Marketing data processing method and platform based on large model

Abstract

The invention provides a marketing data processing method based on a large model. According to the method, marketing behavior data of a target user is obtained, feature extraction is carried out on the marketing behavior data to form a first feature vector set, and if a preset identification vector of which the similarity with a first identification vector is greater than a preset similarity threshold exists in a preset identification vector mapping table, the target user is identified. If yes, processing the first feature vector set by using a preset feature vector processing model to generate a second feature vector set, and then determining the user feature behavior corresponding to the target user by using a user behavior recognition model according to the second feature vector set, thereby reducing the model calculation complexity, and improving the user experience by optimizing the feature combination. The expression ability and the recognition precision of the feature vectors are improved, and accurate capture of the user behavior features is realized.

Description

Technical Field

[0001] The present application relates to data processing technology, and in particular to a marketing data processing method and platform based on a large model. Background Art

[0002] In the current digital marketing field, companies are faced with massive and complex user behavior data. These data not only come from a variety of channels (such as social media, email, website browsing, etc.), but also have highly unstructured and semi-structured characteristics. In order to effectively analyze and utilize these data to formulate accurate marketing strategies, companies need to rely on advanced data processing and analysis technologies.

[0003] Traditional marketing data processing methods usually directly analyze raw user behavior data. The current processing method has high computational complexity, especially when processing large-scale data sets, which leads to low computational efficiency. Summary of the invention

[0004] The present application provides a marketing data processing method and platform based on a large model to reduce the complexity of model calculation and improve model processing efficiency.

[0005] In a first aspect, the present application provides a marketing data processing method based on a large model, comprising: Acquire marketing behavior data of target users, and perform feature extraction on the marketing behavior data to form a first feature vector set, wherein the identification vector in the first feature vector set is a first identification vector, and the feature dimension of the first feature vector set is a first dimension; If there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed by using a preset feature vector processing model to generate a second feature vector set, wherein the feature dimension of the second feature vector set is a second dimension, and the second dimension is smaller than the first dimension; A user behavior recognition model is used, and user characteristic behaviors corresponding to the target user are determined according to the second feature vector set.

[0006] In the above scheme, the marketing behavior data of the target user is obtained and features are extracted from the marketing behavior data to form a first feature vector set. If a preset identification vector mapping table contains a preset identification vector whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed using a preset feature vector processing model to generate a second feature vector set. Then, the user behavior recognition model is used to determine the user characteristic behavior corresponding to the target user based on the second feature vector set, thereby reducing the model calculation complexity. By optimizing the feature combination, the expression ability and recognition accuracy of the feature vector are improved, thereby achieving accurate capture of user behavior characteristics.

[0007] Optionally, processing the first feature vector set by using a preset feature vector processing model includes: Determine a set of associated vectors of the first identification vector by using a preset characteristic vector associated mapping table, wherein the preset characteristic vector associated mapping table is used to establish an associated relationship between each characteristic vector, wherein the associated relationship includes an associated direction and an associated weight value; The second feature vector set is determined according to the association vector set and the first feature vector set, wherein the second feature vector set is a difference set between the first feature vector set and the association vector set.

[0008] In the above scheme, a set of vectors associated with the first identification vector is identified and determined through a preset feature vector association mapping table. These associated vectors represent other repeated feature information related to the target user behavior. Therefore, by calculating the difference between the first feature vector set and the associated vector set, those associated vectors that already exist in the first feature vector set and may be redundant can be removed, thereby reducing the complexity and amount of calculation of data processing. The vectors retained in the first feature vector set but not in the associated vector set are unique feature information that is directly related to the target user behavior. After dimensionality reduction processing, these feature vectors can more accurately reflect the characteristic behavior of the target user, providing more valuable information for subsequent user behavior identification and marketing strategy formulation.

[0009] Optionally, the determining the association vector set of the first identification vector by using a preset characteristic vector association mapping table includes: Determine, according to the preset characteristic vector association mapping table, a first candidate vector set corresponding to the first identification vector, wherein an association relationship between candidate vectors in the first candidate vector set and the first identification vector has a first target association direction, and the first target association direction is used to represent that the candidate vectors in the first candidate vector set are directly associated with the first identification vector; According to the preset characteristic vector association mapping table, candidate vectors having direct association weight values ​​greater than preset association weight values ​​are determined from the first candidate vector set to add to a first association vector set, the association vector set including the first association vector set.

[0010] In the above scheme, according to the preset characteristic vector association mapping table, the first candidate vector set directly associated with the first identification vector can be determined. These candidate vectors represent the characteristic information directly related to the target user behavior. Through further screening, it can be ensured that only those vectors that have an important impact on the target user behavior identification are retained. By setting the preset association weight value, the candidate vectors with a direct association weight value greater than the threshold can be screened out from the first candidate vector set to form a first association vector set. This step avoids processing a large amount of unnecessary or redundant data, thereby improving the efficiency of data processing. Moreover, since the vectors in the first association vector set are all screened, they are more closely associated with the target user behavior. This means that in the subsequent user behavior identification and marketing strategy formulation process, these feature vectors can provide more accurate information.

[0011] Optionally, the determining the association vector set of the first identification vector by using a preset characteristic vector association mapping table further includes: Determine a second candidate vector set of the first identification vector according to the preset characteristic vector association mapping table, wherein the association relationship between the candidate vectors in the second candidate vector set and the first identification vector has a second target association direction, and the second target association direction is used to represent that the candidate vector is indirectly associated with the first identification vector; According to the preset characteristic vector association mapping table, a candidate vector whose indirect association weight value is greater than the preset association weight value is determined from the second candidate vector set to add it to the second association vector set, the association vector set including the second association vector set, and the indirect association weight value is the product of each association weight value in the association path from the candidate vector in the second candidate vector set to the first identification vector.

[0012] In the above scheme, by introducing the concept of indirect association, those candidate vectors that are not directly related to the target user behavior but are indirectly related to the target user behavior through other feature vectors can be identified. These vectors may contain additional and valuable information, which helps to more accurately understand the characteristic behavior of the target user. In the process of marketing data processing, only considering the directly related feature vectors may miss some important information. By introducing the concept of indirect association, the feature vector set can be supplemented and improved to improve the integrity and comprehensiveness of the feature vector. Although the vectors in the second candidate vector set are indirectly related to the first identification vector, by setting a preset association weight value, those candidate vectors whose indirect association weight values ​​are greater than the threshold can be screened out to form a second association vector set. The correlation between these vectors and the target user behavior is still very strong, and there is a large degree of duplication. Therefore, by using the difference set with the second association vector set, it is possible to further avoid processing a large amount of unnecessary or redundant data, thereby improving the efficiency of data processing.

[0013] Optionally, after determining the association vector set of the first identification vector by using the preset characteristic vector association mapping table, the method further includes: Determine the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table; The preset association weight value is determined according to the highest similarity and a preset initial association weight value, wherein.

