CN115099344A

CN115099344A - Model training method and device, user portrait generation method and device, and equipment

Info

Publication number: CN115099344A
Application number: CN202210739784.4A
Authority: CN
Inventors: 萧梓健
Original assignee: Ping An Life Insurance Company of China Ltd
Current assignee: Ping An Life Insurance Company of China Ltd
Priority date: 2022-06-28
Filing date: 2022-06-28
Publication date: 2022-09-23

Abstract

The embodiment of the application provides a model training method and device, a user portrait generation method, device and equipment, and belongs to the technical field of artificial intelligence. The model training method comprises the following steps: acquiring sample image data of a sample object; carrying out feature extraction on the sample portrait data to obtain sample portrait features; constructing a decision sub-tree according to the sample portrait characteristics; the decision sub-tree comprises leaf nodes, and each leaf node is one of the sample portrait characteristics; analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; wherein each sample knowledge item comprises one of the leaf nodes; inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree; and obtaining a target decision tree model according to the weight of the knowledge bars of the K decision subtrees. The target decision tree model can identify required portrait characteristics from massive portrait data, and the training efficiency and accuracy of the model are improved.

Description

Model training method and device, user portrait generation method and device, and equipment

Technical Field

The application relates to the technical field of artificial intelligence, in particular to a model training method and device, a user portrait generation method and device, and equipment.

Background

User representation technology in artificial intelligence generally generates a virtual representation of a user, i.e. a tagged user model of abstract processing, according to user data, including user attributes, preferences, habits, behaviors and the like; user portraits, typically represented synthetically by data tags, such as [ gender: male, age: 25-30, study: this family), and the like. In the background of the development of large data technologies such as data warehouse, the user portrait technology is fully applied, however, how to select the required user tags from the massive portrait tags of the user becomes a technical problem to be solved urgently.

Disclosure of Invention

The embodiment of the application mainly aims to provide a model training method and device, a user portrait generation method and device, and equipment, wherein the target decision tree model obtained by the model training method can be used for identifying required portrait characteristics from massive portrait data.

In order to achieve the above object, a first aspect of an embodiment of the present application provides a model training method, where the model training method includes:

obtaining sample portrait data of a sample object;

performing feature extraction on the sample portrait data to obtain sample portrait features;

constructing a decision sub-tree according to the sample portrait characteristics; said decision sub-tree comprises leaf nodes, each of said leaf nodes being one of said sample representation features;

analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; wherein each sample knowledge bar comprises one of the leaf nodes;

inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree;

obtaining a target decision tree model according to the knowledge bar weights of the K decision subtrees; wherein the target decision tree model comprises the K decision sub-trees.

In some embodiments, the constructing a decision sub-tree from the sample portrait features comprises:

screening out a parent feature from the sample portrait features;

performing feature division on at least one father feature to obtain at least one sub-feature; wherein each of said parent features comprises at least one of said child features;

obtaining the decision sub-tree according to the at least one parent feature and the at least one child feature; wherein the parent feature is a branch node of the decision sub-tree, and each of the leaf nodes is one of the child features.

In some embodiments, the filtering out parent features from the sample portrait features comprises:

calculating the information gain of the sample image characteristics;

obtaining a maximum information gain from the information gain as a target gain;

the parent feature is filtered from the sample portrait features according to a target gain.

In some embodiments, said deriving a target decision tree model from the knowledge bar weights of K of the decision sub-trees comprises:

obtaining K decision subtrees;

acquiring the weight of a knowledge bar of each decision sub-tree of the K decision sub-trees;

and obtaining the target decision tree model according to the classification model and the weight of the knowledge bar of each decision sub-tree.

To achieve the above object, a second aspect of an embodiment of the present application proposes a user representation generation method, including:

acquiring target portrait data of a target user; inputting the target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; wherein the predicted knowledge item set comprises at least two knowledge items hit by the target user, and the target decision tree model is obtained by training according to the model training method of the first aspect;

inputting the prediction knowledge strip set into the classification model for prediction to obtain a target weight;

traversing the prediction knowledge strip set according to the target weight to obtain a target knowledge strip; wherein the target knowledge bar is a user representation of the target user.

In some embodiments, the traversing the set of predicted knowledge bars according to the target weight to obtain a target knowledge bar includes:

acquiring an absolute value of each target weight;

sorting the target weights according to the absolute values from large to small to obtain weight sorting;

and traversing the target weight in sequence according to the weight reordering order to obtain the target knowledge bar.

In order to achieve the above object, a third aspect of an embodiment of the present application provides a model training apparatus, including:

the sample data acquisition module is used for acquiring sample portrait data of a sample object;

the sample feature extraction module is used for extracting features of the sample portrait data to obtain sample portrait features;

the decision tree construction module is used for constructing a decision sub-tree according to the sample portrait characteristics; said decision sub-tree comprising leaf nodes, said each of said leaf nodes being one of said sample representation features;

the analysis module is used for analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; wherein each sample knowledge bar comprises one of the leaf nodes;

the model training module is used for inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree;

the model generation module is used for obtaining a target decision tree model according to the weight of the knowledge bars of the K decision subtrees; wherein the target decision tree model comprises the K decision sub-trees.

