Intelligent e-commerce data processing method and system based on multi-source data fusion

Abstract

The embodiment of the invention provides an intelligent e-commerce data processing method and system based on multi-source data fusion, and belongs to the technical field of computers, and the method comprises the steps: obtaining various original e-commerce data; dividing the original e-commerce data into normal phase data and reverse phase data; establishing a basic double-data-flow model through the normal-phase data and the reverse-phase data; calculating a two-phase exchange coefficient of the original e-commerce data; correcting the basic double-data-flow model according to the two-phase exchange coefficient to obtain a final double-data-flow model; inputting new original e-commerce data into the final double-data-stream model to obtain feature output data; and inputting the feature output data into a data turbulence model to obtain simulation fusion data, realizing dynamic adjustment of the data fusion weight according to the context through two-phase exchange coefficients, providing multi-aspect interpretability through an attention mechanism, and improving the reasoning effect of the model.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an intelligent e-commerce data processing method and system based on multi-source data fusion, a computer device, and a storage medium. Background Technology

[0002] E-commerce data now includes not only transaction records but also user behavior, content, external data, and more. These data vary greatly and have high real-time requirements, creating a need for multi-source data fusion. Current technologies struggle to accurately align unstructured review text and images with structured transaction records across time and entity (e.g., user / product) dimensions, and are computationally expensive. Data from different sources contains significant noise, missing information, duplication, and conflicts (e.g., the same product has different prices on different channels), requiring complex cleaning and governance processes to ensure a single source of truth. Existing models for fusing multi-source data, such as multimodal learning and graph neural networks, typically have large parameters, are time-consuming and resource-intensive for training and inference, and have poor inference performance. Summary of the Invention

[0003] In view of the above problems, embodiments of the present invention are proposed to provide an intelligent e-commerce data processing method based on multi-source data fusion, an intelligent e-commerce data processing system based on multi-source data fusion, a computer device, and a storage medium to overcome or at least partially solve the above problems.

[0004] To address the aforementioned problems, this invention discloses an intelligent e-commerce data processing method based on multi-source data fusion, comprising: We obtained various types of raw e-commerce data; The original e-commerce data is divided into positive phase data and negative phase data; A basic dual-data-stream model is established using the positive-phase data and the negative-phase data; Calculate the two-phase exchange coefficients of the original e-commerce data; The basic dual-data-stream model is modified based on the two-phase exchange coefficients to obtain the final dual-data-stream model; The new raw e-commerce data is input into the final dual data stream model to obtain feature output data; The feature output data is input into the data turbulence model to obtain simulation fusion data.

[0005] Preferably, dividing the original e-commerce data into positive-phase data and negative-phase data includes: Identify the data structure type, source stability parameters, and generation frequency of the original e-commerce data; Based on the data structure type, source stability parameters, and generation frequency, the original e-commerce data is classified into positive phase data and negative phase data.

[0006] Preferably, establishing a basic dual-datastream model using the positive-phase data and the negative-phase data includes: The positive phase data is input into a fully connected network to obtain the constructed positive phase data encoder; The inverted data is input into the BERT model to obtain the constructed inverted data encoder; The positive-phase data encoder and the negative-phase data encoder together form a basic dual-data-stream model.

[0007] Preferably, the calculation of the two-phase exchange coefficients of the original e-commerce data includes: Obtain the positive phase vector features output by the positive phase data encoder; Obtain the inverted vector features output by the inverted data encoder; The positive and negative phase vector features are normalized and dimensionality reduced to obtain the attention weight matrix. The coefficients corresponding to the attention weight matrix are determined as two-phase exchange coefficients.

[0008] Preferably, the step of modifying the basic dual-datastream model according to the two-phase exchange coefficients to obtain the final dual-datastream model includes: The positive-phase vector features and the negative-phase vector features are combined to generate fused features; the two-phase exchange coefficients are set as the initial coefficients of the fused features to obtain the modified fused features; The positive and negative data encoders are trained using the modified fusion features to obtain the final dual data stream model.

[0009] Preferably, the step of inputting the feature output data into the data turbulence model to obtain simulation fusion data includes: The feature output data is subjected to random perturbation injection to obtain the first feature data; The first feature data is subjected to Dropout to obtain the second feature data; Adversarial samples are generated based on the feature output data, and third feature data is obtained by combining the adversarial samples with the second feature data. The third feature data is input into the task classification network to obtain the predicted simulation fusion data.

[0010] Preferably, classifying the original e-commerce data into positive-phase data and negative-phase data based on the data structure type, source stability parameters, and generation frequency includes: The data structure type of the original e-commerce data is identified as structured data or unstructured data, resulting in a first type of data and a second type of data; Identify the confidence levels of the first type of data and the second type of data; Identify the generation frequencies of the first type of data and the second type of data; Data from the first type of data with a confidence level and generation frequency higher than a preset threshold are selected and identified as positive phase data; data from the second type of data with a confidence level and generation frequency lower than a preset threshold are selected and identified as negative phase data.

