Method for constructing user portraits of O2O platforms based on deep learning

Abstract

The present invention discloses a method for constructing a user portrait of an O2O platform based on deep learning, which relates to the technical field of constructing user portraits by deep learning. The method for constructing a user portrait of an O2O platform based on deep learning classifies users into groups through classification and regression trees, identifies user behavior characteristics, constructs a user portrait decision tree model, predicts user behavior data, generates user feature vectors, constructs a user portrait of the O2O platform based on the user feature vectors, analyzes the matching degree between the subsequent behavior data of users and the user portrait of the O2O platform, obtains an optimization index of the platform user portrait through comprehensive analysis with business indicators, establishes a mechanism for real-time update and iterative optimization, updates the user portrait decision tree model, optimizes the user portrait of the O2O platform, improves the accuracy of the user portrait, helps the O2O platform better understand the behavior patterns and preferences of users, thereby optimizing marketing strategies, helping the platform better meet user needs, and improving user loyalty and market share.

Description

Technical Field

[0001] The present invention relates to the technical field of constructing user portraits by deep learning, and specifically to a method for constructing user portraits of an O2O platform based on deep learning. Background Art

[0002] With the popularization of the Internet and smart phones, the lifestyle of users has changed significantly. More and more users are accustomed to shopping, making restaurant reservations, arranging travel, etc. through online platforms. The user portraits of O2O platforms need to reflect this transformation of the digital lifestyle in order to accurately capture user needs. Deep learning, through a multi-layer neural network structure, can achieve excellent performance in fields such as image recognition, natural language processing, and speech recognition. The progress of these technologies provides technical support for the accurate construction of user portraits of O2O platforms.

[0003] For example, the Chinese invention patent with the publication number CN116523545B discloses a method for constructing user portraits based on big data. By collecting a number of conditional features triggered by the user side at each preset collection period, calculating the conditional feature parameters corresponding to the portrait tags to determine whether to store the portrait tags and the number of conditional features triggered by the user side in the user portrait set corresponding to the user side, and determining the tag type of the portrait tags in the user portrait set. When the ratio of the number of fuzzy portrait tags to the number of standard portrait tags in the user portrait set is higher than the preset ratio comparison threshold, the relevance between each fuzzy portrait tag and the standard portrait tag is determined, and the fuzzy portrait tags in the user portrait set are optimized. When the number of continuously stored duplicate portrait tags is greater than the preset storage threshold, the preset collection period is adjusted, improving the efficiency and accuracy of constructing portrait tags for the user side.

[0004] Traditional methods for constructing user portraits mostly rely on rule engines or simple machine learning algorithms, and are difficult to process large-scale, multi-source, heterogeneous data. Moreover, there are limitations in capturing dynamic user behaviors, and they rely on manual experience and simple statistical analysis, making it difficult to comprehensively and accurately capture the complexity and dynamics of user behaviors. At the same time, in the process of determining the relevance between each fuzzy portrait tag and the standard portrait tag and optimizing the fuzzy portrait tags in the user portrait set, real-time update and iterative optimization cannot be carried out. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a method for constructing user portraits of an O2O platform based on deep learning, which solves the problems in the above background art.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for constructing a user portrait on an O2O platform based on deep learning, comprising the following steps: S1. Collect user behavior data in the data source of the O2O platform, and establish a comprehensive user behavior dataset through a clustering algorithm; S2. Based on the comprehensive user behavior dataset, classify users into groups through a classification and regression tree, identify user behavior characteristics, and construct a user portrait decision tree model; S3. Use the user portrait decision tree model to predict user behavior data, generate user feature vectors, and construct a user portrait on the O2O platform based on the user feature vectors; S4. Obtain the optimization index of the platform user portrait through the matching degree between the subsequent behavior data of users in the O2O platform data source and the O2O platform user portrait and business indicator data, establish a mechanism for real-time update and iterative optimization, update the user portrait decision tree model, and optimize the O2O platform user portrait.

[0007] Further, the specific process of establishing a comprehensive user behavior dataset through a clustering algorithm is as follows: Obtain user behavior data from the O2O platform data source, and perform data cleaning and extract user behavior characteristics. User behavior characteristics include user activity, user purchasing power, and browsing product types. Use the K-means algorithm to perform clustering analysis on users with similar user behavior characteristics, divide users into different clusters according to the clustering results, calculate the average value of the characteristics of each cluster as the cluster center, and generate a comprehensive user behavior dataset.