[0014] In the above scheme, by introducing the highest similarity to dynamically adjust the preset association weight value, the feature vector closely related to the target user behavior can be more accurately screened. When the similarity between the first identification vector and a preset identification vector is very high, it means that the behavior of the target user is very close to the behavior of the user group represented by the preset identification vector, so the preset association weight value should be reduced so that more feature vectors related to the target user behavior can be excluded in the subsequent analysis. On the contrary, when the similarity is low, the preset association weight value should be increased to ensure that more feature vectors related to the target user behavior are included in the subsequent identification. The above scheme can dynamically adjust the preset association weight value according to the marketing behavior data of different target users, thereby improving the adaptability and flexibility of the method. This helps to perform more accurate and personalized marketing data processing and analysis according to the characteristics and needs of different user groups in practical applications. In addition, by dynamically adjusting the preset association weight value, the data processing efficiency can be optimized while ensuring the accuracy of data processing. When the similarity is high, by reducing the preset association weight value, more feature vectors can be included for subsequent analysis, thereby reducing unnecessary feature vector analysis, thereby improving data processing efficiency. When the similarity is low, by increasing the preset association weight value, more feature vectors can be included in the subsequent analysis, thereby increasing the accuracy of recognition and helping to capture the characteristic behaviors of target users more comprehensively.

[0015] Optionally, before the step of processing the first feature vector set by using a preset feature vector processing model to generate a second feature vector set, the method further includes: It is determined that the first dimension corresponding to the first feature vector set is greater than a preset dimension threshold.

[0016] In the above scheme, by judging whether the dimension of the first feature vector set is greater than the preset dimension threshold, it is possible to screen out those user behavior data with higher feature dimensions and richer information. For feature vector sets with lower dimensions and less information content, subsequent processing or other processing methods can be directly performed, thereby avoiding unnecessary calculations and waste of resources, and improving the pertinence and efficiency of data processing. When the dimension of the first feature vector set is higher, it often means that the set contains more user behavior feature information. At this time, using the preset feature vector processing model to process it can more effectively extract feature vectors closely related to the target user behavior, form a second feature vector set, and then help optimize the feature vector processing efficiency. It can be seen that by setting a preset dimension threshold, the input feature vector set can be preliminarily screened and filtered, which can avoid poor processing effects or errors caused by too low dimensions or insufficient information content, and at the same time, when the dimension is too high, it can also improve the processing speed and efficiency through dimensionality reduction processing.

[0017] Optionally, after the first feature vector set is processed by using a preset feature vector processing model to generate a second feature vector set, the method further includes: If the feature difference ratio between the second dimension and the first dimension is greater than a preset feature difference ratio threshold, the first feature vector set and the second feature vector set are added to the recognition model training database, and the recognition model training database is used to iteratively train the user behavior recognition model.

[0018] In the above scheme, by continuously adding new feature vector sets (including the first feature vector set and the second feature vector set) to the recognition model training database and using them for iterative training of the model, the model's training sample library can be continuously enriched and optimized. This helps to improve the model's recognition accuracy and generalization ability for user behavior, so that it can better adapt to different application scenarios and user groups. In actual applications, user behavior data may change with time and environmental changes. By continuously iteratively training the model, the model can continuously adapt to new data distributions and feature patterns, thereby enhancing the robustness and stability of the model. This helps to ensure that the model always maintains high recognition performance and stability during long-term use.

[0019] It is worth noting that by setting the feature difference ratio between the second dimension and the first dimension to be greater than the preset feature difference ratio threshold as a condition for adding the corresponding data to the model training database, it is possible to screen out those data sets whose features change significantly during the processing process. These data sets often contain richer and more diverse user behavior information, which is of great significance for improving the recognition ability of the model. Setting the preset feature difference ratio threshold as a screening condition can ensure that only those data sets with sufficiently large feature changes will be added to the model training database for subsequent iterative training, which helps to avoid the introduction of redundant or low-quality data, thereby optimizing the quality of training samples, and thus improving the pertinence and efficiency of subsequent iterative training. In addition, due to the diversity and complexity of user behavior data, traditional fixed models are often difficult to adapt to all situations. By continuously introducing new data sets with significant feature changes for iterative training, the model can continuously learn and adapt to new data distributions and feature patterns, thereby helping to improve the adaptability and robustness of the model when processing complex and variable user behavior data, so that it can better meet the needs of practical applications. In addition, by setting the feature difference ratio threshold, unnecessary training samples can be reduced, thereby reducing the training cost and time of the model, helping to improve the training efficiency of the model, so that it can adapt to new data and needs more quickly. At the same time, since the screened data set is of high quality, it can also reduce the risk of overfitting of the model during training and further improve the generalization ability of the model.

[0020] It is worth noting that the above-mentioned first feature vector set can also be added to the verification set in the recognition model training database, and the second feature vector set can be added to the training set in the recognition model training database. By using the feature vector sets before and after processing for training and verification respectively, data resources can be used more efficiently. Adding the first feature vector set and the second feature vector set to the verification set and the training set respectively provides rich data support for the iterative optimization of the model. By using the verification set to evaluate the performance of the model, we can promptly discover the instability that may occur in the model during the training process, such as performance fluctuations, overfitting, etc. At the same time, adding the processed feature vector set (the second feature vector set) to the training set can further enrich the training samples of the model and improve the stability and robustness of the model.

[0021] In a second aspect, the present application provides a marketing data processing platform based on a large model, comprising: an acquisition module, configured to acquire marketing behavior data of a target user, and perform feature extraction on the marketing behavior data to form a first feature vector set, wherein an identification vector in the first feature vector set is a first identification vector, and a feature dimension of the first feature vector set is a first dimension; a processing module, configured to process the first feature vector set using a preset feature vector processing model to generate a second feature vector set when there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, wherein the feature dimension of the second feature vector set is a second dimension, and the second dimension is smaller than the first dimension; A determination module is used to use a user behavior recognition model to determine the user characteristic behavior corresponding to the target user according to the second feature vector set.

[0022] Optionally, the processing module is specifically used to: Determine a set of associated vectors of the first identification vector by using a preset characteristic vector associated mapping table, wherein the preset characteristic vector associated mapping table is used to establish an associated relationship between each characteristic vector, wherein the associated relationship includes an associated direction and an associated weight value; The second feature vector set is determined according to the association vector set and the first feature vector set, wherein the second feature vector set is a difference set between the first feature vector set and the association vector set.

[0023] Optionally, the processing module is specifically used to: Determine, according to the preset characteristic vector association mapping table, a first candidate vector set corresponding to the first identification vector, wherein an association relationship between candidate vectors in the first candidate vector set and the first identification vector has a first target association direction, and the first target association direction is used to represent that the candidate vectors in the first candidate vector set are directly associated with the first identification vector; According to the preset characteristic vector association mapping table, candidate vectors having direct association weight values ​​greater than preset association weight values ​​are determined from the first candidate vector set to add to a first association vector set, the association vector set including the first association vector set.

[0024] Optionally, the processing module is specifically used to: Determine a second candidate vector set of the first identification vector according to the preset characteristic vector association mapping table, wherein the association relationship between the candidate vectors in the second candidate vector set and the first identification vector has a second target association direction, and the second target association direction is used to represent that the candidate vector is indirectly associated with the first identification vector; According to the preset characteristic vector association mapping table, a candidate vector whose indirect association weight value is greater than the preset association weight value is determined from the second candidate vector set to add it to the second association vector set, the association vector set including the second association vector set, and the indirect association weight value is the product of each association weight value in the association path from the candidate vector in the second candidate vector set to the first identification vector.