[0011] This invention discloses an intelligent e-commerce data processing system based on multi-source data fusion, comprising: The first acquisition module is used to acquire various types of raw e-commerce data; The segmentation module is used to divide the original e-commerce data into positive phase data and negative phase data; A module is established to build a basic dual-datastream model using the positive-phase data and the negative-phase data; The calculation module is used to calculate the two-phase exchange coefficients of the original e-commerce data; The correction module is used to correct the basic dual-data-stream model according to the two-phase exchange coefficients to obtain the final dual-data-stream model. The input module is used to input new raw e-commerce data into the final dual data stream model to obtain feature output data; The simulation fusion data module is used to input the feature output data into the data turbulence model to obtain simulation fusion data.

[0012] This invention also discloses a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described steps of intelligent e-commerce data processing based on multi-source data fusion.

[0013] This invention also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described steps of intelligent e-commerce data processing based on multi-source data fusion.

[0014] The embodiments of the present invention have the following advantages: In this embodiment of the invention, the intelligent e-commerce data processing method based on multi-source data fusion includes: acquiring multiple types of raw e-commerce data; and dividing the raw e-commerce data into positive-phase data and negative-phase data. A basic dual-dataflow model is established using the positive-phase and negative-phase data; the two-phase exchange coefficients of the original e-commerce data are calculated; the basic dual-dataflow model is modified according to the two-phase exchange coefficients to obtain the final dual-dataflow model; new original e-commerce data is input into the final dual-dataflow model to obtain feature output data; the feature output data is input into the data turbulence model to obtain simulated fusion data. The two-phase exchange coefficients enable dynamic adjustment of data fusion weights according to the context, overcoming the limitations of static fusion. The robust design of the entire process from data partitioning and feature fusion to final output significantly improves the system's performance in complex real-world environments. While maintaining the high performance of the deep learning model, the attention mechanism provides multifaceted interpretability, improving the model's inference performance. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the steps of an embodiment of an intelligent e-commerce data processing method based on multi-source data fusion according to an embodiment of the present invention. Figure 2 This is a structural block diagram of an embodiment of an intelligent e-commerce data processing system based on multi-source data fusion according to an embodiment of the present invention; Figure 3 This is an internal structural diagram of a computer device according to one embodiment. Detailed Implementation

[0017] To make the technical problems, technical solutions, and beneficial effects solved by the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0018] In this embodiment of the invention, general raw e-commerce data is classified to obtain positive phase master data and negative phase auxiliary data. A data encoder is trained based on the positive phase master data and negative phase auxiliary data to obtain feature vectors corresponding to the two data streams. The positive phase data and negative phase data are regarded as continuous aggregated data that permeate and interact with each other. The data transmission is quantified by the two-phase exchange coefficient. Then, by adding a data turbulence layer to the data turbulence model, corresponding operations of the data turbulence layer are performed on the output data, including random perturbation injection, Dropout application, and generation of adversarial examples, etc., to introduce random perturbation, enhance the model's generalization ability, reduce the uncertainty of simulated data, and generate more robust and diverse fused data.

[0019] Reference Figure 1 This document illustrates a flowchart of an embodiment of an intelligent e-commerce data processing method based on multi-source data fusion according to the present invention. The method may specifically include the following steps: Step 101: Obtain various types of raw e-commerce data; In this embodiment of the invention, the original e-commerce data may include transaction records, product information, user basic information, user behavior logs, user-generated content, and external data. All of this original e-commerce data is publicly available and can be obtained from the internet through methods such as web scraping. The product information may include product name, price, and inventory. User basic information may include user profiles and registration information. User behavior logs may include click information, browsing information, and favorites information. User-generated content may include comment information, rating information, image information, and video information. External data may include competitor prices and weather data. This embodiment of the invention does not impose excessive restrictions on the types of original e-commerce data. Furthermore, in this embodiment of the invention, the intelligent e-commerce data processing method based on multi-source data fusion can be applied to various terminals, which can be various computers. This embodiment of the invention does not limit the specific type of electrical equipment. The operating system of the terminal can include Android, Harmony OS, IOS, Windows Phone, Windows, etc. This invention does not impose too many restrictions on this.

[0020] Step 102: Divide the original e-commerce data into positive phase data and negative phase data; In practical application to this embodiment of the invention, the original e-commerce data is classified into positive phase data and negative phase data. Positive phase data and negative phase data are data streams with different characteristics and functions. Positive phase data refers to continuous, stable, structured, high-confidence, and real-time data with a fixed pattern that can be directly used for numerical calculations. Negative phase data refers to discrete, dynamic, unstructured or semi-structured, low-confidence, low-frequency data without a fixed format that requires preprocessing to extract features.

[0021] Specifically, in the embodiments of the invention, the original e-commerce data is divided into positive phase data and negative phase data, including: identifying the data structure type, source stability parameters and generation frequency of the original e-commerce data; and classifying the original e-commerce data into positive phase data and negative phase data according to the data structure type, source stability parameters and generation frequency.

[0022] In one specific embodiment, classifying the original e-commerce data into positive-phase data and negative-phase data based on the data structure type, source stability parameters, and generation frequency includes: The data structure type of the original e-commerce data is identified as structured data or unstructured data, resulting in a first type of data and a second type of data; Identify the confidence levels of the first type of data and the second type of data; Identify the generation frequencies of the first type of data and the second type of data; Data from the first type of data with a confidence level and generation frequency higher than a preset threshold are selected and identified as positive phase data; data from the second type of data with a confidence level and generation frequency lower than a preset threshold are selected and identified as negative phase data.