To achieve the above object, a fourth aspect of the embodiments of the present application provides a user representation generating apparatus, comprising:

the target data acquisition module is used for acquiring target portrait data of a target user; the decision module is used for inputting the target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; wherein the set of predicted knowledge bars comprises at least two knowledge bars hit by the target user, and the target decision tree model is trained according to the model training method of the first aspect;

the prediction module is used for inputting the prediction knowledge strip set into the classification model for prediction to obtain a target weight;

the traversal module is used for traversing the prediction knowledge bar set according to the target weight to obtain a target knowledge bar; wherein the target knowledge bar is a user representation of the target user.

In order to achieve the above object, a fifth aspect of embodiments of the present application provides an electronic device, which includes a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for implementing connection communication between the processor and the memory, wherein the program, when executed by the processor, implements the model training method according to the first aspect or the user representation generation method according to the second aspect.

In order to achieve the above object, a sixth aspect of the embodiments of the present application proposes a storage medium, which is a computer-readable storage medium for computer-readable storage, and the storage medium stores one or more programs, which are executable by one or more processors to implement the model training method of the first aspect or the user representation generation method of the second aspect.

The model training method and device, the user portrait generation method and device, the electronic equipment and the storage medium provided by the application are used for obtaining sample portrait data of a sample object and performing feature extraction on the sample portrait data to obtain sample portrait features; constructing a decision sub-tree according to the sample portrait characteristics, wherein the decision sub-tree comprises leaf nodes, and each leaf node is one of the sample portrait characteristics; analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bar of each decision sub-tree; therefore, a target decision tree model can be obtained according to the weight of the knowledge bars of the K decision subtrees, the target decision tree model obtained in the mode comprises the K decision subtrees, the required portrait characteristics can be identified from massive portrait data through the target decision tree model, and the model training efficiency and accuracy are improved.

Drawings

FIG. 1 is a flow chart of a model training method provided by an embodiment of the present application;

FIG. 2 is a flow chart of step 103 of FIG. 1;

FIG. 3 is a diagram illustrating the structure of a decision sub-tree, according to one embodiment;

FIG. 4 is a diagram illustrating the structure of a decision sub-tree according to another embodiment;

FIG. 5 is a flowchart of step 201 in FIG. 2;

FIG. 6 is a flowchart of step 106 in FIG. 1;

FIG. 7 is a flowchart of a user representation generation method provided by an embodiment of the present application;

FIG. 8 is a flowchart of step 704 in FIG. 7;

FIG. 9 is a schematic structural diagram of a model training apparatus according to an embodiment of the present application;

FIG. 10 is a schematic diagram of a user representation generating apparatus according to an embodiment of the present application;

fig. 11 is a hardware structure diagram of an electronic device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

It should be noted that although functional blocks are partitioned in a schematic diagram of an apparatus and a logical order is shown in a flowchart, in some cases, the steps shown or described may be performed in a different order than the partitioning of blocks in the apparatus or the order in the flowchart. The terms first, second and the like in the description and in the claims, as well as in the drawings described above, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.

First, several terms referred to in the present application are resolved:

artificial Intelligence (AI): is a new technical science for researching and developing theories, methods, technologies and application systems for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produces a new intelligent machine that can react in a manner similar to human intelligence, and research in this field includes robotics, language recognition, image recognition, natural language processing, and expert systems, among others. The artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results.

Natural Language Processing (NLP): NLP uses computer to process, understand and use human language (such as chinese, english, etc.), and belongs to a branch of artificial intelligence, which is a cross discipline between computer science and linguistics, also commonly called computational linguistics. Natural language processing includes parsing, semantic analysis, chapter understanding, and the like. Natural language processing is commonly used in the technical fields of machine translation, character recognition of handwriting and print, speech recognition and text-to-speech conversion, information intention recognition, information extraction and filtering, text classification and clustering, public opinion analysis and viewpoint mining, and relates to data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, linguistic research related to language calculation and the like related to language processing.

Logistic Regression (LR): also known as logistic regression analysis, is a classification learning method, which is generally applied to two scenarios: the first scenario is used for prediction, and the second scenario is used for finding influencing factors of dependent variables. The general principle is as follows: and predicting the probability of the future result occurrence through the representation of the historical data.

Gradient Boosting Decision Tree (GBDT): the method is a common model in machine learning, is an iterative decision tree algorithm, and has the main idea that an optimal model is obtained by iterative training of a weak classifier (decision tree).

XGBoost: XGboost uses a pre-sorted algorithm to more accurately find the data separation points.