[0008] Further, the specific process of classifying users into groups through a classification and regression tree based on the comprehensive user behavior dataset is as follows: Load the comprehensive user behavior dataset, divide the comprehensive user behavior dataset into a training set and a test set, use the classification and regression tree algorithm to traverse the user behavior characteristics of the training set, and perform group feature division according to the user behavior characteristics. Calculate the Gini index after each feature division, select the feature with the smallest Gini index as the division feature, and construct a classification and regression tree layer by layer through the O2O platform product classification. Each node represents a user group, and each leaf node represents a user group classification.

[0009] Further, the specific process of identifying user behavior characteristics and constructing a user portrait decision tree model is as follows: Select the combination of a deep neural network and a classification and regression tree as the benchmark model of the user portrait decision tree model, input the user behavior characteristics of the training set, use the deep neural network to perform multi-level abstraction and non-linear feature combination on the user behavior characteristics, extract high-level behavior characteristics, and use the user behavior characteristics of the test set to train the benchmark model of the user portrait decision tree model to construct the user portrait decision tree model.

[0010] Further, the specific process of using the user portrait decision tree model to predict user behavior data and generate user feature vectors is as follows: Traverse the user behavior data through the user portrait decision tree model. Starting from the root node, perform node division layer by layer according to the feature values until reaching the leaf node; at each node, compare the user behavior feature values with the node division features, select the corresponding child node, and output the user feature vectors.

[0011] Further, the specific process of obtaining the platform user portrait optimization index through the matching degree between the subsequent user behavior data in the O2O platform data source and the O2O platform user portrait and the business metric data is as follows: Extract similar features from the subsequent user behavior data and the O2O platform user portrait, calculate the cosine similarity between the user behavior feature vector and the user portrait feature vector, and obtain the feature vector matching index; Perform a comprehensive operation on the user conversion rate and user click-through rate in the business metric data with the standard values of the user conversion rate and the user click-through rate to obtain the business compliance index; Perform a comprehensive operation on the feature vector matching index and the business compliance index to obtain the platform user portrait optimization index.

[0012] Further, the specific process of establishing a mechanism for real-time update and iterative optimization is as follows: Analyze the platform user portrait optimization index in real time, compare the platform user portrait optimization index with the set platform user portrait optimization update level threshold to obtain the O2O platform user portrait optimization information; When the platform user portrait optimization index is less than the first-level update threshold, mark this user portrait as a high update and iterative optimization portrait; When the platform user portrait optimization index is less than the second-level update threshold and greater than the first-level update threshold, mark this user portrait as a medium update and iterative optimization portrait; When the platform user portrait optimization index is less than the third-level update threshold and greater than the second-level update threshold, mark this user portrait as a low update and iterative optimization portrait.

[0013] Further, the specific process of updating the user portrait decision tree model and optimizing the O2O platform user portrait is as follows: Identify the user portrait features that need to be improved through the O2O platform user portrait optimization information, use the incremental learning algorithm to integrate the optimization information and feedback data into the user portrait decision tree model, perform re-training of the model, and verify the updated user portrait decision tree model through cross-validation and model evaluation methods.

[0014] The present invention has the following beneficial effects:

[0015] (1) The method for constructing a user portrait of an O2O platform based on deep learning classifies users into groups through a classification and regression tree, identifies user behavior characteristics, constructs a user portrait decision tree model, enhances the O2O platform's understanding of user behavior, can significantly improve the accuracy and efficiency of user analysis, and at the same time enhances the ability of personalized recommendation and user experience optimization, thereby enhancing the competitive advantage and market performance of the O2O platform. At the same time, through the matching degree of the subsequent behavior data of users in the O2O platform data source and the business indicator data of the O2O platform user portrait, the platform user portrait optimization index is obtained, and a mechanism for real-time update and iterative optimization is established to improve the accuracy of user portrait recognition.

[0016] (2) The method for constructing a user portrait of an O2O platform based on deep learning uses a user portrait decision tree model to predict user behavior data, generates user feature vectors, and constructs an O2O platform user portrait based on the user feature vectors, which can help the O2O platform better understand the behavior patterns and preferences of users, thereby optimizing marketing strategies. Having an accurate user portrait can help the platform better meet user needs and improve user loyalty and market share.