[0023] Specifically, the structured data can include database tables such as user ID, order amount, product price, etc., and configuration data in JSON / XML format; check whether the data has a predefined schema, or can be queried directly via SQL or DataFrame; Unstructured data can refer to comment text, product images, and user behavior logs. The content structure can be initially analyzed using file type detection or natural language processing.

[0024] The source stability parameters may include parameters such as confidence level, data integrity rate, and historical consistency rate. This embodiment of the invention does not impose too many restrictions on these parameters. Among them, the data integrity rate can refer to the ratio of the number of complete fields to the total number of fields, the historical consistency rate can refer to the ratio of the number of consistent records to the total number of records, and the confidence level can also be a preset value, such as 0.95 for internal system data, 0.7 for third-party API data, and 0.6 for crawler data.

[0025] For example, the real-time transaction records in the original e-commerce data have a structured structure, a high stability parameter, and a high generation frequency, and are therefore classified as positive phase data. In addition, the user review text in the original e-commerce data is unstructured, has a medium stability parameter, and a medium generation frequency, which is determined to be inverted data. In this embodiment of the invention, the preset threshold can be any value set by those skilled in the art based on the actual situation, and this embodiment of the invention does not impose too many restrictions on it; Step 103: Establish a basic dual-datastream model using the positive-phase data and the negative-phase data; In a specific example of an embodiment of the present invention, the step of establishing a basic dual-datastream model using the positive-phase data and the negative-phase data includes: inputting the positive-phase data into a fully connected network to obtain a constructed positive-phase data encoder; inputting the negative-phase data into a BERT model to obtain a constructed negative-phase data encoder; and using the positive-phase data encoder and the negative-phase data encoder to form a basic dual-datastream model.

[0026] Among them, the positive phase data encoder refers to a neural network module specifically designed to process structured, high-confidence, and real-time main data streams. The core function of this encoder is to transform raw positive phase data, such as transaction records, basic user information, and product attributes, into dense low-dimensional vector representations that can fully characterize their intrinsic features.

[0027] For example, the positive phase data is first preprocessed, including normalization of numerical features and embedding of categorical features. Then, the processed features are input into a fully connected network. The fully connected network can consist of multiple linear layers and non-linear activation functions (such as ReLU), progressively extracting high-level abstract features through forward propagation. Finally, a fixed-dimensional vector is output as the encoded representation of the positive phase data. For instance, assuming the positive phase data has 100 feature dimensions, after passing through three fully connected layers (dimensions 100->64, 64->32, 32->16 respectively), a 16-dimensional feature vector is output.

[0028] Specifically, the input to a fully connected network can include numerical features, categorical features, and temporal features. The numerical features include price, quantity, duration, etc., the categorical features include user level, product category, etc., and the temporal features include the time series of the most recent N transactions, etc. The output is a fixed-dimensional feature vector, which can contain key semantic information of the positive phase data.

[0029] In addition, the inverse data encoder refers to a neural network module specifically designed to process unstructured or semi-structured auxiliary data streams with relatively low confidence levels and variable update frequencies. The core function of this encoder is to extract meaningful semantic features from complex unstructured data, such as user review text, product images, and browsing behavior sequences.

[0030] First, the inverted text data is preprocessed, including word segmentation and conversion to BERT input format. Specifically, [CLS] and [SEP] tags can be added, and corresponding token IDs and attention masks can be generated. Then, the processed data is input into a pre-trained BERT model, and the contextual representation of each token is obtained through the forward propagation of the BERT model. This invention takes the output vector corresponding to the [CLS] tag as the semantic representation of the entire text. A fully connected layer can be added after the BERT model to adjust the dimension, so that its output dimension is consistent with the output dimension of the positive data encoder. For example, the [CLS] vector output by the BERT model is 768-dimensional, which can be mapped to 16 dimensions through a fully connected layer, thereby aligning it with the 16-dimensional vector output by the positive data encoder. The input to this BERT model can include text sequences, image data, and behavior sequences; the text sequence can refer to user comments, the image data can refer to product images, and the behavior sequence can refer to clickstream logs. The output is a feature vector with the same dimension as the output of the positive encoder, thus achieving alignment of the feature space. The fully connected network and BERT model in this embodiment are merely examples of this embodiment. Other neural networks can also be used for training, and this embodiment does not impose excessive limitations on them.

[0031] In this embodiment of the invention, the forward-phase data encoder and the backward-phase data encoder are core concepts of the dual-data-stream model. They respectively undertake the feature extraction tasks for different types of data: the forward-phase encoder processes structured data through a fully connected network, focusing on modeling numerical relationships, statistical features, and business logic; the backward-phase encoder processes unstructured data through pre-trained models such as BERT, focusing on semantic understanding, contextual analysis, and sentiment mining. The output features of both are dimensionally aligned, laying the foundation for subsequent two-phase exchange coefficient calculation and feature fusion. The architecture design of this embodiment of the invention retains the characteristics of their respective data types while achieving a unified representation of data from different modalities.

