CN110263235A

CN110263235A - Information pushes object updating method, device and computer equipment

Info

Publication number: CN110263235A
Application number: CN201910486490.3A
Authority: CN
Inventors: 杨春风
Original assignee: Shenzhen Tencent Computer Systems Co Ltd
Current assignee: Shenzhen Tencent Computer Systems Co Ltd
Priority date: 2019-06-05
Filing date: 2019-06-05
Publication date: 2019-09-20
Anticipated expiration: 2039-06-05
Also published as: CN110263235B

Abstract

This application involves a kind of information push object updating method, device, computer readable storage medium and computer equipments, which comprises obtains the target identification collection of target information transmission service；The feature that the target identification concentrates the corresponding user of each mark is extracted, prediction model is obtained according to feature training；The feature of the user to be selected is input to the prediction model and predicted by the mark and feature for obtaining user to be selected, obtains the prediction numerical value of the user to be selected；The mark that the user to be selected of preset quantity is chosen according to prediction numerical value, obtains identification sets to be selected；The target identification collection of the target information transmission service is updated according to the identification sets to be selected.This programme can be realized automatically updating for target information push object.

Description

Information push object updating method and device and computer equipment

Technical Field

The present application relates to the field of computer technologies, and in particular, to an information push object updating method, an information push object updating apparatus, a computer-readable storage medium, and a computer device.

Background

With the development of computer technology, information is spread and shared more and more rapidly. The information can be transmitted in a mode of pushing the information to a specific crowd, and the information can be pushed in a targeted mode so as to increase the exposure rate of the information.

However, currently, the pushing of information needs to be adjusted continuously according to the demands of people, and feedback data needs to be collected manually to adjust the pushing object of information.

Disclosure of Invention

Based on this, it is necessary to provide an information push object updating method, apparatus, computer-readable storage medium, and computer device for automatically adjusting a push object in response to a technical problem that the push object needs to be adjusted by manually collecting feedback data.

An information push object updating method comprises the following steps:

acquiring a first identifier set of a target information push service;

extracting the characteristics of the user corresponding to each identification in the target identification set, and training according to the characteristics to obtain a prediction model;

acquiring the identification and the characteristics of a user to be selected, and inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected;

selecting the identifiers of a preset number of users to be selected according to the predicted numerical value to obtain an identifier set to be selected;

and updating the target identification set of the target information push service according to the identification set to be selected.

An information push updating device, the device comprising:

the acquisition module is used for acquiring a target identification set of a target information push service;

the training module is used for extracting the characteristics of the user corresponding to each identifier in the target identifier set and obtaining a prediction model according to the characteristic training;

the prediction module is used for acquiring the identification and the characteristics of the user to be selected, inputting the characteristics of the user to be selected into the prediction model for prediction, and obtaining the prediction value of the user to be selected;

the selection module is used for selecting the identifiers of the users to be selected in the preset number according to the prediction value to obtain an identifier set to be selected;

and the updating module is used for updating the target identification set of the target information pushing service according to the identification set to be selected.

A computer-readable storage medium, storing a computer program which, when executed by a processor, causes the processor to perform the steps of any of the methods described above.

A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of any of the methods described above.

According to the information push object updating method, the information push object updating device, the computer readable storage medium and the computer equipment, the target identification set of the target information push service is obtained at regular time, the characteristics of the user corresponding to each identification in the target identification set are extracted, and the prediction model is obtained according to characteristic training. And inputting the characteristics of the user to be selected into a prediction model for prediction by acquiring the identification and the characteristics of the user to be selected, so as to obtain a prediction value of the user to be selected. And selecting the identifiers of the preset number of users to be selected according to the predicted numerical value to obtain an identifier set to be selected, and updating a target identifier set of the target information push service according to the identifier set to be selected. According to the scheme, the released target identification set is obtained and used as a sample to automatically train and update the prediction model so as to automatically update the target identification set without manual operation, and therefore automatic updating of a target information pushing object is achieved.

Drawings

FIG. 1 is a diagram of an application environment of an information push object update method in an embodiment;

FIG. 2 is a flowchart illustrating an information push object update method according to an embodiment;

FIG. 3 is a diagram illustrating an interface for updating an information push object according to an embodiment;

FIG. 4 is a flowchart illustrating the steps of obtaining a predicted value for a candidate user in one embodiment;

FIG. 5 is a schematic flow chart diagram illustrating the steps of training a predictive model in one embodiment;

FIG. 6 is a flow chart illustrating the steps of detecting the number of positive samples in one embodiment;

FIG. 7 is a schematic flow chart diagram illustrating the steps of a stitching process for features in one embodiment;

FIG. 8 is a diagram illustrating automatic updating of a trained predictive model in one embodiment;

FIG. 9 is a flowchart illustrating the steps of updating a target identification set in one embodiment;

FIG. 10 is a diagram of automatically updating push objects in one embodiment;

FIG. 11 is a diagram illustrating a comparison between the present scheme and a conventional information push object update method in one embodiment;

FIG. 12 is a block diagram of an apparatus for pushing an object for updating information according to an embodiment;

FIG. 13 is a block diagram of a computer device in one embodiment.