[0025] Optionally, the processing module is specifically used to: Determine the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table; The preset association weight value is determined according to the highest similarity and a preset initial association weight value, wherein the highest similarity and the preset association weight value are negatively correlated.

[0026] Optionally, the processing module is specifically used to: It is determined that the first dimension corresponding to the first feature vector set is greater than a preset dimension threshold.

[0027] Optionally, the processing module is specifically used to: If the feature difference ratio between the second dimension and the first dimension is greater than a preset feature difference ratio threshold, the first feature vector set and the second feature vector set are added to the recognition model training database, and the recognition model training database is used to iteratively train the user behavior recognition model.

[0028] In a third aspect, the present application provides an electronic device, including: processor; and, A memory, configured to store executable instructions of the processor; The processor is configured to perform any possible method described in the first aspect by executing the executable instructions.

[0029] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement any possible method described in the first aspect.

[0030] The marketing data processing method and platform based on a large model provided in the present application obtain the marketing behavior data of the target user and perform feature extraction on the marketing behavior data to form a first feature vector set. If a preset identification vector mapping table contains a preset identification vector whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed using a preset feature vector processing model to generate a second feature vector set. Then, a user behavior recognition model is used to determine the user characteristic behavior corresponding to the target user based on the second feature vector set, thereby reducing the model calculation complexity, and by optimizing the feature combination, the expression ability and recognition accuracy of the feature vector are improved, thereby achieving accurate capture of user behavior characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0032] Figure 1 is a flowchart of a marketing data processing method based on a large model according to an exemplary embodiment of the present application; Figure 2 is a flowchart of a marketing data processing method based on a large model according to another exemplary embodiment of the present application; Figure 3 is a structural diagram of a marketing data processing platform based on a large model according to an exemplary embodiment of the present application; Figure 4 It is a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present application.

[0033] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0035] In order to solve the above problems, the embodiment provided by the present application first obtains the marketing behavior data of the target user and extracts the features of the data to form a first feature vector set. This process aims to convert the original user behavior data into a numerical representation that can be understood by the machine, laying the foundation for subsequent processing and analysis. The system obtains the user's marketing behavior data from the data source, and uses feature extraction technology (such as text mining, image recognition, etc.) to convert unstructured or semi-structured data into structured data, forming a set containing multiple feature vectors, each of which represents a certain aspect of user behavior.

[0036] When there is a preset identification vector in the preset identification vector mapping table that has a high similarity with the first identification vector, the first feature vector set is processed using the preset feature vector processing model to generate a second feature vector set with a lower feature dimension. This step aims to reduce the complexity and amount of data processing through dimensionality reduction processing, while retaining the feature information most relevant to the target user behavior. The system compares the similarity between the first identification vector and the preset identification vector in the preset identification vector mapping table. When the similarity exceeds the preset threshold, the preset feature vector association mapping table is used to determine the associated vector set of the first identification vector. These associated vectors represent other repeated feature information related to the target user behavior. Then, by calculating the difference between the first feature vector set and the associated vector set, a second feature vector set is generated to achieve dimensionality reduction and optimization of the feature vector.

[0037] When determining the association vector set, not only the candidate vectors directly associated with the first identification vector (first candidate vector set) are considered, but also the candidate vectors indirectly associated with the first identification vector (second candidate vector set). This step is intended to more comprehensively identify the feature vectors related to the target user behavior and improve the integrity and comprehensiveness of the feature vectors. The system determines the first candidate vector set and the second candidate vector set respectively according to the preset feature vector association mapping table. For the first candidate vector set, the candidate vectors whose direct association weight value is greater than the preset association weight value are directly screened out. For the second candidate vector set, the indirect association weight value of the candidate vector indirectly associated with the first identification vector (that is, the product of each association weight value in the association path) is calculated, and the candidate vectors whose indirect association weight value is greater than the preset association weight value are screened out. These candidate vectors together constitute the association vector set.

[0038] After determining the set of association vectors, the preset association weight value is dynamically adjusted according to the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table. This step aims to dynamically optimize the feature vector processing process according to the marketing behavior data of different target users and improve the adaptability and flexibility of the method. The system first determines the highest similarity, and then calculates a new preset association weight value based on the negative correlation between the preset initial association weight value and the highest similarity. When the highest similarity is high, the preset association weight value is lowered to reduce redundant processing; when the highest similarity is low, the preset association weight value is increased to ensure the comprehensiveness of the identification. The adjusted preset association weight value will be used for subsequent feature vector screening and processing.

[0039] Then, the user behavior recognition model is used to determine the user characteristic behavior corresponding to the target user based on the processed second feature vector set. At the same time, the model is iteratively trained by continuously adding new feature vector sets to the recognition model training database to improve the recognition accuracy and generalization ability of the model. The system uses the trained user behavior recognition model to analyze the second feature vector set and identify the characteristic behavior of the target user. In addition, when the feature difference ratio between the second feature vector set and the original first feature vector set is greater than the preset threshold, the two feature vector sets are added to the training set and verification set in the recognition model training database for iterative training and optimization of the model. This step helps the model adapt to the ever-changing user behavior data and improve the accuracy and stability of recognition.

[0040] Figure 1 FIG. 1 is a flowchart of a marketing data processing method based on a large model according to an exemplary embodiment of the present application. Figure 1 As shown, the marketing data processing method based on the big model provided in this embodiment includes: S101. Obtain marketing behavior data of target users, and perform feature extraction on the marketing behavior data to form a first feature vector set.

[0041] In this step, the marketing behavior data of the target user is obtained, and features are extracted from the marketing behavior data to form a first feature vector set, wherein the identification vector in the first feature vector set is the first identification vector, and the feature dimension of the first feature vector set is the first dimension.

[0042] Specifically, firstly, the marketing behavior data of the target user is obtained through the data collection module. Such data may include but is not limited to the user's purchase history, browsing history, click history, search history, etc. The data collection module may be an internal database system of the enterprise or an external data collection platform.

[0043] After obtaining the marketing behavior data of the target user, the feature extraction module is used to process the data. The main task of the feature extraction module is to extract information that can reflect the characteristics of user behavior from the original data and convert it into a feature vector. In this step, the identification vector in the feature vector set is the first identification vector, which is used to identify a user or a type of user behavior. At the same time, the feature dimension of the first feature vector set is the first dimension, and the size of this dimension depends on the number of extracted features.

[0044] For text-based marketing behavior data (such as user comments, social media posts, etc.), natural language processing technology can be used for feature extraction. For example, the bag-of-words model, TF-IDF (Term Frequency-Inverse Document Frequency), or word embedding methods can be used to convert text into numerical feature vectors.

[0045] Optionally, for time series marketing behavior data (such as user purchase time series), statistical features (such as mean, standard deviation, maximum, minimum, etc.) and trend features (such as moving average, exponential smoothing, etc.) can be extracted. If the marketing behavior data contains image information (such as product images, photos uploaded by users, etc.), computer vision technology (such as convolutional neural networks) can be used to extract image features. For continuous user behavior sequences, recurrent neural networks or their variants (such as long short-term memory networks) can be used to extract sequence features and capture the dynamic changes in user behavior.