[0032] Step 104: Calculate the two-phase exchange coefficients of the original e-commerce data; In this embodiment of the invention, calculating the two-phase exchange coefficients of the original e-commerce data includes: obtaining the positive-phase vector features output by the positive-phase data encoder; obtaining the negative-phase vector features output by the negative-phase data encoder; normalizing and reducing the dimensions of the positive-phase vector features and the negative-phase vector features to obtain an attention weight matrix; and determining the coefficients corresponding to the attention weight matrix as the two-phase exchange coefficients.

[0033] Specifically, the two-phase exchange coefficient can be a quantitative indicator measuring the degree and direction of information interaction between positive and negative data streams in the context of intelligent e-commerce data fusion. More specifically, the two-phase exchange coefficient is a dynamic weighting parameter that determines the degree to which negative auxiliary data (such as user comment sentiment) corrects and enhances positive master data (such as transaction records) in a specific business scenario. This two-phase exchange coefficient is not a fixed value but is dynamically calculated based on real-time data characteristics, reflecting the relativity and credibility of the two types of data sources at the current moment. In e-commerce data fusion, the two-phase exchange coefficient determines the magnitude of the correction force of negative data on positive data.

[0034] When predicting future sales of a product, if the product has recently received a large number of negative reviews (i.e., negative data), even if historical sales data (i.e., positive data) is good, a high exchange coefficient is needed to significantly reduce the predicted sales value due to the negative sentiment characteristics of the reviews. Conversely, if the review data is of poor quality or irrelevant to the current context, the exchange coefficient will automatically decrease, reducing its impact on the main data.

[0035] In one specific example, feature representations are first extracted from the pre-trained dual encoders. The positive data encoder outputs a 16-dimensional feature vector, such as [0.32, -0.15, 0.78, ...], representing structured features such as the user's current purchasing power and consumption preferences. The negative data encoder outputs an equally 16-dimensional vector, such as [0.45, 0.22, -0.31, ...], representing semantic features such as the sentiment of user comments and keyword density.

[0036] To avoid calculation errors caused by differences in the dimensions and numerical ranges of different features, it is necessary to normalize the two vectors. In this embodiment of the invention, L2 normalization can be used, which involves dividing the value of each dimension of each vector by its magnitude, i.e., the square root of the sum of the squares of all dimensions.

[0037] For example, if the positive phase vector is [3, 4], its magnitude is √(3²+4²)=5, which becomes [0.6,0.8] after normalization; similarly, the negative phase vector [1, 2] is normalized to [0.447, 0.894]. This process eliminates the influence of vector length, making subsequent similarity calculations entirely based on direction rather than magnitude.

[0038] In a preferred embodiment, for batch data, the entire feature matrix is ​​batch normalized to ensure that the data distribution of each batch is stable.

[0039] Furthermore, although the dimensions of the positive and negative feature vectors are aligned, as in the 16-dimensional array mentioned above, dimensionality reduction is required to further extract core features and reduce computational complexity. This can be achieved through Principal Component Analysis (PCA) or a learnable linear transformation layer.

[0040] Specifically, the PCA method can be used to reduce 16-dimensional features to 8-dimensional features. This involves calculating the covariance matrix of the feature vectors, finding the 8 principal component directions with the highest contribution rates, and projecting the original vectors onto these directions. For example, the original positive phase vectors may contain multiple related features such as "purchase frequency," "average order value," and "discount sensitivity," which can be integrated into comprehensive dimensions such as "consumer activity" and "price sensitivity" through dimensionality reduction.

[0041] Similarity scores are calculated using dot product operations on the dimensionality-reduced positive and negative features. Each positive feature dimension is paired with each negative feature dimension to form an attention weight matrix.

[0042] In a specific example, the dimensionality-reduced positive features are 8-dimensional vectors Q=[q1, q2, ..., q8]; the negative features are 8-dimensional vectors K=[k1, k2, ..., k8]; the attention scores are calculated as follows: for each qᵢ, the dot product with all kⱼ is calculated, i=1, 2...8, j=1, 2...8, resulting in 8×8=64 initial scores; the scores in each row are converted into a probability distribution using the softmax function: weight Qᵢⱼ = exp(qᵢ·kⱼ) / Σⱼexp(qᵢ·kⱼ); In this embodiment of the invention, the final two-phase exchange coefficients are key parameters extracted from the attention weight matrix. This can be done in the following way: First, calculate the average value of all elements in the attention weight matrix and use this average value as the two-phase exchange coefficient; alternatively, the main diagonal elements or the top k largest weight values ​​of the attention weight matrix can be used to form a set of two-phase exchange coefficients; this embodiment of the invention does not impose too many restrictions on this.

[0043] For example, in an e-commerce recommendation scenario, a user's positive data, specifically their historical purchase records, includes features such as "frequently purchased categories" and "purchase time period," while their negative data, specifically their browsing and review behavior, includes features such as "paying attention to negative reviews" and "reading depth." An attention weight matrix shows that the "frequently purchased categories" feature has a high-weighted connection with the "paying attention to negative reviews" feature (e.g., 0.7), indicating that users pay particular attention to negative reviews of frequently purchased products; while the "purchase time period" feature has a low-weighted connection with the "reading depth" feature (e.g., 0.1), indicating that browsing time and review reading depth are not significantly related.