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.

Fig. 1 is an application environment diagram of an information push object updating method in an embodiment. Referring to fig. 1, the information push object updating method is applied to an information push object updating system. The information push object updating system comprises a terminal 110 and a server 120. The terminal 110 and the server 120 are connected through a network. The terminal 110 may specifically be a desktop terminal or a mobile terminal, and the mobile terminal may specifically be at least one of a mobile phone, a tablet computer, a notebook computer, and the like. The server 120 may be implemented as a stand-alone server or a server cluster composed of a plurality of servers.

In this embodiment, the terminal 110 may obtain a target identifier set of the target information push service, and extract characteristics of a user corresponding to each identifier in the target identifier set. Next, the terminal 110 sends the characteristics of the user corresponding to each identifier in the extracted target identifier set to the server 120. The server 120 receives the characteristics sent by the terminal 110 and obtains a prediction model according to the characteristics training. Then, the terminal 110 obtains the candidate user and the identifier of the candidate user, and extracts the feature of the candidate user. The terminal 110 sends the identification of the user to be selected and the corresponding characteristics of the user to be selected to the server 120, the server 120 receives the identification and the characteristics of the user to be selected, the characteristics are input to the prediction model, and the prediction value of the user to be selected is obtained through prediction of the prediction model. The terminal 110 receives the identifiers of the users to be selected and the corresponding predicted values returned by the server 120, and selects the identifiers of the users to be selected in a preset number according to the predicted values to obtain the identifier set to be selected. Then, the terminal 110 updates the target identifier set of the target information push service according to the identifier set to be selected.

In this embodiment, the terminal 110 may send the identifier of the user to be selected and the corresponding feature of the user to be selected to the server 120, the server 120 obtains a predicted value of the user to be selected through prediction by using a prediction model, and the server 120 performs the steps of selecting the identifiers of the preset number of users to be selected, obtaining the identifier set to be selected, and updating the target identifier set of the target information push service.

In this embodiment, the terminal 110 may obtain a target identifier set of the target information push service at regular time, extract characteristics of a user corresponding to each identifier in the target identifier set, and train according to the characteristics to obtain a prediction model. Then, the terminal 110 obtains the identifier and the feature of the user to be selected, and inputs the feature of the user to be selected into the prediction model for prediction, so as to obtain a prediction value of the user to be selected. The terminal 110 selects the identifiers of the users to be selected in a preset number according to the predicted values to obtain the identifier set to be selected. Then, the terminal 110 updates the target identifier set of the target information push service according to the identifier set to be selected, so as to obtain an updated target identifier set. The method comprises the steps of obtaining a released target identification set at regular time, taking the released target identification set as a sample, automatically training and updating a prediction model so as to automatically update the target identification set, and achieving automatic updating of a target information pushing object without manual operation.

As shown in fig. 2, in one embodiment, an information push object update method is provided. The embodiment is mainly illustrated by applying the method to the terminal 110 (or the server 120) in fig. 1. Referring to fig. 2, the information push object updating method specifically includes the following steps:

step 202, a target identification set of a target information push service is obtained.

The target information pushing service refers to information needing to be pushed to a user. The target information may be, but is not limited to, advertisements, news of various fields, information. The form of the push may be, but is not limited to, text, links, video, audio, and the like. The target identification set is a package which is being released, the package comprises identifications of users, and one user corresponds to one identification. The user's identification may be a nickname of the user, an account identification of the user, or the like.

Specifically, the computer device may determine target information that needs to be pushed to the user, and determine a target identifier set corresponding to the project object information pushing service.

In this embodiment, when the computer device receives a push instruction of target information, a target identifier set corresponding to the target information push service is obtained.

In this embodiment, obtaining a target identifier set of a target information push service includes: and acquiring a target identification set of the target information push service at regular time.

Specifically, the computer device may monitor the current time in real time and compare the current time with a preset time. And when the current time is the same as the preset time, the computer equipment acquires a target identification set corresponding to the push service.

For example, a target identifier set of a target information push service is preset to be acquired at 0 hour, 0 minute and 0 second every day, the computer device monitors the current time in real time, and when the current time reaches 0 hour, 0 minute and 0 second, the computer device acquires the target identifier set of the target information push service.

And 204, extracting the characteristics of the user corresponding to each identifier in the target identifier set, and training according to the characteristics to obtain a prediction model.

