CN112328909B

CN112328909B - Information recommendation method and device, computer equipment and medium

Info

Publication number: CN112328909B
Application number: CN202011288944.5A
Authority: CN
Inventors: 刘波
Original assignee: Ping An Life Insurance Company of China Ltd
Current assignee: Ping An Life Insurance Company of China Ltd
Priority date: 2020-11-17
Filing date: 2020-11-17
Publication date: 2021-07-02
Anticipated expiration: 2040-11-17
Also published as: CN112328909A

Abstract

The invention relates to the field of communication, and discloses an information recommendation method, an information recommendation device, computer equipment and a medium, wherein the method comprises the following steps: the method comprises the steps of acquiring initial data of each user, preprocessing the initial data and evaluating stability of the initial data to obtain stable data, performing feature derivation on features contained in the stable data to obtain derived features, predicting feature values of the derived features in a machine learning mode, supplementing numerical values of the derived features according to the predicted numerical values to obtain supplemented data, adding the supplemented data and the stable data into a target data set as target data, predicting user preference in a recall and sequencing mode based on the target data set, determining recommendation information according to the predicted user preference and a service strategy, and pushing the recommendation information to the user.

Description

Information recommendation method and device, computer equipment and medium

Technical Field

The present invention relates to the field of data processing, and in particular, to an information recommendation method, apparatus, computer device, and medium.

Background

In the existing information recommendation field, when we recommend information to users, it is a traditional practice to make recommendations according to existing user filling information (including but not limited to information of age, sex, income, occupation, etc.) and user historical behavior data. In this mode, the targets of the general optimization are all single, that is, the tasks are all target optimization in a single scene. In an actual scenario, the optimization target of interest is a specific index, such as click rate, purchase rate, exposure rate, forwarding amount, etc. of the user.

When product information needs to be accurately and individually recommended facing different user genders, ages and regions, especially when a user is a tourist identity or a new user identity (user attribute information is almost blank), the recommendation difficulty is greatly increased. In the traditional method, a collaborative filtering (user-based collaborative filtering, article-based collaborative filtering, content-based recommendation) algorithm is adopted, and the methods have the disadvantages of weak personalization and wide range of recommendation results, are prone to recommending similar commodities, and have low recommendation accuracy, so that an accurate information recommendation method is urgently needed.

Disclosure of Invention

The embodiment of the invention provides an information recommendation method, an information recommendation device, computer equipment and a storage medium, and aims to improve the accuracy of information recommendation.

In order to solve the above technical problem, an embodiment of the present application provides an information recommendation method, including:

acquiring initial data of each user, and performing data preprocessing and stability evaluation on the initial data to obtain stable data, wherein the characteristics contained in the obtained stable data are stable characteristics;

performing characteristic derivation treatment on the stable characteristics by adopting a characteristic derivation mode to obtain derived characteristics;

predicting information of the missing characteristic values in the derived characteristics in a machine learning mode to obtain a prediction result;

filling the missing characteristic values according to a prediction result to obtain filled data, taking the filled data and the stable data as target data, and storing the target data into a target data set;

inputting the target data and each recall method into a preset recommendation model aiming at any target data of the target data set, and scoring each recall method through the preset recommendation model to obtain a prediction score;

sorting according to the sequence of the prediction scores from large to small, and sequentially selecting a preset number of recall methods from the sorted prediction scores to add into a recall method sequence;

and generating recommendation information according to a preset recommendation condition and by combining the recall method sequence, and recommending the recommendation information to a user corresponding to the target data.

Optionally, the performing data preprocessing and stability evaluation on the initial data to obtain stable data includes:

and classifying the initial data according to a preset label type to obtain initial class information.

Carrying out missing value processing on each initial category information to obtain basic category information, wherein each basic category information at least comprises one basic feature;

and calculating the stability of the basic features of each basic category information, screening out the basic features of which the stability exceeds a preset stability threshold value as stable features, and taking the basic features contained in the basic features as the basic category information of the stable features as stable data.

Optionally, the method may further include the steps of calculating a stability of the basic feature of each basic category information, and screening out the basic feature of which the stability exceeds a preset stability threshold, where the step of screening out the basic feature as the stable feature includes:

calculating an information value IV of each basic characteristic, and performing characteristic screening according to the information value IV to obtain a key characteristic;

and calculating the stability index PSI of the key feature in a preset mode, and taking the stability index PSI as the key feature when the stability index PSI exceeds a preset threshold value.