Light Gradient decision model (light Gradient Boosting Machine, LightGBM): LightGBM is an advance of the GBDT model. LightGBM is an upgraded version of XGboost, and unlike XGboost, LightGBM uses histogram algorithm (histogram algorithm); LightGBM employs a leaf-wise growing strategy to find one leaf with the largest splitting gain (and generally the largest data volume) from all the current leaves at a time, then split, and so on.

User representation technology in artificial intelligence is generally a virtual representation of a user, i.e. an abstractly processed tagged user model, generated according to user data, including user attributes, preferences, habits, behaviors, and the like; user portrayal, typically represented synthetically by data tags, such as [ gender: male, age: 25-30, learning the calendar: this family), and the like. In the background of the development of big data technologies such as a data warehouse at present, user portrait technology is fully expanded and expanded, and the problems derived from the situation are as follows: how to select the main portrait which can represent the characteristics of the client most from massive portrait labels of the user. The main image has the function of identifying most characteristics (such as the risk of complaints of more than 70% of users) of the users on the basis of a small number of user portrait (main portrait) labels (such as portrait labels less than 1%) by positioning the main image of the users in a specific theme (such as user purchase, user retention, user complaints and the like). At present, the problems that personalized adaptation cannot be achieved and multi-feature intersection of users cannot be accurately depicted generally exist in a main portrait mining method of users. The application scenario related to the user multi-feature intersection is as follows: in a scenario of retention by insurance agents, there is generally a higher probability of retention in a multi-feature crossover scenario where user income exceeds the average level of society and pre-occupation is sales.

Based on this, embodiments of the present application provide a model training method and apparatus, a user portrait generation method and apparatus, an electronic device, and a storage medium, where a target decision tree model obtained by the model training method can identify a required portrait feature from a large amount of portrait data, and can improve training efficiency and accuracy of the model.

The embodiment of the application can acquire and process related data based on an artificial intelligence technology. Among them, Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best result.

The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

The embodiment of the application provides a model training method or a user portrait generation method, and relates to the technical field of artificial intelligence. The model training method or the user portrait generation method provided by the embodiment of the application can be applied to a terminal, a server side and software running in the terminal or the server side. In some embodiments, the terminal may be a smartphone, tablet, laptop, desktop computer, or the like; the server side can be configured into an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and cloud servers for providing basic cloud computing services such as cloud service, a cloud database, cloud computing, cloud functions, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (content delivery network) and big data and artificial intelligence platforms; the software may be an application that implements a model training method or a user representation generation method, etc., but is not limited to the above form.

The application is operational with numerous general purpose or special purpose computing system environments or configurations. For example: personal computers, server computers, hand-held or portable devices, tablet-type devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

In each embodiment of the present application, when data related to the identity or characteristics of a user, such as user information, user behavior data, user image data, user history data, and user location information, is processed, permission or consent of the user is obtained, and the collection, use, and processing of the data comply with relevant laws and regulations and standards in relevant countries and regions. In addition, when the embodiment of the present application needs to acquire sensitive personal information of a user, individual permission or individual consent of the user is obtained through a pop-up window or a jump to a confirmation page, and after the individual permission or individual consent of the user is definitely obtained, necessary user-related data for enabling the embodiment of the present application to operate normally is acquired.

The model training method and apparatus, the user portrait generation method and apparatus, the electronic device, and the storage medium provided in the embodiments of the present application are specifically described with reference to the following embodiments, in which the model training method in the embodiments of the present application is first described.

Fig. 1 is an alternative flowchart of a model training method provided in an embodiment of the present application, and the method in fig. 1 may include, but is not limited to, steps 101 to 106.

Step 101, obtaining sample image data of a sample object;

102, extracting characteristics of sample image data to obtain sample image characteristics;

103, constructing a decision sub-tree according to the sample portrait characteristics; the decision subtree comprises leaf nodes, wherein each leaf node is one of the sample portrait characteristics;

step 104, analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; wherein each sample knowledge item comprises one of the leaf nodes;

step 105, inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree;

step 106, obtaining a target decision tree model according to the weight of the knowledge bars of the K decision subtrees; wherein the target decision tree model comprises K decision sub-trees.

In steps 101 to 106 of the embodiment of the application, sample image data of a sample object is obtained, and feature extraction is performed on the sample image data to obtain sample image features; constructing a decision sub-tree according to the sample portrait characteristics, wherein the decision sub-tree comprises leaf nodes, and each leaf node is one of the sample portrait characteristics; analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree; therefore, a target decision tree model can be obtained according to the weight of the knowledge bars of the K decision subtrees, required portrait characteristics can be identified from massive portrait data through the target decision tree model, and the training efficiency and accuracy of the model can be improved.

In step 101 of some embodiments, the sample object may be a buyer of the shopping platform, or may be a user of a related application, such as a user of a music playing program, or an applicant in the insurance field, etc.; the sample portrait data of the sample object may include user basic information of the sample object, may also include purchase data of the sample object, may also include complaint data of the sample object, and so on; the user basic information of the sample object may include the name, sex, age, etc. of the sample object, and may also include academic information, professional information, etc. of the sample object. The sample portrait data in step 101 may include massive portrait data, and may be applied to various scenarios, such as a shopping platform, a friend-making platform, etc., and may also be in the insurance field, the house rental field, the electric power payment field, etc.

