerous electronic invoice data. For example, if the target data is the sales data of a specific model of laptop computer, then those electronic invoice data fragments that are also related to the sales of the model of laptop computer will be determined as matching power invoice fragments. Then, the cross-platform electronic invoice integration data containing these power invoice fragments are arranged in descending order according to the data volume of the power invoice fragments, and the cross-platform electronic invoice integration data having a business order linkage relationship with the target cross-platform electronic invoice integration data is generated. In other words, if the matching power invoice fragment data contained in a certain cross-platform electronic invoice integration data is large, it will be ranked in front, so that a set of cross-platform electronic invoice integration data ranked according to the degree of association with the target data is obtained.

[0107] The target cross-platform focus mining network is used to extract the corresponding graph knowledge structure encoding results for the cross-platform electronic invoice integration data with business order linkage. For example, for the cross-platform electronic invoice integration data that is ranked first (which may be related data about the large-scale sales of a popular model of laptop computers), the target cross-platform focus mining network will analyze the electronic invoice integration data, order path data and other information contained therein, and generate the graph knowledge structure encoding results according to the previous processing flow (such as analyzing semantic vectors, performing fusion operations, building graph structures, etc.). This result is the focus logical feature of the cross-platform electronic invoice integration data, which may contain the hot factors that customers pay attention to in the sales of the laptop computer (such as the logical relationship between factors such as price, performance, and brand, etc.).

[0108] Based on the logical features of the cross-platform electronic invoice integration data with business order linkage, determine the logical features of the target cross-platform electronic invoice integration data. Assume that in the cross-platform electronic invoice integration data with business order linkage, there are multiple data about laptop sales, and the logical features of the focus of some data show that price and brand are the main factors that customers care about, while the logical features of the focus of other data show that performance and after-sales service are the main factors. Based on this information, the logical features of the focus of the target cross-platform electronic invoice integration data are determined through some statistical or analytical methods (such as weighted average, majority voting, etc.). For example, it is finally determined that for the target laptop sales data, price, performance and brand are the main logical features of the focus.

[0109] Based on the focus logic features of the target cross-platform electronic invoice integration data, the focus logic features are searched from the focus logic feature library to generate a focus logic feature sequence. The focus logic feature library is generated by extracting the corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform focus mining network. For example, if the focus logic features of the target cross-platform electronic invoice integration data are price, performance and brand, then searching for related logical features in the focus logic feature library may find logical features related to price, performance and brand in the sales of other similar products (such as tablets, smart phones, etc.). These found focus logic features are arranged in a certain order to generate a focus logic feature sequence.

[0110] Cluster the focus logic feature sequence to generate multiple cluster members and the cluster center of each cluster member. For example, the price-related logic features in the focus logic feature sequence may be divided into one cluster, the performance-related logic features into another cluster, and the brand-related logic features into a third cluster. These clusters are cluster members. The cluster center in each cluster member is obtained by calculating the focus features of the focus logic features in the cluster. For example, in the price-related cluster members, the cluster center may be a value or vector representing the most typical price-focus logic feature, which can represent the commonality of most focus logic features in this cluster.

[0111] Finally, the attention point logical features with the smallest feature distance from the corresponding cluster center are searched from multiple cluster members respectively, and the cross-platform electronic invoice integration data corresponding to the searched multiple attention point logical features are used as the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data. For example, among the price-related cluster members, the attention point logical feature with the smallest distance from the cluster center (the most typical price attention logical feature) is found. This attention point logical feature may come from the cross-platform electronic invoice integration data of a laptop computer sales, and then this cross-platform electronic invoice integration data is determined as the data that has a matching relationship with the target cross-platform electronic invoice integration data. In this way, the server can use the existing data and mining network to conduct an in-depth analysis of the target cross-platform electronic invoice integration data, find the relevant data that matches it, and provide a basis for further decision-making, analysis or data mining.

[0112] In a possible implementation, the method further includes:

[0113] Get a global, cross-platform electronic invoice database.

[0114] The target cross-platform focus mining network is used to extract graph knowledge structure encoding results of each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database, and the graph knowledge structure encoding results are the focus logic features of the cross-platform electronic invoice integration data.

[0115] Based on the focus point logical features of each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database, a focus point logical feature library is determined.