Optionally, the preset model is a gradient boosting decision tree model, and scoring each recall method through the preset recommendation model to obtain a prediction score includes:

taking a simulation result corresponding to each recall method as a reference feature, taking a data feature contained in the target data as a target feature, and taking the reference feature and the target feature as training features;

inputting the training features into a gradient boost decision tree model, and training the training features through the gradient boost decision tree model to obtain n decision trees;

and taking the characteristics contained in the path of each decision tree as independent variables, and predicting the result of a preset event based on a binary logistic regression model to obtain a prediction score corresponding to each recall method.

Optionally, the training features through the gradient boosting decision tree model to obtain n decision trees includes:

generating an original decision tree by adopting a classification regression tree algorithm for the training characteristics;

putting the original decision tree into a decision tree model, and taking the original decision tree as a current decision tree;

calculating a residual vector of the current decision tree based on the training features;

fitting a new decision tree according to the residual vector, and putting the new decision tree into the decision tree model;

if the total number of decision trees in the decision tree model is lower than a preset threshold value, taking the new decision tree as the current decision tree, returning to the step of calculating the residual vector of the current decision tree based on the training characteristics, and continuing to execute the step;

and if the total number of the decision trees in the decision tree model reaches a preset threshold value n, stopping fitting the new decision tree to obtain the decision tree model comprising n decision trees.

In order to solve the above technical problem, an embodiment of the present application further provides an information recommendation apparatus, including:

the device comprises a characteristic acquisition module, a stability evaluation module and a data preprocessing module, wherein the characteristic acquisition module is used for acquiring initial data of each user, and carrying out data preprocessing and stability evaluation on the initial data to obtain stable data, and the acquired stable data comprises characteristics which are stable characteristics;

the characteristic derivation module is used for performing characteristic derivation processing on the stable characteristic in a characteristic derivation mode to obtain a derived characteristic;

the characteristic value prediction module is used for predicting information of the characteristic values missing from the derived characteristics in a machine learning mode to obtain a prediction result;

the characteristic completion module is used for filling the missing characteristic values according to a prediction result to obtain filled data, taking the filled data and the stable data as target data, and storing the target data into a target data set;

the data evaluation module is used for inputting the target data and each recall method into a preset recommendation model aiming at any target data of the target data set, and scoring each recall method through the preset recommendation model to obtain a prediction score;

the score sorting module is used for sorting according to the sequence of the prediction scores from large to small, and sequentially selecting a preset number of recall methods from the sorted prediction scores to be added into a recall method sequence;

and the information recommendation module is used for generating recommendation information according to a preset recommendation condition and by combining the recall method sequence, and recommending the recommendation information to a user corresponding to the target data.

Optionally, the feature obtaining module includes:

and the classification unit is used for classifying the initial data according to the type of a preset label to obtain initial class information.

The preprocessing unit is used for carrying out missing value processing on each initial category information to obtain basic category information, wherein each basic category information at least comprises one basic feature;

and the stability screening unit is used for calculating the stability of the basic characteristics of each basic category information, screening out the basic characteristics of which the stability exceeds a preset stability threshold value as the stable characteristics, and using the basic characteristics contained in the basic characteristics as the basic category information of the stable characteristics as stable data.

Optionally, the stability screening unit comprises:

the key characteristic determining subunit is used for calculating an information value IV of each basic characteristic and performing characteristic screening according to the information value IV to obtain key characteristics;

and the stability calculating subunit is used for calculating the stability index PSI of the key feature in a preset mode, and taking the stability index PSI as the key feature when exceeding a preset threshold value.

Optionally, the data evaluation module comprises:

a training feature extraction unit, configured to use a simulation result corresponding to each recall method as a reference feature, use a data feature included in the target data as a target feature, and use the reference feature and the target feature as training features;

the decision tree construction unit is used for inputting the training characteristics into a gradient lifting decision tree model, and training the training characteristics through the gradient lifting decision tree model to obtain n decision trees;

and the score prediction unit is used for predicting the result of a preset event by taking the characteristics contained in the path of each decision tree as independent variables based on a binary logistic regression model to obtain the prediction score corresponding to each recall method.