[0044] In this embodiment of the invention, the exchange coefficient changes in real time with the input data. The interaction intensity of positive and negative data for the same product is automatically adjusted at different times and under different market conditions. This improves noise robustness. Compared with simple feature splicing, the attention mechanism avoids dimensionality explosion through weighted fusion and can be efficiently implemented through matrix operations, thus achieving computational efficiency.

[0045] Step 105: Correct the basic dual-data-stream model according to the two-phase exchange coefficients to obtain the final dual-data-stream model; In a practical application, the step of modifying the basic dual-datastream model based on the two-phase exchange coefficients to obtain the final dual-datastream model includes: The positive-phase vector features and the negative-phase vector features are combined to generate fused features; the two-phase exchange coefficients are set as the initial coefficients of the fused features to obtain the modified fused features; The positive and negative data encoders are trained using the modified fusion features to obtain the final dual data stream model.

[0046] Specifically, the first step is to combine the feature vectors output by the positive-phase data encoder and the negative-phase data encoder to generate preliminary fused features. In the feature fusion layer, the negative-phase feature vector is added to the positive-phase feature vector with the weights of the two-phase exchange coefficients, forming a new, corrected fused feature vector, thus obtaining the corrected fused features. In a specific example of this invention, the calculated two-phase exchange coefficients can be an attention weight matrix, where each element αᵢⱼ represents the correlation strength between the i-th dimension of the positive-phase feature and the j-th dimension of the negative-phase feature. A coefficient composed of multiple weighted combinations of αᵢⱼ can be assigned to each dimension of the final fused feature. A mapping from coefficients to weights is established by weighting the positive-phase feature vectors and the negative-phase feature vectors. The combined weights are directly determined by the corresponding row or column vectors in the exchange coefficient matrix. For example, the k-th dimension of the fused feature can be modified by combining 70% of the k-th dimension of the positive-phase feature and 30% of the k-th dimension of the negative-phase feature. This 70% and 30% ratio is the initial coefficient, which can be the normalized ratio of the sum of the attention weights of the k-th dimension of the positive-phase feature with respect to all negative-phase dimensions to the sum of the attention weights of the k-th dimension of the negative-phase feature with respect to all positive-phase dimensions in the attention weight matrix. This invention does not impose excessive limitations on this.

[0047] Furthermore, when the two features are structurally recombined to generate the modified fusion feature, a new fusion dimension is specifically created. The value of this fusion dimension is mainly derived from the weighted calculation of the two highly related original dimensions. Through the above mapping and recombination, the positive phase feature vector and the negative phase feature vector are combined linearly or nonlinearly to obtain the final modified fusion feature. Each dimension of the feature inherently encodes a specific interaction pattern between the positive and negative phase data streams.

[0048] After obtaining the corrected fusion features, the forward and backward data encoders are trained from this starting point. Using the backpropagation algorithm, the parameters of both encoders are updated simultaneously, training both the forward and backward data encoders to obtain the final dual-data-stream model. The entire training process is performed iteratively on a large number of samples. In each iteration, the two-phase exchange coefficients are dynamically calculated based on the new data, guiding the new round of feature fusion and encoder parameter updates.

[0049] Step 106: Input the new raw e-commerce data into the final dual data stream model to obtain feature output data; Step 107: Input the feature output data into the data turbulence model to obtain simulation fusion data.

[0050] In a preferred embodiment of the present invention, the step of inputting the feature output data into a data turbulence model to obtain simulation fusion data includes: performing a random perturbation injection operation on the feature output data to obtain first feature data; performing a Dropout operation on the first feature data to obtain second feature data; generating adversarial examples based on the feature output data; combining the adversarial examples with the second feature data to obtain third feature data; and inputting the third feature data into a task classification network to obtain predicted simulation fusion data.

[0051] The data turbulence model in this embodiment of the invention implements a multi-layered processing procedure that simulates data fluctuations and uncertainties in a real business environment. Its core is to introduce controlled randomness and adversarial perturbations, allowing the model to undergo sufficient testing during the training phase, thereby improving its generalization ability and robustness in complex real-world scenarios.

[0052] The data turbulence model in this embodiment of the invention includes a data turbulence layer, which may include a random perturbation injection operation sublayer, a Dropout application operation sublayer, and an adversarial example sublayer. Specifically, in the random perturbation injection operation sublayer, for the feature vector output by the dual data stream model, a random noise vector of the same dimension and conforming to a Gaussian distribution can be generated. The intensity of this noise is an adjustable hyperparameter. The original feature vector and the noise vector are added element by element to obtain the first feature data. In the Dropout application operation sublayer, in order to prevent the model from becoming overly dependent on certain specific feature dimensions, the output values ​​of a certain part of the feature vector can be temporarily forced to zero according to a preset probability. By repeatedly and randomly masking different feature subsets in multiple rounds of training, the model is forced to learn a more balanced and robust feature combination representation, avoiding the fragility of relying on a single strong feature for decision-making, thus obtaining the second feature data.