[0116] In this embodiment, the server first obtains a global cross-platform electronic invoice database. In this e-commerce service provider, the global cross-platform electronic invoice database covers all electronic invoice-related data of multiple platforms under its umbrella (such as the previously mentioned platform A selling electronic products, platform B selling fashion apparel, platform C selling household goods, etc.). These data contain sales records of different types of products, from high-end electronic products to ordinary daily necessities, and each transaction has detailed electronic invoice information. For example, the sales records of laptops in platform A include sales of computers of different brands, different configurations, and different price ranges, including detailed information of the buyers, sales time, sales channels, etc.; the sales records of fashion apparel in platform B contain information such as styles, sizes, colors, and sales seasons of various types of clothing; the sales records of household goods in platform C contain relevant data such as the material, size, and style of furniture. These rich data constitute a huge global cross-platform electronic invoice database.

[0117] Next, the target cross-platform focus mining network is used to process the cross-platform electronic invoice integration data in the global cross-platform electronic invoice database. Taking the laptop computer sales data of platform A as an example, the target cross-platform focus mining network will operate according to the previously defined process. For each laptop computer sales transaction, its electronic invoice integration data includes basic invoice information, order path data, etc. The mining network first analyzes these data, such as processing the electronic invoice semantic vector, fusing it with the order path graph vector, and then encodes the fused features into a graph knowledge structure. For example, for a laptop computer transaction with a high sales price and high-end configuration, its graph knowledge structure encoding result may reflect that in this case, customers are more concerned about the brand, after-sales service, and the relationship between product performance and price performance. Similarly, for the fashion clothing sales of platform B, the graph knowledge structure encoding result may reflect the customer's focus on popular styles, color matching, and brand image. Logical features; for the home furnishings sales of platform C, it may be the focus on material quality, matching degree with the overall home style, and price. Logical features.

[0118] By performing the above operations on all cross-platform electronic invoice integration data in the global cross-platform electronic invoice database, the server determines the focus logic feature library based on the focus logic features of each cross-platform electronic invoice integration data. This library is like a knowledge treasure house, storing the hot logical relationships that customers may pay attention to in various product sales situations. For example, in the electronic product category, the customer's focus logic on factors such as performance, brand, and after-sales service in different price ranges may be stored; in the fashion apparel category, there is the customer's focus logic on color, brand image, and popular elements in different seasons and styles; in the household goods category, there is the customer's focus logic on material, size, price, etc. in different home styles. These focus logic feature libraries will provide an important basis for subsequent data analysis and matching.

[0119] In a possible implementation, the method further includes:

[0120] Obtain target cross-platform electronic invoice integration data selected from the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data.

[0121] Obtain the target cluster member where the logical feature of the focus point of the target cross-platform electronic invoice integration data is located.

[0122] The focus logic features of the cross-platform electronic invoice integration data that has a business order linkage relationship with the target cross-platform electronic invoice integration data and the focusing features of the focus logic features in the target cluster members are used as the focus logic features for adjusting the target cross-platform electronic invoice integration data.

[0123] Based on the adjusted focus logic features and the focus logic feature library, cross-platform electronic invoice integration data that matches the target cross-platform electronic invoice integration data is searched from the global cross-platform electronic invoice database, and the found cross-platform electronic invoice integration data is the existing processed data.

[0124] In this embodiment, the server obtains target cross-platform electronic invoice integration data selected from the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data. Assume that in the previous operation, a batch of cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data (such as the sales data of a series of laptop computers in platform A) has been determined, and now a specific target cross-platform electronic invoice integration data is selected from them for further analysis. At the same time, the target cluster member where the focus logical feature of this target cross-platform electronic invoice integration data is located is obtained. For example, if the focus logical feature has been clustered before, the focus logical feature of the target cross-platform electronic invoice integration data (such as the logical relationship in terms of price and performance) may be located in a specific cluster member, and this cluster member may contain other focus logical features with similar logical relationships.

[0125] Then, the attention logic features of the cross-platform electronic invoice integration data that has a business order linkage relationship with the target cross-platform electronic invoice integration data and the focus features of the attention logic features in the target cluster members are used as the attention logic features of the target cross-platform electronic invoice integration data adjustment. Taking the sales of laptops as an example, assuming that the target cross-platform electronic invoice integration data is about the sales of a new model of laptops, the cross-platform electronic invoice integration data that has a business order linkage relationship with it may include the sales data of other models of the same brand or laptops with similar configurations. The attention logic features of these related data (such as customers' attention to the cost-effectiveness of different configurations, the comparison logic of different models under the brand, etc.) and the focus features of the attention logic features in the target cluster members (such as the typical logical relationship in the price-performance relationship cluster) are combined to form the attention logic features of the target cross-platform electronic invoice integration data adjustment. This may discover some new logical relationships, such as the customer's expectation logic of the performance improvement of the new model of laptops at a specific price due to the competition of other models of the same brand in the market.