Optionally, the decision tree building unit includes:

a first decision tree generating subunit, configured to generate an original decision tree by using a classification regression tree algorithm for the training features;

a decision tree alternation subunit, configured to place the original decision tree into a decision tree model, and use the original decision tree as a current decision tree;

a residual calculating subunit, configured to calculate a residual vector of the current decision tree based on the training features;

a new decision tree generation subunit, configured to fit a new decision tree according to the residual vector, and place the new decision tree into the decision tree model;

an update iteration subunit, configured to, if the total number of decision trees in the decision tree model is lower than a preset threshold, use the new decision tree as the current decision tree, return to the step of calculating a residual vector of the current decision tree based on the training features, and continue to execute the step;

and the decision tree model generation subunit is used for stopping fitting the new decision tree if the total number of the decision trees in the decision tree model reaches a preset threshold value n, so as to obtain the decision tree model comprising n decision trees.

In order to solve the technical problem, an embodiment of the present application further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the information recommendation method when executing the computer program.

In order to solve the above technical problem, an embodiment of the present application further provides a computer-readable storage medium, where a computer program is stored, and the computer program, when executed by a processor, implements the steps of the information recommendation method.

The information recommendation method, the device, the computer equipment and the storage medium provided by the embodiment of the invention acquire initial data of each user, perform data preprocessing and stability evaluation on the initial data to obtain stable data, further perform feature derivation on features contained in the stable data to obtain derived features, predict the feature values of the derived features in a machine learning mode, further supplement the values of the derived features according to the predicted values to obtain supplemented data, add the supplemented data and the stable data into a target data set as target data, adopt the target data with higher quality and richer features to be beneficial to improving the accuracy of preference prediction, further predict user preferences in a recall and sorting mode based on the target data set, further according to the predicted user preferences and service strategies, and the recommendation information is determined and pushed to the user, so that the accuracy of information recommendation is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without inventive labor.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow diagram of one embodiment of an information recommendation method of the present application;

FIG. 3 is a schematic block diagram of an embodiment of an information recommendation device according to the present application;

FIG. 4 is a schematic block diagram of one embodiment of a computer device according to the present application.

Detailed Description

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 in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "including" and "having," and any variations thereof, in the description and claims of this application and the description of the above figures are intended to cover non-exclusive inclusions. The terms "first," "second," and the like in the description and claims of this application or in the above-described drawings are used for distinguishing between different objects and not for describing a particular order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, as shown in fig. 1, a system architecture 100 may include

terminal devices

101, 102, 103, a network 104 and a server 105. The network 104 serves as a medium for providing communication links between the

terminal devices

101, 102, 103 and the server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, to name a few.

The user may use the

terminal devices

101, 102, 103 to interact with the server 105 via the network 104 to receive or send messages or the like.

The

terminal devices

101, 102, 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, E-book readers, MP3 players (Moving Picture E interface shows a properties Group Audio Layer III, motion Picture experts compress standard Audio Layer 3), MP4 players (Moving Picture E interface shows a properties Group Audio Layer IV, motion Picture experts compress standard Audio Layer 4), laptop portable computers, desktop computers, and the like.

The server 105 may be a server providing various services, such as a background server providing support for pages displayed on the

terminal devices

101, 102, 103.

The information recommendation method provided by the embodiment of the present application is executed by a server, and accordingly, the information recommendation apparatus is disposed in the server.

It should be understood that the number of terminal devices, networks, and servers in fig. 1 is merely illustrative. Any number of terminal devices, networks and servers may be provided according to implementation needs, and the

terminal devices

101, 102 and 103 in this embodiment may specifically correspond to an application system in actual production.

Referring to fig. 2, fig. 2 shows an information recommendation method according to an embodiment of the present invention, which is described by taking the method applied to the server in fig. 1 as an example, and is detailed as follows:

s201: the method comprises the steps of collecting initial data of each user, and conducting data preprocessing and stability evaluation on the initial data to obtain stable data, wherein the obtained stable data comprise stable characteristics.

Specifically, initial data of each user is acquired in a data acquisition mode, the initial data are classified according to the type of a preset label to obtain initial category information, missing value checking and stability evaluation are carried out according to the characteristics of each piece of initial category information, and stable data with stable data characteristics are obtained.

The data acquisition mode specifically includes but is not limited to: the method comprises the following steps of crawling web crawlers, reading background logs, collecting distributed data based on big data, reading a database and the like, wherein the crawling can be specifically selected according to actual conditions, and the crawling is not limited here.

The preset tag type may be set according to an actual situation, for example, in a specific embodiment, the preset tag type includes a basic attribute, a browsing attribute, a consuming attribute, a credit attribute, and the like.