Wherein the characteristic of the user may be at least one of attribute information of the user, personal interest of the user, and vertical industry concerned by the user.

Specifically, the computer device determines a user corresponding to each identifier in the target identifier set, and extracts the characteristics of the user corresponding to each identifier. And then, the computer equipment trains a prediction model according to the extracted characteristics corresponding to each user to obtain the trained prediction model.

In this embodiment, the computer device may obtain personal information of a user corresponding to each identifier in the target identifier set, and extract, from the personal information of the user, attribute information, personal interest, vertical industry concerned by the user, and other features of the user. Further, the computer device may extract features of the user from the registration information of the user and the personalized signature of the user.

And step 206, acquiring the identification and the characteristics of the user to be selected, and inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected.

The candidate user refers to a user waiting for inputting the prediction model for prediction, and the candidate user may be a user corresponding to the identifier in the target identifier set or a user corresponding to the identifier in the non-target identifier set. The predicted value refers to a value obtained after the user to be selected is predicted by the prediction model, and the predicted value can be a score or a percentage.

Specifically, the computer device obtains the identification of the user to be selected, and extracts the feature of each user to be selected. And then, inputting the characteristics of each user to be selected into the prediction model by the computer equipment for prediction to obtain a prediction value output by the prediction model and corresponding to each user to be selected.

And 208, selecting the identifiers of the users to be selected in a preset number according to the predicted numerical value to obtain an identifier set to be selected.

The candidate identification set refers to a set of identifications corresponding to each selected candidate user, and the candidate identification set includes identifications of the preset number of candidate users.

Specifically, the computer device may select a preset number of identifiers of the users to be selected according to the size of the predicted value. Further, the computer device may sort the predicted values corresponding to each user to be selected, and select the identifiers of the users to be selected in a preset number from the sorted predicted values to obtain the identifier set to be selected.

In this embodiment, selecting the identifiers of the preset number of users to be selected according to the predicted value to obtain the identifier set to be selected includes: and sequentially selecting the identifications of the users to be selected from high to low according to the predicted numerical value to obtain an identification set to be selected, wherein the identification set to be selected comprises the identifications of the users to be selected with the preset number.

Specifically, the computer device sorts the predicted values corresponding to each user to be selected, and obtains a preset number of users to be selected, where the preset number may be, but is not limited to, 200, 500, or 1000. The preset number can be adjusted according to the requirement of target information needing to be pushed. And then, the computer equipment sequentially selects the users to be selected according to the sequence of the predicted numerical values from high to low to obtain the preset number of the users to be selected. Further, the predicted values can be sorted from high to low or from low to high, and the users to be selected are sequentially selected from the sorted predicted values from high to low, so that the preset number of the users to be selected is obtained. The computer equipment acquires the identifiers of the selected preset number of the to-be-selected users, so that a to-be-selected identifier set is obtained, and the to-be-selected identifier set comprises the preset number of the identifiers.

And step 210, updating the target identification set of the target information push service according to the identification set to be selected.

Specifically, the computer device updates the target identifier set of the target information push service according to the obtained identifier set to be selected, so as to obtain an updated target identifier set of the target information push service.

According to the information push object updating method, the target identification set of the target information push service is obtained, the characteristics of the user corresponding to each identification in the target identification set are extracted, and the prediction model is obtained according to characteristic training. And after the target information is pushed to the users corresponding to the identifiers in the target identifier set, the feedback data are automatically recovered. And training the prediction model according to the characteristics of the extracted feedback data, so that the training model is adjusted and updated according to the feedback data recovered each time. And then, acquiring the identification and the characteristics of the user to be selected, and inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected. The user to be selected is predicted by using the updated prediction model, so that the predicted result is more accurate. And then selecting the identifiers of the users to be selected with the preset number according to the predicted numerical value to obtain an identifier set to be selected, and updating the target identifier set of the target information push service according to the identifier set to be selected. Such that the updated target identification set increases the likelihood of exposure of the target information. According to the scheme, the released target identification set is obtained and used as a sample to automatically train and update the prediction model so as to automatically update the target identification set without manual operation, and therefore automatic updating of a target information pushing object is achieved.

In one embodiment, before a target identifier set of a target information push service is periodically acquired, the information push object updating method further includes: obtaining an initial training sample; extracting the characteristics of the initial training sample, and training according to the characteristics of the initial training sample to obtain a prediction model; acquiring the identification and the characteristics of a user to be selected, and inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected; sequentially selecting the identification of the user to be selected as a target identification from high to low according to the predicted numerical value, and generating a target identification set of the target information push service, wherein the target identification set comprises a preset number of target identifications; and pushing the target information to the identifiers in the target identifier set.