In step 102 of some embodiments, the method for extracting features from the sample image data may be an existing method for extracting features, and this embodiment is not limited thereto. The sample portrait characteristics can be academic characteristics, and the academic can comprise primary school, junior high school, college, major, subject, master level and the like; sample portrait features may also be user-defined with respect to basic information features such as gender features, age features, native place features, and the like.

Referring to fig. 2, step 103 in some embodiments may include, but is not limited to including, steps 201 through 203:

step 201, screening out a parent feature from sample portrait features;

step 202, performing feature division on at least one parent feature to obtain at least one child feature; wherein each parent feature comprises at least one child feature;

step 203, obtaining a decision sub-tree according to the at least one parent characteristic and the at least one child characteristic; wherein, the father characteristic is a branch node of the decision sub-tree, and each leaf node is one of the child characteristics.

In step 201 of some embodiments, if the sample representation features include a scholarly calendar, then the "scholarly calendar" in the sample representation features is taken as a parent feature; in step 202, the parent feature "academic calendar" is divided to obtain two sub-features, wherein one sub-feature is: the other sub-feature is "major, this family, Master and above". Referring to fig. 3, through step 203, a decision sub-tree is constructed, where fig. 3 illustrates a decision sub-tree with a two-layer structure, and the parent feature is the first layer and is also the "academic calendar" of the parent node; two sub-features are the second layer and two sub-features are two sub-nodes. According to the embodiment of the application, all the characteristics can be enumerated in a greedy manner so as to carry out characteristic division, and at least one sub-characteristic is obtained.

In some application scenarios, the decision sub-tree is a binary tree; in the embodiment of the present application, a binary tree is taken as an example for illustration. Referring to fig. 4, fig. 4 illustrates a decision sub-tree with a four-layer structure, which is repeatedly executed 202 to continuously perform feature segmentation, wherein after feature segmentation is performed on one sub-feature "primary school, junior middle school, and senior middle school", two sub-features of a next layer are obtained: primary school, junior middle school and high school, at this time, the child feature primary school, junior middle school and high school becomes the father feature; the other sub-feature "major, this family, master and above" is divided into two sub-features of the next layer: "Benke, Master and above" and "major", the child features "major, Benke, Master and above" become parent features. Similarly, the sub-feature "primary school, junior middle school" is divided into two sub-features of the next layer: primary school and junior middle school; similarly, the sub-features "this family, Master and above" are divided into two sub-features of the next layer: "this family" and "Master and above". In the decision sub-tree, the parent feature is also a parent node, the child feature is also a branch node, the child feature at the end is a leaf node, and the leaf node in fig. 4 includes: the sub-feature "high school" of the third layer, the sub-feature "elementary school" of the fourth layer, the sub-feature "junior school" of the fourth layer, the sub-feature "major expert" of the third layer, the sub-feature "master and above" of the fourth layer, and the sub-feature "home" of the fourth layer.

Referring to fig. 5, step 201 in some embodiments may include, but is not limited to including, steps 501 through 503:

step 501, calculating the information gain of the sample image characteristics;

step 502, obtaining the maximum information gain from the information gain as a target gain;

step 503, screen out parent features from the sample portrait features according to the target gain.

Information gain, which is used for measuring the degree of the data becoming more orderly and purer; the larger the information gain, the purer the data, so when feature selection is performed, the feature with the largest information gain will be selected. In steps 501 to 503 of this embodiment, the maximum information gain is obtained from the information gains as the target gain, so that the sample portrait feature to which the target gain is applied is taken as the parent feature.

In step 202 of some embodiments, each time feature division is performed, only division of one node in a decision sub-tree is determined, and a decision sub-tree is finally generated only by performing feature division for multiple times in a recursive manner; stopping feature division when a preset stop condition is reached; the preset stopping condition may be that feature division is stopped by setting a hyper-parameter, setting a level threshold of a node, or setting a depth threshold of a decision sub-tree exceeding a value of the hyper-parameter, or exceeding a level threshold of the node, or exceeding a depth threshold of the decision sub-tree.

In some embodiments, each decision sub-tree includes a root node, and step 104 in some embodiments may include, but is not limited to including:

traversing the decision subtree from the root node to obtain leaf nodes;

and analyzing the leaf nodes to obtain a sample knowledge strip.