[0017] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the method for constructing a user portrait of an O2O platform based on deep learning according to the present invention.

[0019] Figure 2 It is a flowchart of constructing a user portrait decision tree model according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In the embodiments of the present application, through the method for constructing a user portrait of an O2O platform based on deep learning, the limitations of traditional methods relying on rule engines or simple machine learning algorithms, which cannot effectively process large-scale, multi-source, and heterogeneous data, are solved.

[0021] The general idea for the problems in the embodiments of the present application is as follows:

[0022] Collect user behavior data in the O2O platform data source, and establish a comprehensive user behavior dataset through a clustering algorithm.

[0023] Based on the comprehensive user behavior dataset, classify users into groups through a classification and regression tree, identify user behavior characteristics, and construct a user portrait decision tree model.

[0024] Use the user portrait decision tree model to predict user behavior data, generate user feature vectors, and construct an O2O platform user portrait based on the user feature vectors.

[0025] Based on the matching degree between the subsequent behavior data of users in the O2O platform data source and the O2O platform user portrait, as well as the business metric data, obtain the optimization index of the platform user portrait, establish a mechanism for real-time update and iterative optimization, update the user portrait decision tree model, and optimize the O2O platform user portrait.

[0026] Please refer to Figure 1 , an embodiment of the present invention provides a technical solution: a method for constructing an O2O platform user portrait based on deep learning, including the following steps: S1. Collect user behavior data in the O2O platform data source, and establish a comprehensive user behavior dataset through a clustering algorithm; S2. Based on the comprehensive user behavior dataset, classify users into groups through a classification and regression tree, identify user behavior characteristics, and construct a user portrait decision tree model; S3. Use the user portrait decision tree model to predict user behavior data, generate user feature vectors, and construct an O2O platform user portrait based on the user feature vectors; S4. Based on the matching degree between the subsequent behavior data of users in the O2O platform data source and the O2O platform user portrait, as well as the business metric data, obtain the optimization index of the platform user portrait, establish a mechanism for real-time update and iterative optimization, update the user portrait decision tree model, and optimize the O2O platform user portrait.

[0027] In this implementation plan, in step S1, collect user behavior data from multi-source data of the O2O platform, and use a clustering algorithm to group users with similar user behavior characteristics. The clustering algorithm helps to discover potential patterns and groups in the data, thereby establishing a comprehensive dataset containing diverse behavior characteristics. In step S2, use the classification and regression tree model to divide the comprehensive user behavior dataset into different groups and identify the main behavior characteristics of each group. This step helps to construct the user portrait decision tree model, and through the hierarchical structure of the tree, effectively capture and characterize the behavior habits and preferences of different user groups. In step S3, use the constructed user portrait decision tree model to predict new user behavior data and generate user feature vectors. These feature vectors reflect the personalized characteristics and behavior patterns of users, providing a basis for constructing the O2O platform user portrait. In step S4, use the user feedback data collected in the O2O platform to establish a real-time update and optimization mechanism, and continuously improve the user portrait decision tree model. Through incremental learning and model update, continuously optimize the user portrait to ensure its real-time matching and accuracy with user behavior.

[0028] Specifically, the specific process of establishing the comprehensive user behavior dataset through the clustering algorithm is as follows: Obtain user behavior data from the O2O platform data source, perform data cleaning, and extract user behavior characteristics. User behavior characteristics include user activity, user purchasing power, and browsing product types. Use the K-means algorithm to perform clustering analysis on users with similar user behavior characteristics, divide users into different clusters according to the clustering results, calculate the average value of each cluster's characteristics as the cluster center, and generate the comprehensive user behavior dataset.