It should be noted that, in the present embodiment, each initial category information may include one or more basic features, for example, in the above example, the basic properties may include basic features such as age, gender, height, weight, and native place, and the browsing properties include basic features such as browsing, collecting, forwarding, commenting, and praise.

S202: and performing characteristic derivation treatment on the stable characteristics by adopting a characteristic derivation mode to obtain derived characteristics.

Specifically, according to the category of the application scene and the complexity of the features, a plurality of stability features are selected from the key feature sequence, and feature derivation is performed on the selected stability features to obtain derived features.

In the present embodiment, the feature derivation includes, but is not limited to, feature combination, feature intersection, image feature generation, text feature generation, and the like.

The feature combination can be realized by four arithmetic combinations, logical and or combinations, polynomial structures, differences between the features and the mean value thereof and the like between every two features.

It should be noted that, the initial data of different data sources may have different basic features, and the feature data corresponding to each user can be greatly enriched in a feature derivation manner, which is beneficial to improving the accuracy of subsequent information recommendation.

S203: and predicting the information of the missing characteristic values in the derived characteristics by adopting a machine learning mode to obtain a prediction result.

Specifically, after the derived features are obtained, the derived features are used as part of the features of the user data, and for the lack of numerical values corresponding to some features, numerical value prediction is performed in a machine learning mode according to some existing data to obtain an information prediction result.

The machine learning method for numerical prediction includes, but is not limited to: lightgbm model, cnn + rnn combinatorial algorithm, etc.

For example, since a certain user data includes a feature a (a1, a2, A3), and the feature a becomes a feature a '(a 1, a2, A3, C6) after feature derivation, there is no value corresponding to C6 in the initial data of the user, and therefore, it is necessary to predict the value of C6, input all the data corresponding to the user into a machine learning model, and predict the value of C6 in the feature a' by the machine learning model to obtain a predicted value.

S204: and filling the missing characteristic values according to the prediction result to obtain filled data, taking the filled data and the stable data as target data, and storing the target data into a target data set.

Specifically, according to the obtained information prediction result, filling the missing value of the derived feature to obtain filled data, and storing the filled data and the stable data into a target data set as final target data.

It should be understood that missing values of the derived features are filled, the quality of data corresponding to the derived features can be improved, the filled data can be stored into a target data set as target data, the data diversity of the target data set is improved, and corresponding appropriate information can be well recommended to a user under the condition that the privacy of the user is not actively acquired aiming at the missing important information (such as information of age, gender, region, marital state and the like) of the user, so that the accuracy of information recommendation is improved, and the service experience of the user is improved.

S205: and aiming at any target data of the target data set, inputting the target data and each recall method into a preset recommendation model, and scoring each recall method through the preset recommendation model to obtain a prediction score.

Specifically, a plurality of recall methods (recall strategies) are preset at the server, a simple model is adopted to quickly screen a recommended item candidate set to a specified level (a hundred level or a ten level is determined according to specific product conditions), a preset recommendation evaluation model is adopted for any target data of a target data set, each recall method is scored, and then the recall methods are evaluated according to scores.

Wherein, recommending the preset model includes but is not limited to: wide & Deep model, dien (Deep Interest Evolution Network for click) model, GBDT + LR model, DNN model, etc.

It should be noted that, in this embodiment, the information recommendation process includes a recall stage and a ranking stage, where the recall stage may be understood as a step of roughly selecting a batch of contents to be recommended from massive information for a user according to behavior data of the user, and selecting a small candidate set, which is equivalent to rough ranking. And in the sequencing stage, more accurate calculation is performed on the basis, each content is accurately scored, and the method is equivalent to accurate sequencing. And in the recall stage, a recall method is adopted for coarse sorting, and in the sorting stage, a preset recommendation evaluation model is adopted for fine sorting.

Preferably, in this embodiment, a combination mode of the GBDT decision tree and the two-classification logistic regression model LR is used as the preset recommendation model to evaluate the recall method, and specific implementation details may refer to the description of the subsequent embodiments, and are not described herein again to avoid repetition.

S206: and sorting according to the sequence of the prediction scores from large to small, and sequentially selecting a preset number of recall methods from the sorted prediction scores to add into the recall method sequence.

Specifically, sorting according to the sequence of the prediction scores from large to small, and sequentially selecting a preset number of recall methods from the sorted prediction scores to add into the recall method sequence.

The preset number can be set according to actual requirements, and is not limited here.

S207: and generating recommendation information according to a preset recommendation condition and by combining the recall method sequence, and recommending the recommendation information to a user corresponding to the target data.