In some embodiments, taking the decision subtree as a binary tree as an example for illustration, the manner of traversing the decision subtree may include: traversing the front sequence; wherein, the front-end traversal means that the traversal is started from the root node, then the traversal is carried out on the left node, and finally the traversal is carried out on the right node; in some application scenarios, traversal may be performed according to a conventional front-end traversal manner, which is not limited in this embodiment. In other embodiments, the traversing manner of the decision subtree may further include a middle-order traversal or a subsequent traversal, and may also include a hierarchical traversal, which is not limited in this embodiment; the traversal process of the middle-order traversal, the subsequent traversal, or the hierarchical traversal, which is not limited in this embodiment, may refer to a conventional traversal principle.

As can be seen from the above step 103, each leaf node is one of the sample image features, so the leaf node is analyzed in step 104 to obtain the corresponding sample image feature, and the obtained sample image feature is the sample knowledge bar. For example, in fig. 4, the leaf node may be a sub-feature "high-middle" of the third layer, and the sub-feature "high-middle" is also one of the sample knowledge bars.

In step 105 of some embodiments, the preset classification model may be a logistic regression model, and in other embodiments, the preset classification model may also be another model, which is not limited in this embodiment; the present embodiment takes a logistic regression model as an example for explanation; inputting the sample knowledge strips into a logistic regression model for training, and obtaining the weight of each knowledge strip of the decision subtree; in some application scenarios, the ith knowledge item weight may be expressed as wx _i The model parameters of the logistic regression model may be denoted as w. In step 105, the logistic regression model is trained according to the input sample knowledge strips, first the weight probabilities of the sample knowledge strips are obtained, and then the weight probabilities are analyzed to obtain the weight of the knowledge strip corresponding to each sample knowledge strip, wherein the weight probabilities and the corresponding weight of the knowledge strip are in one-to-one correspondence.

In some embodiments, the preset classification model may be trained by a gradient descent method. The present embodiment does not limit the training mode of the classification model.

Referring to fig. 6, in some embodiments, step 106 may include, but is not limited to including, steps 601-602:

601, acquiring K decision subtrees;

step 602, obtaining a weight of a knowledge bar of each decision sub-tree of the K decision sub-trees;

step 603, a target decision tree model is obtained according to the classification model and the weight of the knowledge bar of each decision sub-tree.

In some application scenarios, taking the example of obtaining K decision subtrees in total, each decision subtree including n sample knowledge bars as an example, K decision subtrees share K × n sample knowledge bars, that is, share K × n knowledge bar weights.

In step 603 of some embodiments, model parameters w of the classification model are obtained, and an ith sample knowledge item x is obtained _i Then multiplying the model parameter w and the ith sample knowledge strip to obtain the weight wx of each knowledge strip _i Then, summing the weights of the n knowledge bars of each decision sub-tree to obtain a logistic regression model, which is shown in the following formula (1):

∑(wx ₁ +wx ₂ +wx _i +... wxn) formula (1)

In equation (1), for each decision sub-tree: multiplying the model parameter w and the ith sample knowledge strip to obtain the weight wx of each knowledge strip _i And summing the weights of the n knowledge bars of each decision sub-tree to obtain a logistic regression model.

In some embodiments, the target decision tree model may be a LightGBM model; in other embodiments, the goal decision tree model may be an XGboost model. In addition, the target decision tree model may also be another decision tree model, and this embodiment is not limited. In this embodiment, the target decision tree model is exemplified as the LightGBM model. The decision tree is a model generated by a greedy strategy, the implemented target decision tree model is a binary tree structure, and at each decision node, the target decision tree model calculates and divides a feature partition which can maximize the current information gain, for example, the academic character partition is divided into two sets (primary school, junior middle school, high school, major, basic department, master and above). In the target decision tree model of this embodiment, a new decision sub-tree is learned according to the loss of the previous step, and a plurality of decision sub-trees are obtained. In addition, each decision sub-tree constituting the objective decision tree model of the present embodiment is a binary tree structure.

In some embodiments, the model training method may further include, but is not limited to, including: the optimization target decision tree model specifically comprises the following steps:

calculating optimal parameters of a decision sub-tree through second-order Taylor expansion;

and optimizing the target decision tree model according to the optimal parameters.

In other embodiments, the optimal parameter of the decision sub-tree may be calculated in other manners, which is not limited in this embodiment. The target decision tree model obtained by the implementation is a combination of the decision tree model and the logistic regression model, so that the optimal parameters of the decision sub-tree are obtained through calculation, and the target decision tree model can be optimized.

In addition, the convergence condition of the target decision tree model may be determined by setting a loss function, which may be a cross entropy loss function, where the cross entropy loss function represents a difference between the prediction probability of the current model and the distribution of the true target value of the sample, and when the model prediction is closer to the true situation, the loss is smaller, and when the loss is the smallest, the model converges.

The obtained target decision tree model may be used for performing feature division according to the user image, and specifically, will be described in detail later, and will not be described herein again.

FIG. 7 is an alternative flowchart of a user representation generation method provided in an embodiment of the present application, and the method in FIG. 7 may include, but is not limited to, steps 701 to 704.