[0046] Then, according to business needs, the initially extracted features are combined and transformed to generate new composite features. For example, the average purchase interval and purchase frequency of users can be calculated.

[0047] The extracted features are represented as numerical vectors to form a first feature vector set. Each feature vector corresponds to the marketing behavior data of a user, and the dimension of the vector is equal to the number of extracted features.

[0048] In the first feature vector set, one or more identification vectors (first identification vectors) are set to uniquely identify each user or user group. These identification vectors can be the user's unique ID, user portrait label, etc. The feature dimension (first dimension) of the first feature vector set is recorded, that is, the number of elements in the vector. This dimension will be used in subsequent steps to determine whether dimensionality reduction processing is required.

[0049] S102: Process the first feature vector set using a preset feature vector processing model to generate a second feature vector set.

[0050] If there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed using a preset feature vector processing model to generate a second feature vector set, and the feature dimension of the second feature vector set is a second dimension, wherein the second dimension is smaller than the first dimension.

[0051] Specifically, after forming the first feature vector set, the first identification vector therein is compared with the preset identification vector mapping table. The preset identification vector mapping table is a database that stores multiple preset identification vectors and their corresponding information. By calculating the similarity between the first identification vector and the preset identification vector, it can be determined whether there is a preset identification vector similar to the first identification vector. The similarity can be calculated using algorithms such as cosine similarity, Euclidean distance, and Manhattan distance. In this step, if the calculated similarity is greater than the preset similarity threshold, it is considered that there are similar preset identification vectors.

[0052] When it is determined that there is a preset identification vector similar to the first identification vector, the first feature vector set is processed using a preset feature vector processing model. The preset feature vector processing model generates a new feature vector set based on the input feature vector set.

[0053] In this step, the model may be a dimensionality reduction model, such as principal component analysis, linear discriminant analysis, etc., which is used to reduce the dimension of the first feature vector set from the first dimension to the second dimension.

[0054] In one possible implementation, principal component analysis can be performed by transforming a set of variables that may be correlated into a set of linearly uncorrelated variables through orthogonal transformation. This new set of variables is called principal components. The main purpose of principal component analysis is data dimensionality reduction, which is achieved by retaining the features that contribute the most to the variance in the data set.

[0055] Specifically, each feature in the first eigenvector set is first standardized, that is, the mean is subtracted and divided by the standard deviation to ensure that all features have the same scale. Then, the covariance matrix is ​​calculated based on the standardized eigenvectors to measure the correlation between the features. The covariance matrix is ​​then subjected to eigenvalue decomposition to obtain eigenvalues ​​and corresponding eigenvectors. Then, the eigenvectors are sorted according to the size of the eigenvalues, and the eigenvectors corresponding to the first k largest eigenvalues ​​are selected as the principal components, where k is the dimension of the new feature space (i.e., the second dimension). The selected k eigenvectors are combined into a matrix, which is the projection matrix. The first eigenvector set is transformed by the projection matrix to obtain the second eigenvector set after dimensionality reduction.

[0056] In another possible implementation, the linear discriminant analysis can be a supervised dimensionality reduction technique, which attempts to find a projection direction so that data points of different categories are as separated as possible after projection, and data points of the same category are as close as possible after projection.

[0057] Specifically, for each category, calculate the covariance matrix of its samples, and then sum them up to get the intra-class scatter matrix. Calculate the covariance matrix between the means of different categories to get the inter-class scatter matrix. Find the optimal projection direction by solving the generalized eigenvalue problem of the inter-class scatter matrix and the intra-class scatter matrix. Sort the eigenvectors according to the size of the generalized eigenvalue, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the projection direction, where k is the dimension of the new feature space (i.e., the second dimension). Transform the first eigenvector set through the selected projection direction to obtain the second eigenvector set after dimensionality reduction.

[0058] S103: using a user behavior recognition model and determining a user characteristic behavior corresponding to the target user according to the second feature vector set.

[0059] Specifically, after the second feature vector set is generated, it is input into the user behavior recognition model. The user behavior recognition model is a trained machine learning or deep learning model, whose task is to identify the characteristic behavior of the target user based on the input feature vector set. In this step, the model can be a classification model, such as a support vector machine, a random forest, etc., which is used to classify the input feature vector into different user characteristic behavior categories.

[0060] In a possible implementation, the implementation of the above-mentioned support vector machine can be to use the second feature vector set after dimensionality reduction as input data, and each vector corresponds to a user feature behavior category label. Then, a kernel function is selected. Commonly used kernel functions include linear kernels, polynomial kernels, and radial basis functions. In practical applications, a suitable kernel function can be selected according to the characteristics of the data. Then, the parameters of the support vector machine (such as regularization parameters and kernel function parameters) are tuned by cross-validation and other methods to improve the classification performance. The support vector machine is then trained using the training data to obtain the optimal hyperplane and support vector. Finally, the new feature vector is input into the trained support vector machine model to obtain its corresponding user feature behavior category.

[0061] In another possible implementation, for the above random forest implementation, samples are randomly drawn from the training data (with replacement sampling) to build each tree. At each split point, a subset is randomly selected from all features, and the optimal split features and thresholds are selected based on indicators such as information gain and Gini impurity. The above process is repeated until the stopping condition is met (such as the tree reaches the maximum depth, the number of samples contained in the node is less than a certain threshold, etc.). For the new feature vector, each tree will give a prediction result. Random forest aggregates these prediction results by voting (for classification problems) or averaging (for regression problems) to obtain the final classification or regression value. Use a validation set or cross-validation to evaluate the performance of the random forest model, and adjust parameters (such as the number of trees, maximum depth, feature subset size, etc.) as needed. Finally, the new feature vector is input into the trained random forest model to obtain its corresponding user feature behavior category.

[0062] Suppose an e-commerce platform wants to identify users' purchasing preferences by analyzing their purchase records. First, the user's purchase record data is obtained through the data collection module. Then, the feature extraction module is used to extract features that can reflect the user's purchase preferences from the purchase records, such as the type, price, and quantity of purchased goods, and convert them into feature vectors. Next, the extracted feature vector is compared with the preset identification vector mapping table to determine whether there are similar preset identification vectors. If there are similar preset identification vectors, the preset feature vector processing model is used to reduce the dimension of the feature vector. Finally, the reduced feature vector is input into the user behavior recognition model to identify the user's purchase preference.

[0063] In this embodiment, the marketing behavior data of the target user is obtained, and features are extracted from the marketing behavior data to form a first feature vector set. If a preset identification vector mapping table contains a preset identification vector whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed using a preset feature vector processing model to generate a second feature vector set. Then, the user behavior recognition model is used to determine the user characteristic behavior corresponding to the target user based on the second feature vector set, thereby reducing the model calculation complexity. By optimizing the feature combination, the expression ability and recognition accuracy of the feature vector are improved, thereby achieving accurate capture of user behavior characteristics.