Specifically, for different users, even though a superior recall method is obtained, it is necessary to combine some service policies, such as various service policies of going to read, recommending diversification, adding advertisements, and the like, and then form a final recommendation result and push recommendation information to the user.

In the embodiment, the initial data of each user is collected, the initial data is subjected to data preprocessing and stability evaluation to obtain stable data, the characteristics contained in the stable data are subjected to characteristic derivation to obtain derived characteristics, the derived characteristics are predicted in a machine learning mode, the numerical values of the derived characteristics are supplemented according to the predicted numerical values to obtain supplemented data, the supplemented data and the stable data are used as target data and added into a target data set, the target data with high quality and rich characteristics are adopted, the preference prediction accuracy is improved, the preference of the user is predicted in a recall and sequencing mode based on the target data set, and then the recommendation information is determined and pushed to the user according to the predicted user preference and service strategies, the accuracy of information recommendation is improved.

In some optional implementation manners of this embodiment, in step S201, performing data preprocessing and stability evaluation on the initial data, and obtaining stable data includes:

and classifying the initial data according to the type of the preset label to obtain initial class information.

The preset label type is a label classified in advance according to requirements.

Specifically, after the obtained data is classified to obtain initial category information, data preprocessing needs to be performed on the data to ensure data quality, and in consideration of the problem of partial data loss caused by reasons of inconsistent data sources, non-timely update and the like, in this embodiment, a missing value is first processed on each initial category information to obtain basic category information.

In this embodiment, each piece of basic category information includes one or more basic features, and after each piece of basic category information is obtained, the stability of the basic features needs to be evaluated, and the basic features with good stability are retained as the stable features.

The preset stability threshold may be set according to actual conditions, and is not limited herein, and in this embodiment, the value is preferably 0.25.

Further, missing value processing is performed on each initial category information to obtain basic category information, and the specific process includes:

acquiring a characteristic value corresponding to each basic characteristic in the initial category information aiming at each initial category information;

carrying out data verification on the characteristic value, and taking the characteristic value failed in verification as a missing value;

and counting the missing values corresponding to each basic feature, taking the basic feature with the ratio of the missing values to the feature values exceeding a preset ratio as an invalid feature, and removing the invalid feature from the initial category information to obtain basic category information.

The data verification is performed on the characteristic value, and specifically includes but is not limited to: null value verification, numerical normalization verification and numerical uniqueness verification.

It should be understood that when the ratio of the missing value to the feature value exceeds the preset ratio, that is, the missing value is more, at this time, it is determined that the basic feature corresponding to the missing value has a quality problem, and the basic feature is taken as an invalid feature and removed from the initial category information, so as to avoid that the basic feature has a negative influence on the information recommendation model prediction subsequently.

The null value check can be realized in a regular expression mode, the numerical value normalization check is judged by matching the numerical value with a preset rule, and the numerical value uniqueness check is to judge whether the same repeated numerical value exists or not.

In the embodiment, the initial data is subjected to data preprocessing and stability evaluation to obtain stable data, so that the data quality is ensured, and the accuracy of subsequent characteristic derivation is improved.

In some optional implementation manners of this embodiment, the stability of the basic feature of each basic category information is calculated, and the basic feature with the stability exceeding the preset stability threshold is screened out, and the screening as the stable feature includes:

calculating an information value IV of each basic characteristic, and screening the characteristics according to the information value IV to obtain key characteristics;

and calculating a stability index PSI of the key feature in a preset mode, and taking the key feature with the stability index PSI exceeding a preset threshold value as the stability feature.

Specifically, for the characteristic that the data type in the basic characteristic is continuous, the box separation processing is carried out, and the continuous characteristic is converted into the discrete characteristic; performing independent thermal coding on all discrete features to obtain a digital variable; according to the digital variables, calculating an information value IV corresponding to each feature, further performing importance sorting on the information values IV to obtain a sorting result of the importance from high to low, further screening attribute features corresponding to the information values IV according to the sorting result to obtain key features, then calculating a stability index PSI of each key feature, and taking the key feature of which the stability index exceeds a preset threshold as the stable feature. And feature combination derivation is carried out through the stable features in the follow-up process, so that the quality of the derived features is improved.

The preset mode may specifically be that data of the key features of each month is selected according to the dimension of time, and the stability index PSI of the key features is calculated month by month.