[0029] In this implementation plan, user activity is the frequency and duration of a user's activity on the platform, which can be measured by indicators such as the number of logins and access duration; purchasing power represents the purchase frequency and amount of a user on the platform, which can be measured by indicators such as the number of orders and order amount; browsing product types represent the types or categories of products that a user is interested in, which can be analyzed by the types of products browsed by the user. The K-means algorithm is a commonly used clustering analysis method for grouping users with similar characteristics into different clusters. The clusters are divided according to the product types of the O2O platform. An example of using the K-means algorithm to perform clustering analysis on users with similar user behavior characteristics, dividing users into different clusters according to the clustering results, calculating the average value of each cluster's characteristics as the cluster center, and generating the comprehensive user behavior dataset is as follows: The two feature tables of user activity A and purchasing power P are as follows

[0030]

[0031]

[0032] Use the K-means algorithm to perform clustering analysis on these users according to their activity and purchasing power. Randomly select 2 initial clusters, cluster 1 and cluster 2, and the corresponding cluster centers u A =(0.6, 0.8), u B =(0.2, 0.3), u A represents the cluster center of cluster 1, u B represents the cluster center of cluster 2. Use the Euclidean distance to calculate the distance from the user to the cluster center; the Euclidean distance formula is as follows: d represents the distance from the user to the cluster center. The distance from user 1 to the cluster u A center is 0, and the distance to the cluster u A center is 0.64. The distance from user 2 to the cluster u A center is 0.22, and the distance to the cluster u A center is 0.72. The distance from user 3 to the cluster u A center is 0.64, and the distance to the cluster u A center is 0. The distance from user 4 to the cluster u A center is 0.22, and the distance to the cluster u AThe distance from the center is 0.42. Users 1, 2, and 4 are the closest to cluster 1 and are assigned to cluster 1, while user 3 is assigned to cluster 2. The formula for updating the cluster center is as follows: In the formula, ud represents the updated cluster center, C represents the number of cluster data points, and x i represents the i-th data point in the cluster. The updated cluster center is u A1 =(0.633, 0.7), u B2 =(0.2, 0.3). u A1 represents the updated cluster center of cluster 1, and u B2 represents the updated cluster center of cluster 2. Each user is assigned to the final cluster, and the center of the cluster is used as the comprehensive dataset of user behavior.

[0033] Specifically, based on the comprehensive dataset of user behavior, the specific process of classifying users through classification and regression trees is as follows: Load the comprehensive dataset of user behavior, divide the comprehensive dataset of user behavior into a training set and a test set, use the classification and regression tree algorithm to traverse the user behavior characteristics of the training set, and perform group feature division according to the user behavior characteristics. Calculate the Gini index after each feature division, select the feature with the smallest Gini index as the division feature, and construct a classification and regression tree layer by layer through the 020 platform product classification. Each node represents a user group, and each leaf node represents the user group classification.

[0034] In this implementation plan, group feature division is performed according to user behavior characteristics, the Gini index after each feature division is calculated, the feature with the smallest Gini index is selected as the division feature, and an example of constructing a classification and regression tree layer by layer through the 020 platform product classification is as follows: Through two feature tables of user activity A and purchasing power P, the activity is used as the division feature, the activity division point is set to 0.65, and cluster 1 in the comprehensive dataset of user behavior is used for division. The activities of users 1 and 4 are less than 0.65 and are divided into the left child node, and user 2 is divided into the right child node. The formula for calculating the Gini index of the node is as follows: Gin = 1 - (p1 2 - p2 2 ). In the formula, Gin represents the Gini index of the left child node, p1 represents the proportion of p1 in the node, and p2 represents the proportion of p2 in the node. The total node Gini index is obtained as 0.333.

[0035] Please refer to Figure 2, specifically, the specific process of identifying user behavior characteristics and constructing a user portrait decision tree model is as follows: Select the combination of a deep neural network and a classification and regression tree as the benchmark model of the user portrait decision tree model. Input the user behavior characteristics of the training set, use the deep neural network to perform multi-level abstraction and non-linear feature combination on the user behavior characteristics, extract high-level behavior characteristics, and use the user behavior characteristics of the test set to train the benchmark model of the user portrait decision tree model to construct the user portrait decision tree model.

[0036] In this implementation plan, the deep neural network can learn complex user behavior characteristics through multi-level abstraction and non-linear feature combination. This enables the model to handle more abstract and high-level feature representations, thereby better capturing the complex patterns behind user behavior. The classification and regression tree algorithm recursively divides the feature space into different regions and classifies or regresses the data within each region. It can provide intuitive decision rules and explain the relationships between features. The deep neural network processes the input user behavior characteristics through multiple hidden layers. Through weight learning, it can extract higher-level and more abstract feature representations from the original data. These features can capture the non-linear relationships and complex patterns of user behavior, improving the expressive ability of the model. The test set is used to evaluate the generalization ability of the model, that is, the performance of the model on unseen data. During the training process, it is necessary to ensure that the model not only performs well on the training set but also can be effectively generalized to the test set, so as to construct a user portrait decision tree model with good generalization ability.