[0064] Specifically, by adopting feature extraction technology, key information is extracted from marketing behavior data to form a first feature vector set, which not only simplifies the amount of data to be processed subsequently, but also retains sufficient information to reflect the real behavior characteristics of users. When there is a preset identification vector with a high similarity to the first identification vector in the preset identification vector mapping table, the first feature vector set is processed by the preset feature vector processing model to generate a second feature vector set with a lower feature dimension. This dimensionality reduction process not only reduces the computational complexity, but also improves the expression ability and recognition accuracy of the feature vector by optimizing the feature combination. Specifically, by eliminating redundant and noise features and retaining the most representative features, subsequent user behavior identification is more accurate. Based on the second feature vector set after dimensionality reduction, the user behavior recognition model is used to determine the user's characteristic behavior. Since the second feature vector set has been optimized, the user behavior recognition model can converge faster and have higher recognition accuracy, which not only realizes the accurate capture of user behavior characteristics, but also provides strong data support for the formulation of the enterprise's marketing strategy.

[0065] Figure 2 FIG. 1 is a flowchart of a marketing data processing method based on a large model according to another exemplary embodiment of the present application. Figure 2 As shown, the marketing data processing method based on the big model provided in this embodiment includes: S201. Obtain marketing behavior data of target users, and perform feature extraction on the marketing behavior data to form a first feature vector set.

[0066] In this step, the marketing behavior data of the target user is obtained, and features are extracted from the marketing behavior data to form a first feature vector set, wherein the identification vector in the first feature vector set is the first identification vector, and the feature dimension of the first feature vector set is the first dimension.

[0067] Specifically, firstly, the marketing behavior data of the target user is obtained through the data collection module. Such data may include but is not limited to the user's purchase history, browsing history, click history, search history, etc. The data collection module may be an internal database system of the enterprise or an external data collection platform.

[0068] After obtaining the marketing behavior data of the target user, the feature extraction module is used to process the data. The main task of the feature extraction module is to extract information that can reflect the characteristics of user behavior from the original data and convert it into a feature vector. In this step, the identification vector in the feature vector set is the first identification vector, which is used to identify a user or a type of user behavior. At the same time, the feature dimension of the first feature vector set is the first dimension, and the size of this dimension depends on the number of extracted features.

[0069] S202: Determine whether a first dimension corresponding to a first feature vector set is greater than a preset dimension threshold.

[0070] Specifically, by judging whether the dimension of the first feature vector set is greater than the preset dimension threshold, it is possible to screen out those user behavior data with higher feature dimensions and richer information. For feature vector sets with lower dimensions and less information content, subsequent processing or other processing methods can be directly performed, thereby avoiding unnecessary calculations and waste of resources, and improving the pertinence and efficiency of data processing. When the dimension of the first feature vector set is higher, it often means that the set contains more user behavior feature information. At this time, using the preset feature vector processing model to process it can more effectively extract feature vectors closely related to the target user behavior, form a second feature vector set, and then help optimize the feature vector processing efficiency. It can be seen that by setting a preset dimension threshold, the input feature vector set can be preliminarily screened and filtered, which can avoid poor processing effects or errors caused by too low dimensions or insufficient information content, and at the same time, when the dimension is too high, it can also improve the processing speed and efficiency through dimensionality reduction processing.

[0071] S203: Determine a set of associated vectors of the first identification vector using a preset characteristic vector associated mapping table.

[0072] In this step, a set of associated vectors of the first identification vector is determined using a preset characteristic vector associated mapping table, wherein the preset characteristic vector associated mapping table is used to establish an associated relationship between each characteristic vector, wherein the associated relationship includes an associated direction and an associated weight value.

[0073] Specifically, the system needs to use a preset characteristic vector association mapping table to determine the association vector set of the first identification vector. The preset characteristic vector association mapping table is a key data structure, which is used to establish association relationships between various characteristic vectors. These association relationships include association directions and association weight values, which together describe the internal connection and importance between characteristic vectors.

[0074] In specific implementation, the system will search for all candidate vectors that have an association relationship with the first identification vector according to the preset characteristic vector association mapping table. These candidate vectors may be directly associated with the first identification vector, or they may be indirectly associated with the first identification vector through other vectors. For directly associated candidate vectors, their association direction is direct association, and the association weight value is directly obtained according to the record in the mapping table. For indirectly associated candidate vectors, the system needs to calculate the product of each association weight value in its association path as the indirect association weight value of the candidate vector.

[0075] After determining all candidate vectors and their associated weight values, the system will screen out those candidate vectors whose associated weight values ​​are greater than the threshold value according to the preset associated weight value threshold value, and form them into an associated vector set. This set contains the characteristic vectors that are most closely related to the first identification vector, which will be used to generate the second characteristic vector set in the subsequent steps.

[0076] In a possible case, a first candidate vector set corresponding to the first identification vector may be determined according to a preset characteristic vector association mapping table, and a first target association direction exists between the association relationship between the candidate vectors in the first candidate vector set and the first identification vector, and the first target association direction is used to represent that the candidate vectors in the first candidate vector set are directly associated with the first identification vector; According to the preset characteristic vector association mapping table, candidate vectors whose direct association weight values ​​are greater than the preset association weight values ​​are determined from the first candidate vector set to add to the first association vector set, and the association vector set includes the first association vector set.

[0077] Specifically, the system searches for candidate vectors directly related to the first identification vector according to the preset characteristic vector association mapping table to form a first candidate vector set. The preset characteristic vector association mapping table is a carefully constructed data structure that records in detail the association relationship between each characteristic vector, including the association direction and the association weight value. In this step, the system traverses the preset characteristic vector association mapping table to find all candidate vectors that have a direct association relationship with the first identification vector. Direct association relationships refer to those relationships whose association directions clearly point to the first identification vector, and these relationships are represented by specific tags or paths in the mapping table. Through traversal and screening, the system can accurately determine the first candidate vector set, which contains all candidate vectors directly related to the first identification vector. After determining the first candidate vector set, the system will then screen out candidate vectors whose direct association weight values ​​are greater than the threshold value according to the preset association weight value threshold value from the set. This step is intended to further optimize the feature vector combination and remove those vectors that are directly related and have large feature repetitiveness and do not contribute much to the target user behavior recognition. In specific implementation, the system will check the candidate vectors in the first candidate vector set one by one to obtain their corresponding direct association weight values ​​in the preset characteristic vector association mapping table. The system then compares these weight values ​​with the preset association weight value threshold. For candidate vectors with weight values ​​greater than the threshold, the system adds them to the first association vector set. The first association vector set is a key intermediate result that contains all feature vectors that are directly related to the first identification vector and have high association weight values. These vectors will play an important role in the subsequent feature vector processing steps, helping to improve the expressiveness and recognition accuracy of the feature vector.

[0078] In the above scheme, according to the preset characteristic vector association mapping table, the first candidate vector set directly associated with the first identification vector can be determined. These candidate vectors represent the characteristic information directly related to the target user behavior. Through further screening, it can be ensured that only those vectors that have an important impact on the target user behavior identification are retained. By setting the preset association weight value, the candidate vectors with a direct association weight value greater than the threshold can be screened out from the first candidate vector set to form a first association vector set. This step avoids processing a large amount of unnecessary or redundant data, thereby improving the efficiency of data processing. Moreover, since the vectors in the first association vector set are all screened, they are more closely associated with the target user behavior. This means that in the subsequent user behavior identification and marketing strategy formulation process, these feature vectors can provide more accurate information.