[0126] Finally, based on the adjusted focus logic features and the focus logic feature library, the cross-platform electronic invoice integration data that matches the target cross-platform electronic invoice integration data is searched from the global cross-platform electronic invoice database, and the cross-platform electronic invoice integration data found is the existing processing data. For example, the adjusted focus logic features indicate the customer's expectation for performance improvement of a new model of laptop computer at a specific price. The server will search for sales data with similar focus logic features in the global cross-platform electronic invoice database. These data may be sales records of other similar products with similar prices and performance. These sales records found are the existing processing data. These existing processing data can provide reference for the analysis of the current target cross-platform electronic invoice integration data. For example, previous marketing strategies, customer feedback processing methods, etc. can be used to better deal with current product sales and market promotion issues.

[0127] In a possible implementation, the method further includes:

[0128] A target order path data document is obtained, where the order path data document is used to search for cross-platform electronic invoice integration data.

[0129] The target cross-platform focus mining network is utilized to extract the graph knowledge structure encoding result of the target order path data document, and the graph knowledge structure encoding result is the focus logical feature of the target order path data document.

[0130] Based on the focus logic features of the target order path data document and the focus logic feature library, matching cross-platform electronic invoice integration data is searched from the global cross-platform electronic invoice database. The focus logic features in the focus logic feature library are generated by extracting corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform focus mining network.

[0131] In this embodiment, the server first obtains the target order path data document. In the e-commerce service provider, the target order path data document contains a wealth of order process related information, which is used to find cross-platform electronic invoice integration data. For example, in platform A (selling electronic products), the target order path data document may record in detail the entire process of customers purchasing laptops. Starting from the first time the customer visits the laptop area of ​​platform A, the different laptop product pages browsed by the customer, the time spent on each page, whether the product is added to the favorites or shopping cart, and other operations are recorded. Then the specific time when the customer places an order, the payment method selected (such as credit card payment, online payment platform, etc.), whether a coupon or promotion is used, and then the merchant's order processing process, including order review time, which warehouse to ship from, and other information, and then to the logistics and transportation links, such as the name of the logistics company, the estimated delivery time and the actual delivery time, and finally the relevant information of the customer confirming the receipt of the goods. These detailed order path data form the target order path data document, which is a comprehensive record of the entire purchase process, which is of great significance for subsequent analysis and data search.

[0132] Then, the server uses the target cross-platform concern mining network to process the target order path data document and extract the graph knowledge structure encoding result of the target order path data document, which is the concern logical feature of the target order path data document. Taking the order path data of a laptop as an example, the target cross-platform concern mining network will analyze according to the pre-set process. First, the network will decompose the elements in the order path data document, such as order time, payment method, logistics transportation and other links as different elements. Then, the importance and relationship of each element in the entire order process are analyzed. For example, the order time may be related to the deadline of the promotion. If the customer places an order when the promotion is about to end, it may imply that the promotion has a greater impact on the customer's order decision; the choice of payment method may be related to the customer's consumption habits and security considerations. If the customer chooses a payment method with specific discounts or high security, it reflects the customer's concern in the payment link. Through the analysis and processing of these elements, the network will construct a graph structure, in which the nodes represent different order process elements and the edges represent the relationship between the elements. Then, the graph structure is encoded. For example, the relationship between the order time and the promotion deadline can be assigned a specific coding value based on factors such as the closeness and the direction of influence. Similar coding is performed for the relationship between the payment method and the customer's consumption habits. Finally, the graph knowledge structure encoding result of the target order path data document is obtained. This result reflects the logical characteristics of the focus points in the order path data document. For example, it may be concluded that in the sale of laptops, the timeliness of the promotion and the preferential nature of the payment method are important concerns of customers in the order path.