[0037] , specifically, the specific process of using the user portrait decision tree model to predict user behavior data and generate user feature vectors is as follows: Traverse the user behavior data through the user portrait decision tree model, starting from the root node, and perform node division layer by layer according to the feature values until reaching the leaf node; at each node, compare the user behavior feature value with the node division feature, select the corresponding child node, and output the user feature vector.

[0038] In this implementation, traversal refers to inputting the user's behavior data into the user portrait decision tree model for processing and prediction. The behavior data of each user can include various behaviors such as clicks, purchases, and browsing. The user portrait decision tree model is a tree-like structure that starts from the root node and divides nodes according to specific feature values. The division basis is usually a specific feature in the user behavior data, such as the number of clicks or the purchase amount. The model will allocate the user behavior data to the corresponding child nodes according to the division features of each node. This process continues until the leaf nodes are reached, and the leaf nodes represent the final classification of user groups. At each node, the model compares the specific behavior feature values of the user with the relationship of the node division features. For example, if the division feature is that the purchase amount is greater than 100 yuan and the user's purchase amount is 120 yuan, then the user is allocated to the eligible child node. Finally, by traversing the entire decision tree, a user feature vector can be obtained according to the user's behavior data. This vector usually contains multiple dimensions of features, such as the user's interests, purchase preferences, and behavior activity, which are used to describe and characterize the user's attributes and behavior patterns.

[0039] Specifically, the specific process of constructing the O2O platform user portrait based on the user feature vector is as follows: Map the user feature vector to the predefined user portrait dimensions, generate the feature values of the user feature vector, identify the group to which the user belongs according to the feature values of the user feature vector, and construct the O2O platform user portrait. The O2O platform user portrait includes the user's key behavior features, preferences, and group information.

[0040] In this implementation, first, map the user feature vector to the predefined user portrait dimensions. The user feature vector is usually a vector containing multiple feature values, and these feature values can represent the user's behavior, preferences, and interests. For each user feature vector, corresponding feature values are generated according to the predefined user portrait dimensions. The feature values are numerical and categorical labels used to describe the user's behavior features and preferences. If the activity level in the user feature vector is 0.6, the purchasing power is 0.8, and the product preference is electronic products, then the generated feature values can be high activity level, high purchasing power, and electronic product preference.

[0041] Specifically, the specific process of obtaining the platform user portrait optimization index through the matching degree between the subsequent behavior data of users in the 020 platform data source and the 020 platform user portrait and the business metric data is as follows: Extract similar features from the subsequent behavior data of users and the 020 platform user portrait, calculate the cosine similarity between the user behavior feature vector and the user portrait feature vector, and obtain the feature vector matching index; Perform a comprehensive operation on the user conversion rate and the user click-through rate in the business metric data with the standard values of the user conversion rate and the user click-through rate to obtain the business compliance index; Perform a comprehensive operation on the feature vector matching index and the business compliance index to obtain the platform user portrait optimization index.

[0042] In this implementation plan, collect, clean, and encode the click, browse, purchase, evaluation, and other behavior data of users from the data source of the 020 platform, and convert it into a user behavior feature vector. The calculation formula for the feature vector matching index is as follows: In the formula, Vdf represents the user behavior feature vector, Vrg represents the user portrait feature vector, εy represents the feature vector matching index, βGB represents the standard value of the cosine similarity between vectors. The calculation formula for the business compliance index is as follows: In the formula, represents the business compliance index, which is used to indicate whether the user's behavior meets the business standards. ER represents the user conversion rate, RG represents the standard value of the user click-through rate, ΔFc represents the allowable deviation value of the user click-through rate, QZ represents the user click-through rate, DB represents the standard value of the user click-through rate, and ΔBN represents the allowable deviation value of the user click-through rate. The calculation formula for the platform user portrait optimization index is as follows: In the formula, ω represents the platform user portrait optimization index, ε 1 represents the weight factor of the feature vector matching index, ε 2 represents the weight factor of the business compliance index.