[0079] In another possible case, a second candidate vector set of the first identification vector is determined according to a preset characteristic vector association mapping table, and a second target association direction exists between the association relationship between the candidate vectors in the second candidate vector set and the first identification vector, and the second target association direction is used to characterize that the candidate vector is indirectly associated with the first identification vector; According to the preset characteristic vector association mapping table, a candidate vector whose indirect association weight value is greater than the preset association weight value is determined from the second candidate vector set to add it to the second association vector set. The association vector set includes the second association vector set, and the indirect association weight value is the product of each association weight value in the association path from the candidate vector in the second candidate vector set to the first identification vector.

[0080] Specifically, the system searches for candidate vectors indirectly associated with the first identification vector according to a preset characteristic vector association mapping table to form a second candidate vector set. Indirect associations refer to vectors that do not directly point to the first identification vector but have an association relationship with the first identification vector through other characteristic vectors. These indirect associations may be represented by multiple paths or chain associations in the mapping table. In this step, the system traverses the preset characteristic vector association mapping table to find all candidate vectors that have an indirect association relationship with the first identification vector. By considering various possible association paths and weight combinations, the system can comprehensively identify all indirectly related candidate vectors and add them to the second candidate vector set. For each candidate vector in the second candidate vector set, the system needs to calculate its indirect association weight value. This weight value is determined by considering the product of each association weight value in the entire association path from the candidate vector to the first identification vector. This ensures that the indirect association relationship is also reasonably quantified and evaluated. In specific implementation, the system finds one or more association paths for each candidate vector, which connect the candidate vector and the first identification vector. Then, the system calculates the product of all the association weight values ​​on each path to obtain the indirect association weight value of the candidate vector. If there are multiple paths, the system may select the path with the largest weight value as a representative, or consider the weight values ​​of all paths for some form of comprehensive evaluation. After determining the second candidate vector set and the indirect association weight value of each candidate vector, the system will next filter out the candidate vectors with indirect association weight values ​​greater than the threshold from the set according to the preset association weight value threshold. This step is intended to ensure that only those indirect association vectors that have a significant contribution to the identification of the target user behavior are retained. In specific implementation, the system will check the candidate vectors in the second candidate vector set one by one, obtain their indirect association weight values, and compare them with the preset association weight value threshold. For candidate vectors with weight values ​​greater than the threshold, the system will add them to the second association vector set. The second association vector set is a set of important feature vectors that are indirectly associated with the first identification vector. Although these vectors are not directly associated with the first identification vector, they provide additional information and perspectives about the target user behavior through association with other feature vectors. In the subsequent feature vector processing step, the second associated vector set will work together with the first associated vector set to jointly optimize the expression ability and recognition accuracy of the feature vector.

[0081] In the above scheme, by introducing the concept of indirect association, those candidate vectors that are not directly related to the target user behavior but are indirectly related to the target user behavior through other feature vectors can be identified. These vectors may contain additional and valuable information, which helps to more accurately understand the characteristic behavior of the target user. In the process of marketing data processing, only considering the directly related feature vectors may miss some important information. By introducing the concept of indirect association, the feature vector set can be supplemented and improved to improve the integrity and comprehensiveness of the feature vector. Although the vectors in the second candidate vector set are indirectly related to the first identification vector, by setting a preset association weight value, those candidate vectors whose indirect association weight values ​​are greater than the threshold can be screened out to form a second association vector set. The correlation between these vectors and the target user behavior is still very strong, and there is a large degree of duplication. Therefore, by using the difference set with the second association vector set, it is possible to further avoid processing a large amount of unnecessary or redundant data, thereby improving the efficiency of data processing.

[0082] S204: Determine the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table.

[0083] First, the system searches the preset identification vector mapping table for the preset identification vector with the highest similarity to the first identification vector, and records the similarity as the highest similarity. The key to this step is to accurately calculate the similarity between the first identification vector and the preset identification vector. This can usually be achieved through cosine similarity, Euclidean distance or other suitable similarity measurement methods.

[0084] S205: Determine a preset association weight value according to the highest similarity and a preset initial association weight value.

[0085] In this step, the preset association weight value may be determined according to the highest similarity and the preset initial association weight value. It is worth noting that the highest similarity and the preset association weight value may be set to be in a negative correlation.

[0086] After obtaining the highest similarity, the system will dynamically adjust the preset association weight value according to this similarity and the preset initial association weight value. The key here is to establish a negative correlation between the highest similarity and the preset association weight value. That is, when the highest similarity is high, it indicates that the behavioral characteristics of the first identification vector and a preset identification vector are very close. At this time, the preset association weight value should be lowered so that more feature vectors related to the target user's behavior can be excluded in subsequent processing, thereby avoiding overfitting and redundant processing. On the contrary, when the highest similarity is low, the preset association weight value should be increased to ensure that more feature vectors related to the target user's behavior can be included to improve the accuracy and comprehensiveness of recognition. In specific implementation, the system can preset an association weight value adjustment formula or algorithm, which calculates a new preset association weight value based on the highest similarity and the initial association weight value. For example, linear interpolation, exponential decay or other suitable mathematical methods can be used to implement this adjustment process.

[0087] Finally, the system will use the adjusted preset association weight value to continue the subsequent feature vector processing steps. This includes screening out qualified association vectors from the candidate vector set according to the adjusted weight value to form the final association vector set, which will be used for subsequent feature vector dimensionality reduction and user behavior recognition.

[0088] In the above scheme, by introducing the highest similarity to dynamically adjust the preset association weight value, the feature vector closely related to the target user behavior can be more accurately screened. When the similarity between the first identification vector and a preset identification vector is very high, it means that the behavior of the target user is very close to the behavior of the user group represented by the preset identification vector, so the preset association weight value should be reduced so that more feature vectors related to the target user behavior can be excluded in the subsequent analysis. On the contrary, when the similarity is low, the preset association weight value should be increased to ensure that more feature vectors related to the target user behavior are included in the subsequent identification. The above scheme can dynamically adjust the preset association weight value according to the marketing behavior data of different target users, thereby improving the adaptability and flexibility of the method. This helps to perform more accurate and personalized marketing data processing and analysis according to the characteristics and needs of different user groups in practical applications. In addition, by dynamically adjusting the preset association weight value, the data processing efficiency can be optimized while ensuring the accuracy of data processing. When the similarity is high, by reducing the preset association weight value, more feature vectors can be included for subsequent analysis, thereby reducing unnecessary feature vector analysis, thereby improving data processing efficiency. When the similarity is low, by increasing the preset association weight value, more feature vectors can be included in the subsequent analysis, thereby increasing the accuracy of recognition and helping to capture the characteristic behaviors of target users more comprehensively.

[0089] S206: Determine a second feature vector set according to the association vector set and the first feature vector set.

[0090] In this step, a second feature vector set is determined according to the association vector set and the first feature vector set, wherein the second feature vector set is a difference set between the first feature vector set and the association vector set.