[0133] Finally, the server searches for matching cross-platform electronic invoice integration data from the global cross-platform electronic invoice database based on the focus logic features of the target order path data document and the focus logic feature library. The focus logic feature library is generated by extracting the corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform focus mining network. For example, the focus logic features of the sales process of various products (including electronic products, fashion clothing, household goods, etc.) are stored in the focus logic feature library. For notebook computer sales, there may be focus logic features of different models, different price ranges, and different sales seasons. For example, some high-end notebook computers may pay more attention to the brand image and after-sales service in the order path in a specific sales season, while some cost-effective notebook computers may pay more attention to promotional activities and logistics speed. When the server wants to find matching cross-platform electronic invoice integration data, it will compare the focus logic features of the target order path data document (such as the timeliness of promotional activities and the preferential nature of payment methods) with various logical features in the focus logic feature library. If similar logical features are found in the focus logic feature library (for example, other products have similar customer focus logic in promotions and payment methods), then the cross-platform electronic invoice integration data associated with these logical features will be regarded as matching cross-platform electronic invoice integration data. These matching cross-platform electronic invoice integration data can provide reference for further business analysis, such as understanding customer purchase behavior and sales trends under similar focus points, thereby helping e-commerce service providers optimize sales strategies and improve product recommendation systems.

[0134] Figure 2 The cross-platform electronic invoice data integration and processing system 100 shown includes: a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the cross-platform electronic invoice data integration and processing system 100 may also include a transceiver 1004, which may be used for data interaction between the server and other servers, such as data transmission and / or data reception. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of the cross-platform electronic invoice data integration and processing system 100 does not constitute a limitation on the embodiments of the present application.

[0135] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0136] The bus 1002 may include a path to transmit information between the above components. The bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 1002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0137] The memory 1003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), magnetic disk storage medium, other magnetic storage devices, or any other medium that can be used to carry or store program code and can be read by a computer, without limitation herein.

[0138] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.

[0139] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are pre-installed. When the processor executes the computer executable instructions, the above cross-platform electronic invoice data integration and processing method is implemented.

[0140] Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A cross-platform electronic invoice data integration and processing method, characterized in that: The method comprises: Obtain sample cross-platform electronic invoice integration data, each of the sample cross-platform electronic invoice integration data includes cross-platform electronic invoice integration data, order path data of the cross-platform electronic invoice integration data, and cross-platform focus tag data; Using the first unit module of the cross-platform focus mining network, feature selection is performed on the basic electronic invoice semantic vector of the cross-platform electronic invoice integration data to generate a translated electronic invoice semantic vector, and first cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the translated electronic invoice semantic vector; Using the second unit module of the cross-platform focus mining network, feature selection is performed on the basic order path graph vector of the order path data to generate a translated order path graph vector, and second cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the translated order path graph vector; Using the third unit module of the cross-platform focus mining network, a differential fusion operation is performed on the translated electronic invoice semantic vector and the translated order path graph vector to generate differential fusion features, and the differential fusion features are encoded in a graph knowledge structure to generate a graph knowledge structure encoding result, and third cross-platform focus prediction data for the sample cross-platform electronic invoice integration data is generated based on the graph knowledge structure encoding result; the graph knowledge structure encoding result is a focus logic feature of the sample cross-platform electronic invoice integration data; Determine a network prediction error based on the first cross-platform focus point prediction data, the second cross-platform focus point prediction data, the third cross-platform focus point prediction data and the cross-platform focus point label data, train and optimize the cross-platform focus point mining network based on the network prediction error until the cross-platform focus point mining network meets the network convergence requirements, and generate a target cross-platform focus point mining network; The step of utilizing the third unit module of the cross-platform focus mining network to perform a differential fusion operation on the translated electronic invoice semantic vector and the translated order path graph vector to generate differential fusion features, performing graph knowledge structure encoding on the differential fusion features, and generating graph knowledge structure encoding results includes: Analyze the position and importance of the semantic concept represented by each element in the translated electronic invoice semantic vector in the entire electronic invoice semantic system, as well as the degree of association between each element in the translated order path graph vector and each link of the order process, and generate a first feature analysis result of the translated electronic invoice semantic vector and a second feature analysis result of the translated order path graph vector; According to the first feature analysis result and the second feature analysis result, the translated electronic invoice semantic vector and the translated order path graph vector are decomposed, and for the translated electronic invoice semantic vector, elements representing related semantic concepts or related in business logic are combined to generate a first sub-vector sequence, and for the translated order path graph vector, related elements are combined according to different stages of the order process to generate a second sub-vector sequence, and the first sub-vector sequence and the second sub-vector sequence are rearranged to make the first sub-vector sequence and the second sub-vector sequence logically adapt to each other; Determine a target sub-vector combination having associated logical meaning in the first sub-vector sequence and the second sub-vector sequence, and align the target sub-vector combination in position to generate a translated electronic invoice semantic vector and a translated order path graph vector after feature alignment; Based on the prior knowledge of business logic, the importance of features in the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment is evaluated to generate corresponding feature importance evaluation results, and the translated electronic invoice semantic vector and the translated order path graph vector after feature alignment are pre-fused and adjusted according to the feature importance evaluation results to generate the translated electronic invoice semantic vector and the translated order path graph vector after pre-fusion and adjustment; The corresponding sub-vectors in the pre-fused and adjusted translated electronic invoice semantic vector and the translated order path graph vector are fused according to the determined weights to obtain fused sub-vectors, and all fused sub-vectors are combined to generate differential fusion features; Classifying the elements in the difference fusion feature, and preliminarily constructing a graph structure according to the element classification results, wherein the elements in each category are used as nodes in the graph structure, the connection relationship between the nodes is determined according to the business logic, and attributes are assigned to each node, the attributes include the numerical information and category identification information of the element in the difference fusion feature, and the preliminarily constructed graph structure is used to reflect the data relationship in the difference fusion feature; Detect whether there are redundant connections or connections that should exist but are not established in the nodes in the initially constructed graph structure, delete the detected redundant connections accordingly, supplement the connections that should exist accordingly, and then optimize the properties of the nodes to generate an optimized and adjusted graph structure; Mining knowledge structure data in the optimized and adjusted graph structure, wherein the knowledge structure data includes a substructure and a path pattern in the graph structure, wherein the substructure is a structure composed of a group of interrelated nodes, representing a business logic unit, and the path pattern is a connection path between nodes in the knowledge structure, reflecting a process pattern in the business process; The mined knowledge structure data are classified and sorted, and the coding rules and scopes are determined for each category of knowledge structure data. According to the determined coding rules and scopes, the knowledge structure data in the graph structure are encoded. For each knowledge structure data, a coding value is assigned according to the category it belongs to and its position in the category to generate the graph knowledge structure coding result.