[0043] Specifically, the specific process of establishing a mechanism for real-time update and iterative optimization is as follows: Analyze the platform user portrait optimization index in real time, compare the platform user portrait optimization index with the set platform user portrait optimization update level threshold to obtain the 020 platform user portrait optimization information; When the platform user portrait optimization index is less than the first-level update threshold, mark this user portrait as a high-update iterative optimization portrait; When the platform user portrait optimization index is less than the second-level update threshold and greater than the first-level update threshold, mark this user portrait as a medium-update iterative optimization portrait; When the platform user portrait optimization index is less than the third-level update threshold and greater than the second-level update threshold, mark this user portrait as a low-update iterative optimization portrait.

[0044] In this implementation solution, by analyzing and optimizing the user profile in real time, it can be ensured that the user profile always reflects the latest user behaviors and preferences, thereby improving the accuracy and reliability of the profile. The real-time update mechanism enables the platform to quickly respond to changes in user behavior, timely adjust marketing strategies and service content, and enhance the flexibility and competitiveness of the platform. Through real-time data analysis and optimization, the platform can make more scientific and efficient decisions based on real data, reducing blindness and decision-making risks.

[0045] Specifically, the process of updating the user profile decision tree model and optimizing the user profile of the O2O platform is as follows: Based on the optimization information obtained from user feedback data, identify the user profile features that need to be improved. Use the incremental learning algorithm to integrate the optimization information and feedback data into the user profile decision tree model, and perform retraining of the model. Through cross-validation and model evaluation methods, verify the updated user profile decision tree model.

[0046] In this implementation solution, use user feedback data to analyze the current user profile model and identify the features that need to be improved. These features include user behavior characteristics, preferences, and group classification. Integrate the identified user profile features that need to be improved and the optimization information in the user feedback data into the user profile decision tree model. The incremental learning algorithm allows the model to gradually learn and improve based on new data, rather than completely retraining the entire model. Retrain the user profile decision tree model using the integrated dataset. Ensure that the model can be updated using the new feedback data and optimization information. Retraining involves adjusting model parameters, optimizing the loss function, and updating the model structure to better reflect the new features and behaviors of users. Use cross-validation techniques to evaluate the performance of the updated user profile decision tree model. Cross-validation can help evaluate the generalization ability of the model and avoid overfitting problems.

[0047] In summary, this application has at least the following effects:

[0048] The method for constructing a user profile of an O2O platform based on deep learning uses a user profile decision tree model to predict user behavior data, generates user feature vectors, and constructs a user profile of the O2O platform based on the user feature vectors, which can help the O2O platform better understand the behavior patterns and preferences of users, thereby optimizing marketing strategies. Having an accurate user profile can help the platform better meet user needs and improve user loyalty and market share.

[0049] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0050] The present invention is described with reference to the flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0051] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0053] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0054] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing user portraits of an O2O platform based on deep learning, characterized in that: The following steps are involved: S1. Collect user behavior data from the O2O platform data source and establish a comprehensive user behavior data set through clustering algorithms; S2. Based on the comprehensive user behavior data set, classify users into groups through classification and regression trees, identify user behavior characteristics, and build a user portrait decision tree model; S3. Use the user portrait decision tree model to predict user behavior data, generate user feature vectors, and build O2O platform user portraits based on the user feature vectors; S4. Obtain the platform user portrait optimization index through the matching degree and business indicator data of the user's subsequent behavior data in the O2O platform data source and the O2O platform user portrait, establish a real-time update and iterative optimization mechanism, update the user portrait decision tree model, and optimize the O2O platform user portrait; The specific process of obtaining the platform user portrait optimization index is as follows: Extract similar features from the user's subsequent behavior data and the O2O platform user portrait, calculate the cosine similarity between the user behavior feature vector and the user portrait feature vector, and obtain the feature vector matching index; Perform comprehensive calculation on the user conversion rate and user click-through rate in the business indicator data and the standard values ​​of the user conversion rate and the standard values ​​of the user click-through rate to obtain the business target index; Perform comprehensive calculations on the feature vector matching index and the business compliance index to obtain the platform user portrait optimization index; The calculation formula of the user behavior feature vector matching index is as follows: , where Vdf represents the user behavior feature vector, Vrg represents the user portrait feature vector, represents the feature vector matching index, βGB represents the standard value of cosine similarity between vectors, and the calculation formula of the business compliance index is as follows , where represents the business compliance index, ER represents the user conversion rate, RG represents the standard value of the user conversion rate, ΔFc represents the allowable deviation value of the user conversion rate, QZ represents the user click rate, DB represents the standard value of the user click rate, Indicates the allowed deviation value of user click rate. The calculation formula of platform user portrait optimization index is as follows: In the formula, ω represents the platform user portrait optimization index, represents the weight factor of the eigenvector matching index, Indicates the weight factor of the business compliance index.