[0091] If there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed using a preset feature vector processing model to generate a second feature vector set, and the feature dimension of the second feature vector set is a second dimension, wherein the second dimension is smaller than the first dimension.

[0092] Specifically, after forming the first feature vector set, the first identification vector therein is compared with the preset identification vector mapping table. The preset identification vector mapping table is a database that stores multiple preset identification vectors and their corresponding information. By calculating the similarity between the first identification vector and the preset identification vector, it can be determined whether there is a preset identification vector similar to the first identification vector. The similarity can be calculated using algorithms such as cosine similarity, Euclidean distance, and Manhattan distance. In this step, if the calculated similarity is greater than the preset similarity threshold, it is considered that there are similar preset identification vectors.

[0093] When it is determined that there is a preset identification vector similar to the first identification vector, the first feature vector set is processed using a preset feature vector processing model. The preset feature vector processing model generates a new feature vector set based on the input feature vector set. In this step, the model can be a dimensionality reduction model, such as principal component analysis, linear discriminant analysis, etc., which is used to reduce the dimension of the first feature vector set from the first dimension to the second dimension.

[0094] S207: Using the user behavior recognition model, and according to the second feature vector set, determining the user characteristic behavior corresponding to the target user.

[0095] Specifically, after the second feature vector set is generated, it is input into the user behavior recognition model. The user behavior recognition model is a trained machine learning or deep learning model, whose task is to identify the characteristic behavior of the target user based on the input feature vector set. In this step, the model can be a classification model, such as a support vector machine, a random forest, etc., which is used to classify the input feature vector into different user characteristic behavior categories.

[0096] It is worth noting that the above-mentioned user behavior recognition model can be various general models used in the prior art to identify user characteristic behaviors, which are not specifically limited here. It can also be a model formed by special training. For example, a recurrent neural network or its variants (such as long short-term memory network, gated recurrent unit) in a deep learning model is selected as the basic architecture of the user behavior recognition model, and then the model is trained using historical user behavior data as input and corresponding user characteristic behavior labels as output. The model parameters are continuously optimized through the back propagation algorithm and the gradient descent method until the performance of the model on the validation set reaches the preset standard to train the recognition model.

[0097] Suppose an e-commerce platform wants to identify users' purchasing preferences by analyzing their purchase records. First, the user's purchase record data is obtained through the data collection module. Then, the feature extraction module is used to extract features that can reflect the user's purchase preferences from the purchase records, such as the type, price, and quantity of purchased goods, and convert them into feature vectors. Next, the extracted feature vector is compared with the preset identification vector mapping table to determine whether there are similar preset identification vectors. If there are similar preset identification vectors, the preset feature vector processing model is used to reduce the dimension of the feature vector. Finally, the reduced feature vector is input into the user behavior recognition model to identify the user's purchase preference.

[0098] On the basis of the above embodiment, after the first feature vector set is processed by using the preset feature vector processing model to generate a second feature vector set, the method may further include: If the feature difference ratio between the second dimension and the first dimension is greater than the preset feature difference ratio threshold, the first feature vector set and the second feature vector set are added to the recognition model training database, the recognition model training database is used to iteratively train the user behavior recognition model, the first feature vector set is added to the verification set in the recognition model training database, and the second feature vector set is added to the training set in the recognition model training database.

[0099] In the above scheme, by continuously adding new feature vector sets (including the first feature vector set and the second feature vector set) to the recognition model training database and using them for iterative training of the model, the model's training sample library can be continuously enriched and optimized. This helps to improve the model's recognition accuracy and generalization ability for user behavior, so that it can better adapt to different application scenarios and user groups. In actual applications, user behavior data may change with time and environmental changes. By continuously iteratively training the model, the model can continuously adapt to new data distributions and feature patterns, thereby enhancing the robustness and stability of the model. This helps to ensure that the model always maintains high recognition performance and stability during long-term use.

[0100] It is worth noting that by setting the feature difference ratio between the second dimension and the first dimension to be greater than the preset feature difference ratio threshold as a condition for adding the corresponding data to the model training database, it is possible to screen out those data sets whose features change significantly during the processing process. These data sets often contain richer and more diverse user behavior information, which is of great significance for improving the recognition ability of the model. Setting the preset feature difference ratio threshold as a screening condition can ensure that only those data sets with sufficiently large feature changes will be added to the model training database for subsequent iterative training, which helps to avoid the introduction of redundant or low-quality data, thereby optimizing the quality of training samples, and thus improving the pertinence and efficiency of subsequent iterative training. In addition, due to the diversity and complexity of user behavior data, traditional fixed models are often difficult to adapt to all situations. By continuously introducing new data sets with significant feature changes for iterative training, the model can continuously learn and adapt to new data distributions and feature patterns, thereby helping to improve the adaptability and robustness of the model when processing complex and variable user behavior data, so that it can better meet the needs of practical applications. In addition, by setting the feature difference ratio threshold, unnecessary training samples can be reduced, thereby reducing the training cost and time of the model, helping to improve the training efficiency of the model, so that it can adapt to new data and needs more quickly. At the same time, since the screened data set is of high quality, it can also reduce the risk of overfitting of the model during training and further improve the generalization ability of the model.

[0101] It is worth noting that the above-mentioned first feature vector set can also be added to the verification set in the recognition model training database, and the second feature vector set can be added to the training set in the recognition model training database. By using the feature vector sets before and after processing for training and verification respectively, data resources can be used more efficiently. Adding the first feature vector set and the second feature vector set to the verification set and the training set respectively provides rich data support for the iterative optimization of the model. By using the verification set to evaluate the performance of the model, we can promptly discover the instability that may occur in the model during the training process, such as performance fluctuations, overfitting, etc. At the same time, adding the processed feature vector set (the second feature vector set) to the training set can further enrich the training samples of the model and improve the stability and robustness of the model.

[0102] Figure 3 FIG. 1 is a schematic diagram of a marketing data processing platform based on a large model according to an exemplary embodiment of the present application. Figure 3 As shown, the marketing data processing platform 300 based on the big model provided in this embodiment includes: An acquisition module 310 is used to acquire marketing behavior data of a target user, and perform feature extraction on the marketing behavior data to form a first feature vector set, wherein the identification vector in the first feature vector set is a first identification vector, and the feature dimension of the first feature vector set is a first dimension; A processing module 320 is configured to process the first feature vector set using a preset feature vector processing model to generate a second feature vector set when there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, wherein the feature dimension of the second feature vector set is a second dimension, and the second dimension is smaller than the first dimension; The determination module 330 is used to use the user behavior recognition model to determine the user characteristic behavior corresponding to the target user according to the second feature vector set.

[0103] Optionally, the processing module 320 is specifically configured to: Determine a set of associated vectors of the first identification vector by using a preset characteristic vector associated mapping table, wherein the preset characteristic vector associated mapping table is used to establish an associated relationship between each characteristic vector, wherein the associated relationship includes an associated direction and an associated weight value; The second feature vector set is determined according to the association vector set and the first feature vector set, wherein the second feature vector set is a difference set between the first feature vector set and the association vector set.