The initial training sample is a training sample obtained for the first time and is used for training and testing the constructed prediction model for the first time.

Specifically, the computer device obtains an initial training sample and extracts features of the initial training sample. And training and testing the constructed prediction model for the first time according to the characteristics of the initial training sample to obtain the trained prediction model. And then, the computer equipment predicts the users to be selected by using the trained prediction model to obtain the prediction value of each user to be selected. And then, the computer equipment sequentially selects the identification of the user to be selected as the target identification from high to low according to the predicted numerical value, and generates a target identification set of the target information pushing service, wherein the target identification set comprises a preset number of target identifications, so that a pushing object of the target information is generated for the first time. And then, the computer equipment pushes the target information to the identifiers in the target identifier set, so that the pushing of the target information is realized.

In this embodiment, the obtaining of the initial training sample includes: an initial training sample is obtained based on an image indexing method.

Where the portrait index refers to a tab created for a user. The portrait indexing method comprises the crowd with basic attributes or interests as tags, the crowd with position information as tags, the crowd with certain application programs as tags, the crowd with advertisements as tags for browsing and the crowd with keywords as tags.

Specifically, the computer device may obtain initial training samples based on an portrait indexing approach. Further, the computer device may obtain a population tagged with the basic attribute or tagged with the interest, a population tagged with the location information, a population tagged with the application program of a certain type, and at least one of a population tagged with an advertisement of a certain type and a population tagged with the keyword as an initial training sample.

For example, the computer device may obtain an initial training sample based on the tags using the online shopping application, and then the computer device obtains the tags as the initial training sample for the user using the online shopping application. The computer device obtains an initial training sample based on the label of liking beautiful makeup, and then the computer device obtains the user of which the label is liking beautiful makeup as the initial training sample. The initial training sample is obtained in a portrait index-based mode, so that the selection range of the initial training sample can be expanded, and various choices are provided, so that the requirements of different target information push services are met.

In this embodiment, the computer device may select the label of the initial training sample according to the type setting of the target information push service. For example, if the target information is a cosmetic advertisement, the computer device may set the selected user tag as a cosmetic favorite or other cosmetic related tag, so as to select an initial training sample related to the target information push service.

In this embodiment, the obtaining of the initial training sample includes: and acquiring an initial training sample based on a similar population expansion mode.

The similar population expansion mode is a technology for searching more similar populations having potential relevance with the seed user through a specific algorithm evaluation model based on the seed user provided by an advertiser, such as collected user samples or equipment identification. For example, if the target information is a recommended advertisement of herbal tea, a purchaser of herbal tea is used as a seed user to search related groups such as working people with high pressure, people staying up at night, guests of hot pot restaurants and the like according to some logic rules such as easy getting on fire and the like.

Specifically, the computer device may obtain a target identifier set, use a user corresponding to each identifier in the target identifier set as a seed user, and search a population related to the seed user based on a similar population expansion manner as an initial training sample. Or the computer device can acquire the pushed target information and search for a pushing object of other information pushing services related to the target information as an initial training sample based on a similar population expansion mode. The initial training sample is obtained in a similar population expansion mode, and a user related to the current user sample or related to the pushed target information can be obtained as the initial training sample, so that the obtained initial training sample is more similar to the user sample, and the requirements of different target information pushing services can be met.

In this embodiment, the obtaining of the initial training sample includes: and acquiring an initial training sample based on at least one of an portrait indexing mode and a similar population expansion mode.

Fig. 3 is a schematic interface diagram illustrating update of an information push object in one embodiment.

Specifically, the release date refers to the time when the target information is pushed to the user. The optimization goal refers to a goal that can be achieved by pushing objects that update the goal information. The optimization target in this embodiment is the click rate, that is, by updating the push object of the target information, more push objects can click the target information, and the click rate of the target information is improved. The seed crowd is a packet corresponding to the history pushing object identifier. "XXA" and "XXB" are names of information push terminals, and "XXC" is an identification of target information to be pushed. The user can check the automatic updating function of the information pushing platform, select the release date, input the target information to be pushed at the information pushing end, and input the identification set corresponding to the target information to be pushed. After selection, the information push platform can automatically acquire a target identification set updating prediction model corresponding to the released target information and automatically update the target identification set.

In one embodiment, the periodically acquiring a target identifier set of a target information push service includes: starting timing after the target identification set is updated; and when the preset time length is reached, acquiring a target identification set of the target information push service.

The preset duration refers to a period of time from the last time the target identifier set is updated to the time the target identifier set is updated again. The preset time can be adjusted according to the requirement, for example, the preset time can be one day, two days, etc.