2. The method for constructing an O2O platform user portrait based on deep learning according to claim 1, characterized in that: The specific process of establishing a comprehensive user behavior data set through clustering algorithms is as follows: User behavior data is obtained through the O2O platform data source, and data cleaning and user behavior features are extracted. User behavior features include user activity, user purchasing power and browsed product types. The K-means algorithm is used to perform cluster analysis on users with similar user behavior features. According to the clustering results, users are divided into different clusters, and the average value of each cluster feature is calculated as the cluster center to generate a comprehensive user behavior data set.

3. The method for constructing an O2O platform user portrait based on deep learning according to claim 2, characterized in that: Based on the comprehensive user behavior data set, the specific process of classifying users into groups through classification and regression trees is as follows: load the comprehensive user behavior data set, divide the comprehensive user behavior data set into a training set and a test set, use the classification and regression tree algorithm to traverse the user behavior features of the training set, and divide the group features according to the user behavior features, calculate the Gini index of each feature after division, select the feature with the smallest Gini index as the division feature, and build a classification and regression tree layer by layer through the O2O platform product classification. Each node represents a user group, and each leaf node represents a user group classification.

4. The method for constructing an O2O platform user portrait based on deep learning according to claim 3 is characterized in that: The specific process of identifying user behavior characteristics and building a user portrait decision tree model is as follows: A combination of deep neural network and classification and regression tree is selected as the benchmark model of the user portrait decision tree model. The user behavior characteristics of the training set are input, and the deep neural network is used to perform multi-level abstraction and nonlinear feature combination on the user behavior characteristics to extract high-level behavior features. The user behavior characteristics of the test set are used to train the benchmark model of the user portrait decision tree model to construct the user portrait decision tree model.

5. The method for constructing an O2O platform user portrait based on deep learning according to claim 4, characterized in that: The specific process of using the user portrait decision tree model to predict user behavior data and generate user feature vectors is as follows: Traverse the user behavior data through the user portrait decision tree model, starting from the root node, and divide the nodes layer by layer according to the feature values ​​until reaching the leaf node; At each node, the user behavior feature value is compared with the node partition feature, the corresponding child node is selected, and the user feature vector is output.

6. The method for constructing an O2O platform user portrait based on deep learning according to claim 5, characterized in that: The specific process of constructing O2O platform user portraits based on user feature vectors is as follows: Map the user feature vector to the predefined user portrait dimension, generate the eigenvalue of the user feature vector, identify the group to which the user belongs based on the eigenvalue of the user feature vector, and build the O2O platform user portrait. The O2O platform user portrait includes the user's key behavioral characteristics, preferences and group information.

7. The method for constructing an O2O platform user portrait based on deep learning according to claim 6, characterized in that: The specific process of establishing a real-time update and iterative optimization mechanism is as follows: Conduct real-time analysis on the platform user portrait optimization index, compare the platform user portrait optimization index with the set platform user portrait optimization update level threshold, and obtain O2O platform user portrait optimization information; When the platform user portrait optimization index is less than the first-level update threshold, the user portrait is marked as a high-update iterative optimization portrait; When the platform user portrait optimization index is less than the second-level update threshold and greater than the first-level update threshold, the user portrait is marked as a medium-update iterative optimization portrait; When the platform user portrait optimization index is less than the third-level update threshold and greater than the second-level update threshold, the user portrait is marked as a low-update iterative optimization portrait.

8. The method for constructing an O2O platform user portrait based on deep learning according to claim 7, characterized in that: The specific process of updating the user portrait decision tree model and optimizing the user portrait of the O2O platform is as follows: Through the O2O platform user portrait optimization information, the user portrait features that need to be improved are identified, and the optimization information and feedback data are integrated into the user portrait decision tree model using the incremental learning algorithm. The model is retrained, and the updated user portrait decision tree model is verified through cross-validation and model evaluation methods.

Images