Step 701, acquiring target portrait data of a target user;

step 702, inputting target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; the prediction knowledge strip set comprises at least two knowledge strips hit by a target user, and a target decision tree model is obtained by training according to the model training method of the embodiment;

step 703, inputting the prediction knowledge set into a classification model for prediction to obtain a target weight;

step 704, traversing the prediction knowledge strip set according to the target weight to obtain a target knowledge strip; wherein the target knowledge bar is a user representation of the target user.

In steps 701 to 704 of the embodiment of the application, target portrait data of a target user is obtained; inputting target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; inputting the prediction knowledge set into a classification model for prediction to obtain a target weight; traversing the prediction knowledge strip set according to the target weight to obtain a target knowledge strip, wherein the target knowledge strip is a user portrait of a target user and is also a main portrait of the target user; therefore, accurate identification of the target user can be realized.

At present, in a user main portrait mining method, problems that personalized adaptation cannot be achieved, multi-feature intersection of users cannot be accurately depicted, and the like generally exist, the user portrait generating method provided by this embodiment is suitable for accurately identifying a user main portrait in a complex user portrait system, a target knowledge bar obtained through step 704 is a main portrait of a target user, and through the main portrait, accurate identification of the target user can be achieved, so that identification of the target user is achieved through features as small as possible. In a complex system, the user portrait generation method of the embodiment can also realize accurate identification of the target user.

The target user in step 701 in some implementations is similar to the sample object in step 101, and the target user may be a buyer of the shopping platform, or a user of a related application, such as a user of a music playing program, or an applicant in the insurance field, etc.; the target portrait data of the target user may include user basic information of the target user, may also include purchase data of the target user, may also include complaint data of the target user, and so on; the user basic information of the target user may include a name, a gender, an age, and the like of the target user, and may also include academic information, professional information, and the like of the target user. The target portrait data in step 101 may include massive portrait data, and may be applied to various scenes, such as a shopping platform, a friend-making platform, etc., and may also be in the insurance field, the house rental field, the electric power payment field, etc.

In some embodiments, at step 702, the target decision tree model is obtained by training according to the model training method of the above embodiment, where the target decision tree model includes K decision subtrees, each decision subtree is a binary tree, target portrait data is input into the target decision tree model, and a decision is made through the target decision tree to obtain a prediction knowledge item set UKB _s The predicted knowledge set UKB _s Is the set of knowledge bars hit by the target user, the predicted knowledge bar set UKB _s Including at least two knowledge bars hit by the target user.

In some implementations, step 703, the classification model is a logistic regression model that will predict the knowledge set of the knowledge sets UKB _s Inputting the target weight into a logistic regression model for prediction to obtain the target weight.

Referring to fig. 8, in some implementations, step 704 may include, but is not limited to, steps 801 to 803:

step 801, acquiring an absolute value of each target weight;

step 802, sorting the target weights according to the absolute value from large to small to obtain weight sorting;

and 803, sequentially traversing the target weights according to the weight sorting to obtain a target knowledge bar.

In some implementations, step 801, for each target weight, its absolute value is calculated.

In some implementations, in step 802, the target weights are ranked from large to small according to the absolute value of each target weight, resulting in a ranked weight rank, which is ranked in order from large to small.

In some implementations, step 803 may include, but is not limited to including:

performing summation calculation on all target weights to obtain a weight sum; in some embodiments, the sum of weights is denoted while _ locations;

sequentially traversing the prediction knowledge strip set according to the weight sorting to obtain traversed knowledge strips; in some embodiments, the sum of traversal weights is denoted sum _ locations _j Where j is the number of traversed knowledge bars that have completed traversal;

summing all the traversed knowledge strips to obtain a traversal weight sum;

comparing the traversing weight and the sum of the weights which is larger than the sum of the weights according to the target weight;

and if the traversal weight sum is larger than the weight sum, stopping traversal to obtain a target knowledge bar.

In a specific application scenario, if the predicted knowledge item set hit by a target user has 10 knowledge items in 100 knowledge items, sorting the 10 knowledge items from large to small according to the absolute values of the weights of the 10 knowledge items hit by the target user, and performing weight accumulation on the sorted 10 knowledge items to obtain a weight sum of while _ logits; and traversing the sequenced 10 knowledge strips in sequence, and summing all the traversed knowledge strips to obtain traversal weights and sum _ locations when traversing one knowledge strip _j If the traversal of the 6 th knowledge bar is completed, the sum of the traversal weights at this time is sum _ locations ₆ 。

Comparing the magnitude between the traversal weight and the sum greater than the weight according to the target weight, including:

acquiring the current weight of the current traversal knowledge bar;

comparing the current weight to a weight threshold; wherein, the weight threshold is used for representing the coverage rate of the portrait characteristics;

if the current weight is greater than the weight threshold, comparing the size between the traversal weight and the sum of the weights;

and if the current weight is less than or equal to the weight threshold, respectively negating the traversing weight and the sum of the weights greater than the traversing weight, and then comparing the sizes.