Specifically, the computer device starts timing after updating the target identification set, and can detect the updating time of the target identification set in real time. The computer equipment obtains the preset time length, compares the detected time with the preset time length, and when the detected time is less than the preset time length, the computer equipment continues timing and continues monitoring in real time. And when the detected time reaches the preset time length, the computer equipment stops timing and acquires a target identification set of the target information push service. By starting timing after updating the target identification set, when the preset duration is reached, the target identification set of the target information push service is obtained. After the target information is pushed to the target identification set for a period of time, the target identification set which is being released can be automatically collected so as to collect feedback data, and therefore the pushed object of the target information can be better adjusted.

In one embodiment, the obtaining of the target identifier set of the target information push service includes: starting timing from pushing the target information to the updated target identifier set; and when the preset time length is reached, acquiring a target identification set of the target information push service.

Specifically, the computer device pushes the target information to the updated target identifier set, and starts timing after the target information is successfully pushed. And when the time for successfully pushing the target information to the updated target identification set reaches the preset time length, the computer equipment acquires the target identification set of the target information pushing service. By starting timing from the pushing of the target information to the updated target identifier set, when the preset duration is reached, the target identifier set of the target information pushing service is obtained. After the target information is pushed to the target identification set for a period of time, the target identification set which is being released can be automatically collected so as to collect feedback data, and therefore the pushed object of the target information can be better adjusted.

In an embodiment, as shown in fig. 4, the inputting the characteristics of the user to be selected into the prediction model for prediction to obtain the predicted value of the user to be selected includes:

step 402, determining the probability that the user to be selected belongs to each classification in the prediction model according to the characteristics of the user to be selected.

Specifically, the computer device inputs the characteristics of the user to be selected into a trained predictive model. And the prediction model receives the characteristics of the user to be selected, and the probability that the user to be selected belongs to each class is calculated according to each constructed class in the prediction model.

Step 404, obtain the weight corresponding to each classification in the prediction model.

And 406, determining a prediction value of the user to be selected according to the probability and the weight.

Specifically, one class in the predictive model corresponds to one weight, and each class corresponds to a different weight. The computer device obtains weights corresponding to the classifications in the predictive model. And then, the computer equipment calculates the prediction value of each user to be selected according to the weight corresponding to each classification and the probability of the user to be selected belonging to each classification.

In this embodiment, the computer device may perform weighted summation on the probability that each user to be selected belongs to each category and the weight of each category to obtain a predicted value of each user to be selected.

In this embodiment, the prediction value may be a score, and the computer device obtains, through calculation, a probability that each user to be selected belongs to each category in the prediction model, and obtains a weight corresponding to each category in the prediction model. And the computer equipment performs weighted summation on the probability of a user to be selected belonging to each classification in the prediction model and the weight corresponding to each classification to obtain the prediction score of the user to be selected. And calculating each user to be selected in the same way to obtain the predicted score of each user to be selected.

According to the information push object updating method, the probability that the user to be selected belongs to each classification in the prediction model is determined according to the characteristics of the user to be selected, the weight corresponding to each classification in the prediction model is obtained, and the prediction value of the user to be selected is determined according to the probability and the weight, so that the prediction value of each user to be selected can be obtained quickly and accurately.

In an embodiment, as shown in fig. 5, the extracting features of the user corresponding to each identifier in the target identifier set, and training according to the features to obtain a prediction model includes:

step 502, classifying users corresponding to each identifier in the target identifier set into a positive sample and a negative sample.

Wherein, the positive sample may be a user who clicks the pushed target information, and the negative sample may be a user who does not click the pushed target information.

Specifically, after the computer device obtains the target identifier set, it is detected whether the user corresponding to each identifier in the target identifier set clicks the pushed target information. The computer device classifies the user according to whether the user clicks the pushed target information. Further, the computer device takes the user who clicked the pushed target information as a positive sample, and takes the user who did not click the pushed target information as a negative sample.

And step 504, extracting features of the positive sample and features of the negative sample, wherein the features of the positive sample and the features of the negative sample comprise at least one of attribute information, personal interests and vertical industries of the user.

The attribute information may include age, gender, academic calendar, occupation, income, and the like. Personal interests refer to general hobbies, such as, for example, enjoying running, enjoying comics, enjoying shopping, and the like. The vertical industry refers to an industry with professionalism which is concerned by users, such as a financial industry, an information technology industry and the like.

Specifically, the computer device extracts at least one feature of attribute information, personal interests, and vertical industry of the positive sample user. And extracting at least one feature of attribute information, personal interests and vertical industries of the negative sample user.

Step 506, training according to the characteristics of the positive sample and the characteristics of the negative sample to obtain a prediction model.

Specifically, the computer device trains a prediction model according to the attribute information of the positive sample and the negative sample, the personal interests and the characteristics in the vertical industry, and continuously adjusts parameters according to the training result of each time to obtain the trained prediction model.