It should be noted that in the information recommendation scene, the data dimension and the data volume involved are large, the key features are screened through the information value IV, the important features are retained, the data computation amount can be greatly reduced, and the efficiency of obtaining the stable features is improved.

Optionally, in this embodiment, the importance of the stability features is sorted from multiple dimensions by a preset feature sorting manner to obtain a key feature sequence, so that when feature derivation processing is performed subsequently, features with the top sorting are preferentially selected to perform feature derivation.

The preset feature sorting mode includes, but is not limited to: the lightgbm algorithm, the xgboost algorithm, and the like, it should be noted that the tree model naturally performs importance ranking on the features to split the data set and construct branches, and then obtains importance ranking according to the scores of the branches, and the specific mode can be selected according to actual needs, which is not limited herein.

It should be noted that after the sequence of each stability feature is calculated by the tree model, the stability indicators with the importance degree lower than the preset value are removed, so as to improve the quality of the features in the key feature sequence.

In some optional implementation manners of this embodiment, in step S205, the preset model is a gradient boosting decision tree model, and scoring is performed on each recall method through the preset recommendation model, and obtaining the prediction score includes:

taking a simulation result corresponding to each recall method as a reference feature, taking a data feature contained in target data as a target feature, and taking the reference feature and the target feature as training features;

inputting the training characteristics into a gradient boost decision tree model, and training the training characteristics through the gradient boost decision tree model to obtain n decision trees;

and (3) taking the characteristics contained in the path of each decision tree as independent variables, and predicting the result of the preset event based on a binary logistic regression model to obtain a prediction score corresponding to each recall method.

The Gradient Boosting Decision Tree (GBDT) algorithm is an iterative Decision Tree algorithm, and the algorithm is composed of a plurality of Decision trees, and the conclusions of all the trees are accumulated to serve as a final prediction result.

The decision tree in the gradient lifting decision tree belongs to a regression tree, a predicted value of the classification feature corresponding to each node is obtained at each node of the trees, and for the classification feature of which the specific numerical value is not determined, the average value of the classification feature is used as the predicted value of the classification feature.

After a decision tree model is generated, feature values of features contained in different paths are combined according to each decision tree to obtain combined features, the values of the same combined features of different trees are accumulated, the final accumulated value is used as the feature value of the combined features, the feature value is used as an independent variable in a binary Regression (LR) model to calculate a probability value, and the prediction score of a recall method corresponding to the combined features is determined according to the probability value.

In some optional implementation manners of this embodiment, training the training features through a gradient boosting decision tree model, and obtaining n decision trees includes:

fitting a new decision tree according to the residual vector, and putting the new decision tree into a decision tree model;

if the total number of the decision trees in the decision tree model is lower than a preset threshold value, taking the new decision tree as the current decision tree, returning to the step of calculating the residual vector of the current decision tree based on the training characteristics, and continuing to execute the step;

In this embodiment, a Classification And Regression Tree (CART) algorithm is also called a least square Regression Tree, And the CART algorithm considers that each node has a possibility of becoming a leaf node, And assigns a class to each node. The method for assigning the classes can use the classes which appear most in the current nodes, can also refer to the classification errors of the current nodes or other more complex methods, and adopts a mode based on binary recursive segmentation. Therefore, the decision tree generated by the CART algorithm is a binary tree with a concise structure, and the CART algorithm is suitable for a scene with the sample characteristics being taken as yes or no.

Each node of the classification regression tree obtains a predicted value, which is equal to the average age of all people belonging to the node, taking age as an example. When branching, the digital variable values corresponding to each feature are exhausted to find the best segmentation point, but the best measurement standard is not the maximum entropy any more, but the minimized square error is used as the segmentation error, namely, the more the number of predicted errors is, the more the error is separated from the spectrum, the larger the segmentation error is, and the most reliable branching basis can be found by using the minimized square error as the segmentation error. If the age of the person at the final leaf node is not unique, the average age of all persons at the node is used as the predicted age of the leaf node.

Specifically, in the gradient lifting decision tree algorithm, a gradient lifting method is adopted to construct a weak classifier, in each iteration, a loss function is used to calculate a loss value of a sample in a current decision tree on each classification feature, and the loss value is used as a predicted value of a next tree to fit and generate a new decision tree, wherein the loss value is an absolute value of a residual vector.

Wherein the loss function includes, but is not limited to: 0-1Loss Function (0-1Loss Function), square Loss Function (Quadratic Loss Function), Absolute Loss Function (Absolute Loss Function), and Logarithmic Loss Function (Logarithmic Loss Function).