In a specific application scenario, the traversal weight and the sum of the weights greater than the traversal weight are respectively negated and then compared with each other, which can be expressed as shown in equation (2):

sign*sum_logits _i >sign while locations formula (2)

Wherein, thre is a weight threshold, sign is a mark for determining whether the sum of the traversal weights and the sum of the weights are inverted, if the current weight is greater than the weight threshold, sign is 1, which indicates that the sum of the traversal weights and the sum of the weights do not need to be inverted; if the current weight is less than or equal to the weight threshold, sign is-1, which means that the sum of the weights needs to be inverted when the weights are traversed. For example, if the weight threshold thre is 0.5, it represents that the coverage rate of the image features is 50%, and also represents that the features cover 50% of the user; if the current weight is 0.6, the current weight is 0.6>If the weight threshold is 0.5, sign is 1, the magnitude between the traversal weight sum and the sum of weights is directly compared: sum _ locations _i >while _ locations _ thre; if the current weight is 0.4, the current weight is 0.4<If the weight threshold is 0.5, sign is-1, and the traversal weight sum and the sum of weights greater than the weight sum are respectively negated and then the magnitudes are compared: -sum _ locations _i <-while _ locations _ thre. When traversing the weights sum _ locations _j And when the sum is larger than the weight sum while _ locations, namely when the inequality of the formula (2) is satisfied, stopping traversing to obtain the target knowledge bar. The target knowledge bar obtained in the mode is the main image characteristic of the target user, so that the target user can be accurately identified.

Referring to fig. 9, an embodiment of the present application further provides a model training apparatus, including:

the sample characteristic extraction module is used for carrying out characteristic extraction on the sample portrait data to obtain sample portrait characteristics;

the decision tree construction module is used for constructing a decision sub-tree according to the sample portrait characteristics; the decision sub-tree comprises leaf nodes, and each leaf node is one of the sample portrait characteristics;

the analysis module is used for analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; wherein each sample knowledge item comprises one of the leaf nodes;

the model generation module is used for obtaining a target decision tree model according to the weight of the knowledge bars of the K decision subtrees; wherein the target decision tree model comprises K decision sub-trees.

The specific implementation of the model training apparatus is substantially the same as the specific implementation of the model training method, and is not described herein again.

Referring to fig. 10, an embodiment of the present application further provides a user representation generating apparatus, which can implement the user representation generating method, where the user representation generating apparatus includes:

the target data acquisition module is used for acquiring target portrait data of a target user;

the decision module is used for inputting target portrait data into the target decision tree model for decision making to obtain a prediction knowledge strip set; the prediction knowledge strip set comprises at least two knowledge strips hit by a target user, and a target decision tree model is obtained by training according to the model training method of the first aspect;

the prediction module is used for inputting the prediction knowledge strip set into the classification model for prediction to obtain target weight;

the traversal module is used for traversing the prediction knowledge strip set according to the target weight to obtain a target knowledge strip; wherein the target knowledge bar is a user representation of the target user.

The specific implementation of the user image generating apparatus is substantially the same as the specific implementation of the user image generating method, and is not described herein again.

An embodiment of the present application further provides an electronic device, where the electronic device includes: the system comprises a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for enabling communications between the processor and the memory, the program when executed by the processor implementing the model training method or the user representation generation method described above. The electronic equipment can be any intelligent terminal including a tablet computer, a vehicle-mounted computer and the like.

Referring to fig. 11, fig. 11 illustrates a hardware structure of an electronic device according to another embodiment, where the electronic device includes:

the processor 110 may be implemented by a general-purpose CPU (central processing unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits, and is configured to execute a relevant program to implement the technical solution provided in the embodiment of the present application;

the memory 112 may be implemented in the form of a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a Random Access Memory (RAM). The memory 112 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present disclosure is implemented by software or firmware, the relevant program codes are stored in the memory 112 and called by the processor 110 to execute the model training method or the user representation generating method according to the embodiments of the present disclosure;

an input/output interface 114 for implementing information input and output;

the communication interface 116 is used for realizing communication interaction between the device and other devices, and may realize communication in a wired manner (e.g., USB, network cable, etc.) or in a wireless manner (e.g., mobile network, WIFI, bluetooth, etc.);

a bus 118 that transfers information between various components of the device (e.g., the processor 110, memory 112, input/output interfaces 114, and communication interfaces 116);

wherein processor 110, memory 112, input/output interface 114, and communication interface 116 are communicatively coupled to each other within the device via bus 118.

Embodiments of the present application further provide a storage medium, which is a computer-readable storage medium for computer-readable storage, and the storage medium stores one or more programs, and the one or more programs are executable by one or more processors to implement the model training method or the user representation generation method.