According to the information push object updating method, the users corresponding to the identifiers in the target identifier set are classified into the positive sample and the negative sample, the characteristics of the positive sample and the characteristics of the negative sample are extracted, and the prediction model is obtained through training according to the characteristics of the positive sample and the characteristics of the negative sample. And the model is trained according to the positive and negative samples, so that the discrimination of the trained prediction model is higher, and the prediction is more accurate.

In an embodiment, as shown in fig. 6, after the acquiring the target identity set of the target information push service at the timing, the method further includes:

step 602, detecting the number of positive samples of users corresponding to each identifier in the target identifier set.

Step 604, comparing the number of positive samples with a preset number of samples.

Wherein, the preset number of samples may be a preset number of required positive samples. The number of the preset samples is less than the number of each mark in the target mark set. The preset number of samples can be adjusted according to the requirement, for example, the preset number of samples can be 50, 100, or 200.

Specifically, after the computer device regularly acquires a target identifier set of the target information push service, the users corresponding to each identifier in the target identifier set are classified into a positive sample and a negative sample, and then the number of the users of the positive sample is detected. Then, the computer device obtains the number of the preset samples, and compares the number of the positive samples with the number of the preset samples to determine whether the number of the positive samples reaches the number of the preset samples.

Step 606, when the number of the positive samples is smaller than the number of the preset samples, obtaining the characteristics of the user corresponding to the information push service with the same type as the target information push service.

The same type refers to an information push service belonging to the same type as the target information push service, and may be the same industry, similar information materials, or similar information promotion plans. For example, the pushed target information is shopping information on a certain platform, and the information pushing service of the same type as the target information can be shopping information on other shopping platforms. The pushed information is an advertisement of a certain cosmetic, and the information of the same type is an advertisement of various cosmetics, or an advertisement which has the same brand, the same function or can be matched with the cosmetics.

Specifically, when the computer device detects that the number of positive samples is less than the preset number of samples, it indicates that a sufficient amount of feedback data has not been collected during the period of time to update the training model. The computer device determines the information push service of the same type as the target information push service, and obtains the user corresponding to the information push service of the same type, and takes the user corresponding to the information push service of the same type as a supplementary positive sample. The computer device then extracts the characteristics of the push user of the same type of information push service.

The extracting of the characteristics of the user corresponding to each identifier in the target identifier set and the training according to the characteristics to obtain a prediction model include:

step 608, extracting the characteristics of the user corresponding to each identifier in the target identifier set.

Step 610, training according to the characteristics of the user corresponding to each identifier in the target identifier set and the characteristics of the user corresponding to the information push service with the same type to obtain a prediction model.

Specifically, the computer device determines a user corresponding to each identifier in the target identifier set, and extracts the characteristics of the user corresponding to each identifier. And then, the computer equipment trains a prediction model according to the extracted characteristics corresponding to each user and the characteristics of the users corresponding to the information push services with the same type, so as to obtain the trained prediction model.

For example, the number of the preset samples is 100, the target information is a cosmetic recommendation advertisement of brand a, and after the target information is pushed to the user corresponding to each identifier in the target identifier set, the computer device collects and acquires the target identifier set after a period of time. The number of positive samples obtained after the computer device detects that the user corresponding to each identifier in the target identifier set is classified is 80, and the number of the positive samples is smaller than the number of preset samples, which indicates that the collected feedback data is insufficient. The computer device may detect the advertisements for the other cosmetics of brand a to determine the identification set to which the advertisements for the other cosmetics of brand a correspond. And obtains from the set of identities 20 identities that are not the same as the target set of identities as complementary positive samples. Then, the computer device extracts three types of features of the attribute type, the personal interest and the vertical industry of the user corresponding to the 20 identifiers, and respectively extracts three types of features of the attribute type, the personal interest and the vertical industry of the user of the positive sample and the negative sample in the target identifier set. And inputting the characteristics of the user corresponding to the 20 identifications as the characteristics of the positive sample, the characteristics of the 80 positive samples in the target identification set and the characteristics of the negative samples in the target identification set into the prediction model together to train the prediction model, so as to obtain the trained prediction model.

According to the information push object updating method, whether sufficient feedback data are collected or not is determined by detecting whether the number of positive user samples corresponding to each identification in the acquired target identification set reaches the preset sample number or not. When the number of the positive samples is smaller than the preset number of samples, the collected number of the positive samples is insufficient, and then a push object corresponding to the information push service of the same type as the target information push service can be used as supplementary data of the positive samples, so that training samples with sufficient number can be obtained.