Preferably, the loss function used in the present invention is a logarithmic loss function, which uses a maximum likelihood estimation method.

Furthermore, a residual vector corresponding to the training features is used as a predicted value of the new decision tree on the classification features, and the new decision tree is fitted, so that the new decision tree further improves the current decision tree, and the accuracy of the decision tree model on the feature description of the sample data is improved.

In the embodiment, an original decision tree is generated through a CART algorithm, the original decision tree is placed in a decision tree model and is used as a current decision tree, a residual vector of the current decision tree is further calculated, a new decision tree is fitted according to the residual vector of the current decision tree and is placed in the decision tree model, the new decision tree is fitted in a circulating mode until the total number of the decision trees in the decision tree model reaches a preset threshold, fitting of the new decision tree is stopped, a gradient lifting decision tree algorithm is adopted in the whole process, each new decision tree is fitted to the current decision tree, errors of the decision tree model are reduced step by step, and prediction accuracy is improved.

It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

Fig. 3 is a schematic block diagram of an information recommendation apparatus corresponding to the information recommendation method according to the above-described embodiment. As shown in fig. 3, the information recommendation apparatus includes a feature acquisition module 31, a feature derivation module 32, a feature value prediction module 33, a feature completion module 34, a data evaluation module 35, a score ranking module 36, and an information recommendation module 37. The functional modules are explained in detail as follows:

the characteristic acquisition module 31 is configured to acquire initial data of each user, and perform data preprocessing and stability evaluation on the initial data to obtain stable data, where a characteristic included in the acquired stable data is a stable characteristic;

the characteristic derivation module 32 is configured to perform characteristic derivation processing on the stable characteristic in a characteristic derivation manner to obtain a derived characteristic;

the characteristic value prediction module 33 is configured to perform information prediction on a characteristic value missing from the derived characteristic by using a machine learning method to obtain a prediction result;

the characteristic completion module 34 is configured to fill the missing characteristic values according to the prediction result to obtain filled data, use the filled data and the stable data as target data, and store the target data into a target data set;

the data evaluation module 35 is configured to, for any target data of the target data set, input the target data and each recall method into a preset recommendation model, and score each recall method through the preset recommendation model to obtain a prediction score;

the score sorting module 36 is used for sorting according to the sequence of the prediction scores from large to small, and sequentially selecting a preset number of recall methods from the sorted prediction scores to add into the recall method sequence;

and the information recommendation module 37 is configured to generate recommendation information according to a preset recommendation condition and by combining with the recall method sequence, and recommend the recommendation information to a user corresponding to the target data.

Optionally, the feature obtaining module 31 includes:

and the classification unit is used for classifying the initial data according to the preset label type to obtain initial class information.

Optionally, the stability screening unit comprises:

the key characteristic determining subunit is used for calculating an information value IV of each basic characteristic and screening the characteristics according to the information value IV to obtain key characteristics;

and the stability calculating subunit is used for calculating the stability index PSI of the key feature in a preset mode, and taking the key feature with the stability index PSI exceeding a preset threshold value as the stability feature.

Optionally, the data evaluation module 35 comprises:

the training feature extraction unit is used for taking a simulation result corresponding to each recall method as a reference feature, taking a data feature contained in the target data as a target feature, and taking the reference feature and the target feature as training features;

the decision tree construction unit is used for inputting the training characteristics into the gradient lifting decision tree model and training the training characteristics through the gradient lifting decision tree model to obtain n decision trees;

and the score prediction unit is used for predicting the result of the preset event based on a binary logistic regression model by taking the characteristics contained in the path of each decision tree as independent variables to obtain the prediction score corresponding to each recall method.

Optionally, the decision tree building unit includes:

the first decision tree generation subunit is used for generating an original decision tree by adopting a classification regression tree algorithm on the training characteristics;

the decision tree alternation subunit is used for putting the original decision tree into the decision tree model and taking the original decision tree as the current decision tree;

a residual error calculating subunit, configured to calculate a residual error vector of the current decision tree based on the training features;

a new decision tree generation subunit, configured to fit a new decision tree according to the residual vector, and place the new decision tree in the decision tree model;

the updating iteration subunit is used for taking the new decision tree as the current decision tree if the total number of the decision trees in the decision tree model is lower than a preset threshold value, returning to the step of calculating the residual vector of the current decision tree based on the training characteristics and continuing to execute the step;

For specific limitations of the information recommendation device, reference may be made to the above limitations of the information recommendation method, which are not described herein again. The modules in the information recommendation device can be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In order to solve the technical problem, an embodiment of the present application further provides a computer device. Referring to fig. 4, fig. 4 is a block diagram of a basic structure of a computer device according to the present embodiment.