The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The model training method and device, the user portrait generation method and device, the electronic equipment and the storage medium provided by the embodiment of the application are used for obtaining sample portrait data of a sample object and performing feature extraction on the sample portrait data to obtain sample portrait features; constructing a decision sub-tree according to the sample portrait characteristics, wherein the decision sub-tree comprises leaf nodes, and each leaf node is one of the sample portrait characteristics; analyzing the decision subtrees to obtain a sample knowledge bar of each decision subtree; inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bars of each decision sub-tree; therefore, a target decision tree model can be obtained according to the weight of the knowledge bars of the K decision subtrees, required portrait characteristics can be identified from massive portrait data through the target decision tree model, and the training efficiency and accuracy of the model can be improved. The target knowledge bar obtained by the user portrait generation method is the main portrait characteristic of the target user, so that the target user can be accurately identified.

In this embodiment, a decision tree model and a logistic regression model are combined to recognize a target user, where the decision tree model is used to perform feature division to obtain knowledge bars, the knowledge bars obtained through the decision tree model are input to the logistic regression model to perform prediction to obtain target weights, and thus, a prediction knowledge bar set is traversed according to the target weights to obtain target knowledge bars, and the target knowledge bars are main images of the target user, so that accurate recognition of the target user can be achieved.

The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute limitations on the technical solutions provided in the embodiments of the present application, and it is obvious to those skilled in the art that the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems with the evolution of technologies and the emergence of new application scenarios.

It will be appreciated by those skilled in the art that the embodiments shown in the figures are not intended to limit the embodiments of the present application and may include more or fewer steps than those shown, or some of the steps may be combined, or different steps may be included.

The above-described embodiments of the apparatus are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may also be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.

One of ordinary skill in the art will appreciate that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof.

The terms "first," "second," "third," "fourth," and the like in the description of the application and the above-described figures, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be understood that, in this application, "at least one" means one or more, "a plurality" means two or more. "and/or" for describing an association relationship of associated objects, indicating that there may be three relationships, e.g., "a and/or B" may indicate: only A, only B and both A and B are present, wherein A and B may be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of the singular or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c may be single or plural.

In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the above-described units is only one type of logical functional division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may also be implemented in the form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a separate product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device) to perform all or part of the steps of the method of the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing programs, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The preferred embodiments of the present application have been described above with reference to the accompanying drawings, and the scope of the claims of the embodiments of the present application is not limited thereto. Any modifications, equivalents and improvements that may occur to those skilled in the art without departing from the scope and spirit of the embodiments of the present application are intended to be within the scope of the claims of the embodiments of the present application.

Claims

1. A model training method, characterized in that the model training method comprises:

obtaining sample portrait data of a sample object;

constructing a decision sub-tree according to the sample portrait characteristics; said decision sub-tree comprising leaf nodes, said each of said leaf nodes being one of said sample representation features;

inputting the sample knowledge bars into a preset classification model for training to obtain the weight of the knowledge bar of each decision sub-tree;

2. The model training method of claim 1, wherein the constructing a decision sub-tree based on the sample portrait features comprises:

screening out parent features from the sample portrait features;

performing feature division on at least one parent feature to obtain at least one sub-feature; wherein each of the parent features comprises at least one of the child features;

3. The model training method of claim 2, wherein the filtering out parent features from the sample portrait features comprises:

calculating the information gain of the sample image characteristics;

4. A method as claimed in any one of claims 1 to 3, wherein the deriving a target decision tree model according to the weights of the knowledge bars of the K decision subtrees comprises:

obtaining K decision subtrees;

5. A user representation generation method, the user representation generation comprising:

acquiring target portrait data of a target user; inputting the target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; wherein the prediction knowledge item set comprises at least two knowledge items hit by the target user, and the target decision tree model is trained according to the model training method of any one of claims 1 to 4;

6. The user representation generation method of claim 5, wherein traversing the set of predicted knowledge bars according to the target weights to obtain a target knowledge bar comprises:

acquiring an absolute value of each target weight;

and sequentially traversing the target weight according to the weight reordering order to obtain the target knowledge bar.

7. Model training apparatus, characterized in that the training apparatus comprises:

8. A user representation generation device, the user representation generation device comprising:

the target data acquisition module is used for acquiring target portrait data of a target user; the decision module is used for inputting the target portrait data into a target decision tree model for decision making to obtain a prediction knowledge strip set; wherein the prediction knowledge item set comprises at least two knowledge items hit by the target user, and the target decision tree model is trained according to the model training method of any one of claims 1 to 4;

9. Electronic device, characterized in that it comprises a memory, a processor, a program stored on said memory and executable on said processor, and a data bus for enabling a connection communication between said processor and said memory, said program, when executed by said processor, realizing:

a step of the model training method according to any one of claims 1 to 4;

alternatively, the first and second electrodes may be,

a user representation generation method as claimed in any one of claims 5 to 6.

10. A storage medium that is a computer-readable storage medium for computer-readable storage, wherein the storage medium stores one or more programs that are executable by one or more processors to implement:

a step of the model training method according to any one of claims 1 to 4;

alternatively, the first and second electrodes may be,

a user representation generation method as claimed in any one of claims 5 to 6.