In one embodiment, the predictive model includes a factorization model and an ensemble tree model; the training according to the features to obtain the prediction model comprises the following steps: performing feature conversion and feature splicing processing on the features through a factorization model; and inputting the output of the factorization machine model into the integrated tree model for model training to obtain a prediction model.

The factor decomposition Machine (FM) model has good learning capacity for sparse data, and can reduce dimensions of some high-dimensional discrete feature vectors, so that continuous low-dimensional feature vectors are obtained. An integrated tree model (xgboost model for short) is a model based on a Gradient lifting algorithm, and a new objective function is obtained by performing second-order taylor expansion on an objective function and fitting a residual error according to first-order derivative and second-order derivative information. And constructing a plurality of trees, defining a complexity structure part and a leaf weight part of each tree, and adding the complexity structure part and the leaf weight part into a new objective function as a regular term. And then obtaining the optimal segmentation points through a greedy algorithm to carry out division, and stopping division until a certain threshold value is met or pure nodes are obtained so as to construct an integrated tree model.

In particular, the predictive model includes a factorization model and an ensemble tree model. After extracting the characteristics of the user corresponding to each identifier in the target identifier set, the computer equipment inputs the characteristics of the user corresponding to each identifier in the target identifier set into a factorization model in the prediction model. The factorization machine model receives the characteristics of the user, and performs characteristic transformation on the characteristics of the user to obtain a characteristic vector corresponding to each characteristic. And then, the factorization machine model carries out splicing processing on the feature vector corresponding to the feature of each user to obtain a spliced feature vector corresponding to each user, and the factorization machine model outputs the spliced feature vector corresponding to each user. And then, the factorization model transmits the output spliced feature vectors corresponding to each user to the integrated tree model. And the integrated tree model receives the spliced characteristic vectors corresponding to each user, performs model training according to the spliced characteristic vectors, adjusts parameters of the model for multiple times to perform repeated training, and finally obtains a prediction model meeting the preset accuracy requirement.

According to the information push object updating method, the characteristics are subjected to characteristic conversion and characteristic splicing through the factorization model, the output of the factorization model is input into the integrated tree model to be subjected to model training, and the prediction model is obtained, so that the obtained prediction model is more accurate, and the recognition degree is higher.

In one embodiment, the feature transforming the feature by the factorization model comprises: respectively matching the characteristics of the user corresponding to each identifier in the target identifier set with preset characteristics; and taking the feature vector of the preset features matched with the features of the user corresponding to each identifier as the feature vector corresponding to each identifier.

The preset features refer to preset feature words, and each feature word corresponds to one feature vector.

Specifically, the computer device selects a user corresponding to any one identifier in the target identifier set, and selects any one feature of the user to match with a preset feature. And when the matching is successful, acquiring a feature vector corresponding to the successfully matched preset feature, and taking the feature vector as a feature vector corresponding to the selected feature. Then, the computer device selects the other features of the user for matching according to the same mode, and a feature vector corresponding to each feature of the user can be obtained, so that a plurality of feature vectors corresponding to the user identifier can be obtained. And executing the same matching operation aiming at each feature of each user so as to obtain a feature vector corresponding to the identifier of each user, wherein the number of the feature vectors corresponding to the identifier of one user is the same as the number of the features of the user. By matching the extracted features with the preset features and taking the feature vectors of the preset features as the feature vectors corresponding to the user features successfully matched, the text features can be converted into corresponding numerical vectors, and therefore feature conversion processing is achieved.

In one embodiment, the computer device may classify the features of the user into discrete features and continuous features, and encode the discrete features of the user by a one-hot encoding method to obtain a feature vector corresponding to the discrete features. The discrete feature refers to a feature having at least two classification values. For example, "gender" is a characteristic that has two values, male and female. The continuous characteristic refers to a characteristic with a certain regular change, such as 'height', and the height of a person is gradually increased and belongs to a continuous characteristic. One-Hot coding, i.e., One-Hot coding, also called One-bit significance coding, is directed to a discrete feature, which uses an N-bit state register to code N states, each state having its own independent register bit, and at any time, only One of the states is in the significance state 1, and the others are 0. Such as unique thermal code 0001,0100,0010,0001. For example, the feature "gender" includes two attribute values: male and female, male denoted 1 and female denoted 2, after unique hot coding, this feature is divided into two parts: male ═ 0,1], female ═ 1, 0. After the one-hot encoding, a feature vector of discrete features can be obtained, and the obtained feature vector is a low-dimensional continuous feature vector.

In one embodiment, as shown in fig. 7, the feature stitching process performed on the feature by the factorization model includes:

step 702, for any identifier in the target identifier set, a first-dimension feature vector in the feature vector corresponding to the identifier is obtained.

The first-dimension feature vector refers to a high-dimension discrete feature vector corresponding to some preset features. For example, the vertical industry concerned by the user in the preset features is the financial industry, and the feature vector corresponding to the financial industry is the first-dimension feature vector.