[0053] In the adversarial example sublayer, features can be perturbed with a very small step size to construct adversarial examples. Then, the original second feature data is linearly interpolated or concatenated with this adversarial example to form third feature data. This third feature data is input into a task classification network, such as a multilayer perceptron, to obtain predicted simulated fusion data. For example, in a product recommendation task, inputting a user's fusion features, the task classification network ultimately outputs a recommendation score that has been stability-corrected and is meaningful for multiple similar user groups. This recommendation score is the simulated fusion data used to drive business decisions. Through training throughout the entire turbulent process, the model ensures that its final output maintains high stability and accuracy when facing data noise, missing features, or potential adversarial interference.

[0054] In this embodiment of the invention, the intelligent e-commerce data processing method based on multi-source data fusion includes: acquiring multiple types of raw e-commerce data; and dividing the raw e-commerce data into positive-phase data and negative-phase data. A basic dual-dataflow model is established using the positive-phase and negative-phase data; the two-phase exchange coefficients of the original e-commerce data are calculated; the basic dual-dataflow model is modified according to the two-phase exchange coefficients to obtain the final dual-dataflow model; new original e-commerce data is input into the final dual-dataflow model to obtain feature output data; the feature output data is input into the data turbulence model to obtain simulated fusion data. The two-phase exchange coefficients enable dynamic adjustment of data fusion weights according to the context, overcoming the limitations of static fusion. The robust design of the entire process from data partitioning and feature fusion to final output significantly improves the system's performance in complex real-world environments. While maintaining the high performance of the deep learning model, the attention mechanism provides multifaceted interpretability, improving the model's inference performance.

[0055] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0056] Reference Figure 2 This diagram illustrates a structural block diagram of an embodiment of an intelligent e-commerce data processing system based on multi-source data fusion according to an embodiment of the present invention, which may specifically include the following modules: The first acquisition module 301 is used to acquire various raw e-commerce data; The segmentation module 302 is used to divide the original e-commerce data into positive phase data and negative phase data; Module 303 is used to establish a basic dual-datastream model using the positive-phase data and the negative-phase data; Calculation module 304 is used to calculate the two-phase exchange coefficients of the original e-commerce data; The correction module 305 is used to correct the basic dual data stream model according to the two-phase exchange coefficients to obtain the final dual data stream model. Input module 306 is used to input new raw e-commerce data into the final dual data stream model to obtain feature output data; The simulation fusion data module 307 is used to input the feature output data into the data turbulence model to obtain simulation fusion data.

[0057] Preferably, the partitioning module includes: The identification submodule is used to identify the data structure type, source stability parameters, and generation frequency of the original e-commerce data. The classification submodule is used to classify the original e-commerce data into positive phase data and negative phase data according to the data structure type, source stability parameters and generation frequency.

[0058] Preferably, the establishment module includes: The first construction submodule is used to input the positive phase data into a fully connected network to obtain the constructed positive phase data encoder; The second construction submodule is used to input the inverted data into the BERT model to obtain the constructed inverted data encoder; The component submodule is used to form a basic dual-data-stream model using the positive-phase data encoder and the negative-phase data encoder.

[0059] Preferably, the computing module includes: The first acquisition submodule is used to acquire the positive phase vector features output by the positive phase data encoder; The second acquisition submodule is used to acquire the inverted vector features output by the inverted data encoder; The dimension reduction submodule is used to normalize and reduce the dimension of the positive phase vector features and the negative phase vector features to obtain the attention weight matrix. The determination submodule is used to determine the coefficients corresponding to the attention weight matrix as two-phase exchange coefficients.

[0060] Preferably, the correction module includes: A generation submodule is used to combine the positive phase vector features and the negative phase vector features to generate fused features; A submodule is configured to set the two-phase exchange coefficients as the initial coefficients of the fusion feature to obtain the modified fusion feature; The training submodule is used to train the positive phase data encoder and the negative phase data encoder through the modified fusion features to obtain the final dual data stream model.

[0061] Preferably, the simulation fusion data module includes: The first feature data submodule is used to perform a random perturbation injection operation on the feature output data to obtain the first feature data. The second feature data submodule is used to perform Dropout operation on the first feature data to obtain the second feature data. The third feature data submodule is used to generate adversarial samples based on the feature output data, and to obtain the third feature data by combining the adversarial samples with the second feature data. The third feature data is input into the task classification network to obtain the predicted simulation fusion data.

[0062] Preferably, the classification submodule includes: The first identification unit is used to identify whether the data structure type of the original e-commerce data is structured data or unstructured data, and to obtain the first type of data and the second type of data. The second identification unit is used to identify the confidence levels of the first type of data and the second type of data; The third identification unit is used to identify the generation frequency of the first type of data and the second type of data; The filtering unit is used to filter out data in the first type of data whose confidence level and generation frequency are higher than a preset threshold, and to identify such data as positive phase data; and to filter out data in the second type of data whose confidence level and generation frequency are lower than a preset threshold, and to identify such data as negative phase data.