The computer device 4 comprises a memory 41, a processor 42, a network interface 43 communicatively connected to each other via a system bus. It is noted that only the computer device 4 having the components connection memory 41, processor 42, network interface 43 is shown, but it is understood that not all of the shown components are required to be implemented, and that more or fewer components may be implemented instead. As will be understood by those skilled in the art, the computer device is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and the hardware includes, but is not limited to, a microprocessor, an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.

The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer equipment can carry out man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch panel or voice control equipment and the like.

The memory 41 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or D interface display memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read Only Memory (ROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a Programmable Read Only Memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the computer device 4. Of course, the memory 41 may also include both internal and external storage devices of the computer device 4. In this embodiment, the memory 41 is generally used for storing an operating system installed in the computer device 4 and various types of application software, such as program codes for controlling electronic files. Further, the memory 41 may also be used to temporarily store various types of data that have been output or are to be output.

The processor 42 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data Processing chip in some embodiments. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is configured to execute the program code stored in the memory 41 or process data, such as program code for executing control of an electronic file.

The network interface 43 may comprise a wireless network interface or a wired network interface, and the network interface 43 is generally used for establishing communication connection between the computer device 4 and other electronic devices.

The present application further provides another embodiment, which is to provide a computer-readable storage medium storing an interface display program, where the interface display program is executable by at least one processor to cause the at least one processor to execute the steps of the information recommendation method as described above.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present application.

It is to be understood that the above-described embodiments are merely illustrative of some, but not restrictive, of the broad invention, and that the appended drawings illustrate preferred embodiments of the invention and do not limit the scope of the invention. This application is capable of embodiments in many different forms and is provided for the purpose of enabling a thorough understanding of the disclosure of the application. Although the present application has been described in detail with reference to the foregoing embodiments, it will be apparent to one skilled in the art that the present application may be practiced without modification or with equivalents of some of the features described in the foregoing embodiments. All equivalent structures made by using the contents of the specification and the drawings of the present application are directly or indirectly applied to other related technical fields and are within the protection scope of the present application.

Claims

1. An information recommendation method, comprising:

generating recommendation information according to a preset recommendation condition and in combination with the recall method sequence, and recommending the recommendation information to a user corresponding to the target data;

wherein, the data preprocessing and stability evaluation of the initial data to obtain stable data comprises:

classifying the initial data according to a preset label type to obtain initial class information;

calculating the stability of the basic features of each basic category information, screening out the basic features of which the stability exceeds a preset stability threshold value as stable features, and taking the basic features contained in the basic features as the basic category information of the stable features as stable data;

the preset recommendation model is a gradient lifting decision tree model, and scoring is performed on each recall method through the preset recommendation model to obtain a prediction score, wherein the step of obtaining the prediction score comprises the following steps:

2. The information recommendation method according to claim 1, wherein the calculating of the stability of the basic feature of each basic category information, and the screening of the basic feature whose stability exceeds a preset stability threshold as the stable feature comprises:

3. The information recommendation method of claim 1, wherein the training features through the gradient boosting decision tree model to obtain n decision trees comprises:

4. An information recommendation apparatus, comprising:

the information recommendation module is used for generating recommendation information according to a preset recommendation condition and in combination with the recall method sequence, and recommending the recommendation information to a user corresponding to the target data;

wherein the feature acquisition module comprises:

the classification unit is used for classifying the initial data according to a preset label type to obtain initial class information;

the stability screening unit is used for calculating the stability of the basic characteristics of each basic category information, screening out the basic characteristics of which the stability exceeds a preset stability threshold value as stable characteristics, and taking the basic characteristics contained in the basic characteristics as the basic category information of the stable characteristics as stable data;

wherein the data evaluation module comprises:

and the score prediction unit is used for predicting the result of the preset event based on the two-classification logistic regression model by taking the characteristics contained in the path of each decision tree as independent variables to obtain the prediction score corresponding to each recall method.

5. The information recommendation device of claim 4, wherein the stability filtering unit comprises:

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the information recommendation method according to any one of claims 1 to 3 when executing the computer program.

7. A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, implements the information recommendation method according to any one of claims 1 to 3.