2. The cross-platform electronic invoice data integration and processing method according to claim 1 is characterized in that: The first unit module includes an initialization electronic invoice semantic extraction unit and an electronic invoice semantic translation unit; the first unit module using the cross-platform focus mining network performs feature selection on the basic electronic invoice semantic vector of the cross-platform electronic invoice integration data to generate a translated electronic invoice semantic vector, including: Utilizing the initialized electronic invoice semantic extraction unit, decomposing the cross-platform electronic invoice integration data to generate an electronic invoice data set, extracting the basic electronic invoice semantic vector of each electronic invoice data in the electronic invoice data set, and generating a basic electronic invoice semantic vector set; The electronic invoice semantic translation unit is used to perform feature domain mapping on the basic electronic invoice semantic vector set to generate a feature domain mapping result, and feature selection is performed on the feature domain mapping result to generate a translated electronic invoice semantic vector.

3. The cross-platform electronic invoice data integration and processing method according to claim 1 is characterized in that: The second unit module includes an initialization graph autoencoder and an order path graph translation unit; the second unit module using the cross-platform focus mining network performs feature selection on the basic order path graph vector of the order path data to generate a translated order path graph vector, including: Utilizing the initialized graph autoencoder, extracting a basic order path graph vector of the order path data; The order path graph translation unit is used to perform feature selection on the basic order path graph vector to generate a translated order path graph vector.

4. The cross-platform electronic invoice data integration and processing method according to claim 1 is characterized in that: The determining of a network prediction error based on the first cross-platform focus point prediction data, the second cross-platform focus point prediction data, the third cross-platform focus point prediction data and the cross-platform focus point label data, and training and optimizing the cross-platform focus point mining network based on the network prediction error includes: Determining a first prediction error based on the first cross-platform focus point prediction data and the cross-platform focus point label data; Determining a second prediction error based on the second cross-platform focus point prediction data and the cross-platform focus point label data; Determining a third prediction error based on the third cross-platform point of interest prediction data and the cross-platform point of interest label data; The first prediction error, the second prediction error and the third prediction error are weightedly calculated to generate the network prediction error.