[0063] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0064] Specific limitations regarding the intelligent e-commerce data processing system based on multi-source data fusion can be found in the limitations of the intelligent e-commerce data processing method based on multi-source data fusion mentioned above, and will not be repeated here. Each module in the aforementioned intelligent e-commerce data processing system based on multi-source data fusion can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0065] The intelligent e-commerce data processing system based on multi-source data fusion provided above can be used to execute the intelligent e-commerce data processing method based on multi-source data fusion provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0066] In one embodiment, a computer device is provided, which may be a control terminal corresponding to an atomizing air pump, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent e-commerce data processing method based on multi-source data fusion. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0067] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0068] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: We obtained various types of raw e-commerce data; The original e-commerce data is divided into positive phase data and negative phase data; A basic dual-data-stream model is established using the positive-phase data and the negative-phase data; Calculate the two-phase exchange coefficients of the original e-commerce data; The basic dual-data-stream model is modified based on the two-phase exchange coefficients to obtain the final dual-data-stream model; The new raw e-commerce data is input into the final dual data stream model to obtain feature output data; The feature output data is input into the data turbulence model to obtain simulation fusion data.

[0069] Preferably, dividing the original e-commerce data into positive-phase data and negative-phase data includes: Identify the data structure type, source stability parameters, and generation frequency of the original e-commerce data; Based on the data structure type, source stability parameters, and generation frequency, the original e-commerce data is classified into positive phase data and negative phase data.

[0070] Preferably, establishing a basic dual-datastream model using the positive-phase data and the negative-phase data includes: The positive phase data is input into a fully connected network to obtain the constructed positive phase data encoder; The inverted data is input into the BERT model to obtain the constructed inverted data encoder; The positive-phase data encoder and the negative-phase data encoder together form a basic dual-data-stream model.

[0071] Preferably, the calculation of the two-phase exchange coefficients of the original e-commerce data includes: Obtain the positive phase vector features output by the positive phase data encoder; Obtain the inverted vector features output by the inverted data encoder; The positive and negative phase vector features are normalized and dimensionality reduced to obtain the attention weight matrix. The coefficients corresponding to the attention weight matrix are determined as two-phase exchange coefficients.

[0072] Preferably, the step of modifying the basic dual-datastream model according to the two-phase exchange coefficients to obtain the final dual-datastream model includes: The positive-phase vector features and the negative-phase vector features are combined to generate fused features; the two-phase exchange coefficients are set as the initial coefficients of the fused features to obtain the modified fused features; The positive and negative data encoders are trained using the modified fusion features to obtain the final dual data stream model.

[0073] Preferably, the step of inputting the feature output data into the data turbulence model to obtain simulation fusion data includes: The feature output data is subjected to random perturbation injection to obtain the first feature data; The first feature data is subjected to Dropout to obtain the second feature data; Adversarial samples are generated based on the feature output data, and third feature data is obtained by combining the adversarial samples with the second feature data. The third feature data is input into the task classification network to obtain the predicted simulation fusion data.

[0074] Preferably, classifying the original e-commerce data into positive-phase data and negative-phase data based on the data structure type, source stability parameters, and generation frequency includes: The data structure type of the original e-commerce data is identified as structured data or unstructured data, resulting in a first type of data and a second type of data; Identify the confidence levels of the first type of data and the second type of data; Identify the generation frequencies of the first type of data and the second type of data; Data from the first type of data with a confidence level and generation frequency higher than a preset threshold are selected and identified as positive phase data; data from the second type of data with a confidence level and generation frequency lower than a preset threshold are selected and identified as negative phase data.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs the following steps: We obtained various types of raw e-commerce data; The original e-commerce data is divided into positive phase data and negative phase data; A basic dual-data-stream model is established using the positive-phase data and the negative-phase data; Calculate the two-phase exchange coefficients of the original e-commerce data; The basic dual-data-stream model is modified based on the two-phase exchange coefficients to obtain the final dual-data-stream model; The new raw e-commerce data is input into the final dual data stream model to obtain feature output data; The feature output data is input into the data turbulence model to obtain simulation fusion data.

[0076] Preferably, dividing the original e-commerce data into positive-phase data and negative-phase data includes: Identify the data structure type, source stability parameters, and generation frequency of the original e-commerce data; Based on the data structure type, source stability parameters, and generation frequency, the original e-commerce data is classified into positive phase data and negative phase data.

[0077] Preferably, establishing a basic dual-datastream model using the positive-phase data and the negative-phase data includes: The positive phase data is input into a fully connected network to obtain the constructed positive phase data encoder; The inverted data is input into the BERT model to obtain the constructed inverted data encoder; The positive-phase data encoder and the negative-phase data encoder together form a basic dual-data-stream model.

[0078] Preferably, the calculation of the two-phase exchange coefficients of the original e-commerce data includes: Obtain the positive phase vector features output by the positive phase data encoder; Obtain the inverted vector features output by the inverted data encoder; The positive and negative phase vector features are normalized and dimensionality reduced to obtain the attention weight matrix. The coefficients corresponding to the attention weight matrix are determined as two-phase exchange coefficients.

[0079] Preferably, the step of modifying the basic dual-datastream model according to the two-phase exchange coefficients to obtain the final dual-datastream model includes: The positive-phase vector features and the negative-phase vector features are combined to generate fused features; the two-phase exchange coefficients are set as the initial coefficients of the fused features to obtain the modified fused features; The positive and negative data encoders are trained using the modified fusion features to obtain the final dual data stream model.