[0104] Optionally, the processing module 320 is specifically configured to: Determine, according to the preset characteristic vector association mapping table, a first candidate vector set corresponding to the first identification vector, wherein an association relationship between candidate vectors in the first candidate vector set and the first identification vector has a first target association direction, and the first target association direction is used to represent that the candidate vectors in the first candidate vector set are directly associated with the first identification vector; According to the preset characteristic vector association mapping table, candidate vectors having direct association weight values ​​greater than preset association weight values ​​are determined from the first candidate vector set to add to a first association vector set, the association vector set including the first association vector set.

[0105] Optionally, the processing module 320 is specifically configured to: Determine a second candidate vector set of the first identification vector according to the preset characteristic vector association mapping table, wherein the association relationship between the candidate vectors in the second candidate vector set and the first identification vector has a second target association direction, and the second target association direction is used to represent that the candidate vector is indirectly associated with the first identification vector; According to the preset characteristic vector association mapping table, a candidate vector whose indirect association weight value is greater than the preset association weight value is determined from the second candidate vector set to add it to the second association vector set, the association vector set including the second association vector set, and the indirect association weight value is the product of each association weight value in the association path from the candidate vector in the second candidate vector set to the first identification vector.

[0106] Optionally, the processing module 320 is specifically configured to: Determine the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table; The preset association weight value is determined according to the highest similarity and a preset initial association weight value, wherein the highest similarity and the preset association weight value are negatively correlated.

[0107] Optionally, the processing module 320 is specifically configured to: It is determined that the first dimension corresponding to the first feature vector set is greater than a preset dimension threshold.

[0108] Optionally, the processing module 320 is specifically configured to: If the feature difference ratio between the second dimension and the first dimension is greater than a preset feature difference ratio threshold, the first feature vector set and the second feature vector set are added to the recognition model training database, and the recognition model training database is used to iteratively train the user behavior recognition model.

[0109] Figure 4 is a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present application. Figure 4 As shown, an electronic device 400 provided in this embodiment includes: a processor 401 and a memory 402; wherein: The memory 402 is used to store computer programs, and the memory may also be a flash memory.

[0110] The processor 401 is used to execute the execution instructions stored in the memory to implement each step in the above method. For details, please refer to the relevant description in the above method embodiment.

[0111] Optionally, the memory 402 may be independent or integrated with the processor 401 .

[0112] When the memory 402 is a device independent of the processor 401, the electronic device 400 may further include: The bus 403 is used to connect the memory 402 and the processor 401 .

[0113] This embodiment further provides a readable storage medium, in which a computer program is stored. When at least one processor of an electronic device executes the computer program, the electronic device executes the methods provided in the above-mentioned various implementation modes.

[0114] This embodiment also provides a program product, which includes a computer program stored in a readable storage medium. At least one processor of the electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device implements the methods provided in the above various embodiments.

[0115] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0116] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A marketing data processing method based on a large model, characterized in that: include: Acquire marketing behavior data of target users, and perform feature extraction on the marketing behavior data to form a first feature vector set, wherein the identification vector in the first feature vector set is a first identification vector, and the feature dimension of the first feature vector set is a first dimension; If there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, the first feature vector set is processed by using a preset feature vector processing model to generate a second feature vector set, wherein the feature dimension of the second feature vector set is a second dimension, and the second dimension is smaller than the first dimension; A user behavior recognition model is used, and the user characteristic behavior corresponding to the target user is determined according to the second feature vector set.

2. The marketing data processing method based on a large model according to claim 1, characterized in that: Processing the first feature vector set using a preset feature vector processing model includes: Determine a set of associated vectors of the first identification vector by using a preset characteristic vector associated mapping table, wherein the preset characteristic vector associated mapping table is used to establish an associated relationship between each characteristic vector, wherein the associated relationship includes an associated direction and an associated weight value; The second feature vector set is determined according to the association vector set and the first feature vector set, wherein the second feature vector set is a difference set between the first feature vector set and the association vector set.

3. The marketing data processing method based on a large model according to claim 2 is characterized in that: The determining the associated vector set of the first identification vector by using a preset characteristic vector associated mapping table includes: Determine, according to the preset characteristic vector association mapping table, a first candidate vector set corresponding to the first identification vector, wherein an association relationship between candidate vectors in the first candidate vector set and the first identification vector has a first target association direction, and the first target association direction is used to represent that the candidate vectors in the first candidate vector set are directly associated with the first identification vector; According to the preset characteristic vector association mapping table, candidate vectors having direct association weight values ​​greater than preset association weight values ​​are determined from the first candidate vector set to add to a first association vector set, the association vector set including the first association vector set.

4. The marketing data processing method based on a large model according to claim 3 is characterized in that: The determining of the associated vector set of the first identification vector by using the preset characteristic vector associated mapping table further includes: Determine a second candidate vector set of the first identification vector according to the preset characteristic vector association mapping table, wherein the association relationship between the candidate vectors in the second candidate vector set and the first identification vector has a second target association direction, and the second target association direction is used to represent that the candidate vector is indirectly associated with the first identification vector; According to the preset characteristic vector association mapping table, a candidate vector whose indirect association weight value is greater than the preset association weight value is determined from the second candidate vector set to add it to the second association vector set, the association vector set including the second association vector set, and the indirect association weight value is the product of each association weight value in the association path from the candidate vector in the second candidate vector set to the first identification vector.

5. The marketing data processing method based on a large model according to claim 3 or 4, characterized in that: After determining the associated vector set of the first identification vector by using the preset characteristic vector associated mapping table, the method further includes: Determine the highest similarity between the first identification vector and the preset identification vector with the highest similarity in the preset identification vector mapping table; The preset association weight value is determined according to the highest similarity and a preset initial association weight value.

6. The marketing data processing method based on a large model according to claim 5 is characterized in that: Before the method uses a preset feature vector processing model to process the first feature vector set to generate a second feature vector set, the method further includes: It is determined that the first dimension corresponding to the first feature vector set is greater than a preset dimension threshold.

7. The marketing data processing method based on a large model according to claim 6 is characterized in that: After the first feature vector set is processed by using a preset feature vector processing model to generate a second feature vector set, the method further includes: If the feature difference ratio between the second dimension and the first dimension is greater than a preset feature difference ratio threshold, the first feature vector set and the second feature vector set are added to the recognition model training database, and the recognition model training database is used to iteratively train the user behavior recognition model.

8. A marketing data processing platform based on a large model, characterized in that: include: an acquisition module, configured to acquire marketing behavior data of a target user, and perform feature extraction on the marketing behavior data to form a first feature vector set, wherein an identification vector in the first feature vector set is a first identification vector, and a feature dimension of the first feature vector set is a first dimension; a processing module, configured to process the first feature vector set using a preset feature vector processing model to generate a second feature vector set when there is a preset identification vector in the preset identification vector mapping table whose similarity with the first identification vector is greater than a preset similarity threshold, wherein the feature dimension of the second feature vector set is a second dimension, and the second dimension is smaller than the first dimension; A determination module is used to use a user behavior recognition model to determine the user characteristic behavior corresponding to the target user according to the second feature vector set.

9. An electronic device, characterized in that: include: processor; as well as, A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images