5. The cross-platform electronic invoice data integration and processing method according to any one of claims 2 to 4, characterized in that: The method further comprises: Obtain target cross-platform electronic invoice integration data; Using the initialized electronic invoice semantic extraction unit, the target cross-platform electronic invoice integration data is decomposed to generate an electronic invoice data set, and the electronic invoice semantic vector of each electronic invoice data in the electronic invoice data set is extracted. For each electronic invoice semantic vector, the semantic deviation between the electronic invoice semantic vector and each electronic invoice semantic vector in the electronic invoice semantic vector library of the candidate cross-platform electronic invoice integration data is calculated, and the electronic invoice semantic vector in the electronic invoice semantic vector library whose semantic deviation from the electronic invoice semantic vector is less than the set deviation is determined; Based on the determined electronic invoice data corresponding to the electronic invoice semantic vector, determine the electric invoice fragments that have a matching relationship with the target cross-platform electronic invoice integration data, arrange the cross-platform electronic invoice integration data where the electric invoice fragments are located in descending order according to the data volume of the electric invoice fragments, and generate cross-platform electronic invoice integration data that has a business order linkage relationship with the target cross-platform electronic invoice integration data; By using the target cross-platform focus mining network, corresponding graph knowledge structure encoding results are respectively extracted for the cross-platform electronic invoice integration data with the business order linkage relationship, and the graph knowledge structure encoding results are the focus logic features of the cross-platform electronic invoice integration data; Determine the focus point logical features of the target cross-platform electronic invoice integration data based on the focus point logical features of each cross-platform electronic invoice integration data having a business order linkage relationship; Based on the point of interest logic features of the target cross-platform electronic invoice integration data, searching the point of interest logic features from the point of interest logic feature library to generate a point of interest logic feature sequence; Clustering the attention point logical feature sequence to generate a plurality of cluster members and a cluster center of each cluster member, wherein the cluster center is a focused feature of the attention point logical features in the cluster members; Search for the focus point logical feature with the smallest feature distance to the corresponding cluster center from the multiple cluster members respectively, and use the cross-platform electronic invoice integration data corresponding to the searched multiple focus point logical features as the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data; the focus point logical features in the focus point logical feature library are generated by extracting the corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform focus point mining network.

6. The cross-platform electronic invoice data integration and processing method according to claim 5 is characterized in that: The method further comprises: Obtain a global cross-platform electronic invoice database; Utilizing the target cross-platform focus mining network, respectively extracting graph knowledge structure encoding results of each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database, wherein the graph knowledge structure encoding results are focus logic features of the cross-platform electronic invoice integration data; Based on the focus point logical features of each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database, a focus point logical feature library is determined.

7. The cross-platform electronic invoice data integration and processing method according to claim 5 is characterized in that: The method further comprises: Acquire target cross-platform electronic invoice integration data selected from the cross-platform electronic invoice integration data that has a matching relationship with the target cross-platform electronic invoice integration data; Acquire the target cluster member where the logical feature of the focus point of the target cross-platform electronic invoice integration data is located; The focus logic features of the cross-platform electronic invoice integration data having a business order linkage relationship with the target cross-platform electronic invoice integration data and the focus features of the focus logic features in the target cluster members are used as the focus logic features of the target cross-platform electronic invoice integration data; Based on the adjusted focus logic features and the focus logic feature library, cross-platform electronic invoice integration data that matches the target cross-platform electronic invoice integration data is searched from the global cross-platform electronic invoice database, and the found cross-platform electronic invoice integration data is the existing processed data.

8. The cross-platform electronic invoice data integration and processing method according to any one of claims 1 to 4, characterized in that: The method further comprises: Obtaining a target order path data document, wherein the order path data document is used to search for cross-platform electronic invoice integration data; Utilizing the target cross-platform focus mining network, extracting the graph knowledge structure encoding result of the target order path data document, wherein the graph knowledge structure encoding result is the focus logic feature of the target order path data document; Based on the focus logic features and the focus logic feature library of the target order path data document, matching cross-platform electronic invoice integration data are searched from the global cross-platform electronic invoice database; the focus logic features in the focus logic feature library are generated by extracting corresponding graph knowledge structure encoding results from each cross-platform electronic invoice integration data in the global cross-platform electronic invoice database using the target cross-platform focus mining network.

9. A cross-platform electronic invoice data integration and processing system, characterized in that: The cross-platform electronic invoice data integration and processing system includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the cross-platform electronic invoice data integration and processing method described in any one of claims 1-8.

Images