[0080] Preferably, the step of inputting the feature output data into the data turbulence model to obtain simulation fusion data includes: The feature output data is subjected to random perturbation injection to obtain the first feature data; The first feature data is subjected to Dropout to obtain the second feature data; Adversarial samples are generated based on the feature output data, and third feature data is obtained by combining the adversarial samples with the second feature data. The third feature data is input into the task classification network to obtain the predicted simulation fusion data.

[0081] Preferably, classifying the original e-commerce data into positive-phase data and negative-phase data based on the data structure type, source stability parameters, and generation frequency includes: The data structure type of the original e-commerce data is identified as structured data or unstructured data, resulting in a first type of data and a second type of data; Identify the confidence levels of the first type of data and the second type of data; Identify the generation frequencies of the first type of data and the second type of data; Data from the first type of data with a confidence level and generation frequency higher than a preset threshold are selected and identified as positive phase data; data from the second type of data with a confidence level and generation frequency lower than a preset threshold are selected and identified as negative phase data.

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of apparatus, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction methods implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0089] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or terminal device that includes said element.

[0090] The present invention has provided a detailed description of an intelligent e-commerce data processing method and system based on multi-source data fusion, a computer device, and a storage medium. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A smart e-commerce data processing method based on multi-source data fusion, characterized in that, include: We obtained various types of raw e-commerce data; The original e-commerce data is divided into positive phase data and negative phase data; A basic dual-data-stream model is established using the positive-phase data and the negative-phase data; Calculate the two-phase exchange coefficients of the original e-commerce data; The basic dual-data-stream model is modified based on the two-phase exchange coefficients to obtain the final dual-data-stream model; The new raw e-commerce data is input into the final dual data stream model to obtain feature output data; The feature output data is input into the data turbulence model to obtain simulation fusion data.

2. The method according to claim 1, characterized in that, The process of dividing the original e-commerce data into positive-phase data and negative-phase data includes: Identify the data structure type, source stability parameters, and generation frequency of the original e-commerce data; Based on the data structure type, source stability parameters, and generation frequency, the original e-commerce data is classified into positive phase data and negative phase data.

3. The method according to claim 1, characterized in that, The establishment of a basic dual-datastream model using the positive-phase and negative-phase data includes: The positive phase data is input into a fully connected network to obtain the constructed positive phase data encoder; The inverted data is input into the BERT model to obtain the constructed inverted data encoder; The positive-phase data encoder and the negative-phase data encoder together form a basic dual-data-stream model.

4. The method according to claim 1, characterized in that, The calculation of the two-phase exchange coefficients of the original e-commerce data includes: Obtain the positive phase vector features output by the positive phase data encoder; Obtain the inverted vector features output by the inverted data encoder; The positive and negative phase vector features are normalized and dimensionality reduced to obtain the attention weight matrix. The coefficients corresponding to the attention weight matrix are determined as two-phase exchange coefficients.

5. The method according to claim 1, characterized in that, The step of modifying the basic dual-datastream model based on the two-phase exchange coefficients to obtain the final dual-datastream model includes: The positive-phase vector features and the negative-phase vector features are combined to generate fused features; the two-phase exchange coefficients are set as the initial coefficients of the fused features to obtain the modified fused features; The positive and negative data encoders are trained using the modified fusion features to obtain the final dual data stream model.

6. The method according to claim 1, characterized in that, The step of inputting the feature output data into the data turbulence model to obtain simulation fusion data includes: The feature output data is subjected to random perturbation injection to obtain the first feature data; The first feature data is subjected to Dropout to obtain the second feature data; Adversarial samples are generated based on the feature output data, and third feature data is obtained by combining the adversarial samples with the second feature data. The third feature data is input into the task classification network to obtain the predicted simulation fusion data.

7. The method according to claim 1, characterized in that, The process of classifying the original e-commerce data into positive-phase data and negative-phase data based on the data structure type, source stability parameters, and generation frequency includes: The data structure type of the original e-commerce data is identified as structured data or unstructured data, resulting in a first type of data and a second type of data; Identify the confidence levels of the first type of data and the second type of data; Identify the generation frequencies of the first type of data and the second type of data; Data from the first type of data with a confidence level and generation frequency higher than a preset threshold are selected and identified as positive phase data; data from the second type of data with a confidence level and generation frequency lower than a preset threshold are selected and identified as negative phase data.

8. An intelligent e-commerce data processing system based on multi-source data fusion, characterized in that, include: The first acquisition module is used to acquire various types of raw e-commerce data; The segmentation module is used to divide the original e-commerce data into positive phase data and negative phase data; A module is established to build a basic dual-datastream model using the positive-phase data and the negative-phase data; The calculation module is used to calculate the two-phase exchange coefficients of the original e-commerce data; The correction module is used to correct the basic dual-data-stream model according to the two-phase exchange coefficients to obtain the final dual-data-stream model. The input module is used to input new raw e-commerce data into the final dual data stream model to obtain feature output data; The simulation fusion data module is used to input the feature output data into the data turbulence model to obtain simulation fusion data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent e-commerce data processing method based on multi-source data fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent e-commerce data processing method based on multi-source data fusion as described in any one of claims 1 to 7.

Images