In particular, one identifier may correspond to three features of one user and one feature corresponds to one feature vector, and one identifier may correspond to three feature vectors. And the computer equipment acquires the characteristic vector corresponding to each identifier in the target identifier set. For each identifier, the computer device classifies the feature vector corresponding to each identifier, and classifies the first-dimension feature vector and the non-first-dimension feature vector. The computer equipment extracts the feature vector as the preset feature of the first-dimension feature vector, and determines the feature of the user corresponding to each identifier in the target identifier set matched with the extracted preset feature, namely, the feature vector corresponding to the feature of the user matched with the extracted preset feature can be determined as the first-dimension feature vector. Then, the computer device obtains a first-dimension feature vector in the feature vectors corresponding to the identifiers in the target identifier set.

Step 704, performing dimension reduction processing on the first-dimension feature vector to obtain a second-dimension feature vector, where the dimension of the second-dimension feature vector is smaller than the dimension of the first-dimension feature vector.

The second-dimension feature vector is a low-dimension continuous feature vector obtained by performing dimension reduction processing on the first-dimension feature vector, and therefore the dimension of the second-dimension feature vector is smaller than that of the first-dimension feature vector.

Specifically, the computer device performs dimension reduction processing on the acquired first-dimension feature vector through the factorization machine model. The factorization machine model maps the first-dimension feature vectors into a low-dimension continuous space to convert the high-dimension discrete feature vectors into low-dimension continuous feature vectors, so that second-dimension feature vectors corresponding to each first-dimension feature vector are obtained.

Step 706, the second-dimension feature vector is spliced with the rest feature vectors except the first-dimension feature vector in the feature vector corresponding to the identifier.

Specifically, the computer device acquires a feature vector corresponding to one identifier in the target identifier set to acquire a second-dimension feature vector corresponding to a first-dimension feature vector in the feature vector. And then, the computer equipment splices the second-dimension characteristic vector corresponding to the identifier and the rest characteristic vectors except the first-dimension characteristic vector through a factorization model to obtain a spliced characteristic vector. After the splicing processing, one identifier corresponds to one spliced feature vector. And then, according to the same processing mode, splicing the second-dimension feature vector corresponding to each identifier with the rest feature vectors except the first-dimension feature vector to obtain a spliced feature vector corresponding to each identifier. And outputting the feature vector which is subjected to splicing processing and corresponds to each identifier by the factorization machine model.

According to the information push object updating method, a first-dimension feature vector in a feature vector corresponding to the identifier is obtained by aiming at any identifier in the target identifier set, and dimension reduction processing is carried out on the first-dimension feature vector to obtain a second-dimension feature vector. The high-dimensional discrete feature vector can be converted into the low-dimensional continuous feature vector, the key information in the first-dimensional feature vector is reserved, and the non-key information is removed, so that the error caused by redundant information in the high-dimensional discrete feature vector is reduced. And splicing the second-dimension feature vector with the rest feature vectors except the first-dimension feature vector in the feature vectors corresponding to the identifiers, and integrating a plurality of feature vectors into one feature vector, so that a plurality of pieces of key information are integrated into one vector, and the accuracy of model identification is improved.

In one embodiment, the factorization machine model performs stitching processing on the second-dimensional feature vector and the rest feature vectors except the first-dimensional feature vector in the feature vector corresponding to the identifier, and the stitching processing includes: and the factorization machine model splices the second-dimensional characteristic vector and the rest characteristic vectors except the first-dimensional characteristic vector in the characteristic vectors corresponding to the identifier according to the row vector direction.

Specifically, the computer device acquires a feature vector corresponding to one identifier in the target identifier set to acquire a second-dimension feature vector corresponding to a first-dimension feature vector in the feature vector. And then, the computer equipment splices the second-dimension characteristic vector corresponding to the identifier and the rest characteristic vectors except the first-dimension characteristic vector according to the row vector direction through a factorization model to obtain a spliced characteristic vector. And then, according to the same processing mode, splicing the second-dimension characteristic vector corresponding to each identifier with the rest characteristic vectors except the first-dimension characteristic vector according to the row vector direction to obtain a spliced characteristic vector corresponding to each identifier. And outputting the feature vector which is subjected to splicing processing according to the row vector direction and corresponds to each identifier by the factorization machine model. By splicing the feature vectors according to the direction of the row vector, the feature vectors can be integrated into one feature vector, so that the key information is integrated into one piece of key information, and the processing efficiency is improved.

For example, one identifier corresponds to the second-dimension feature vector as [345], and the other identifier corresponds to the feature vectors other than the first-dimension feature vector as [12], [768 ]. The factorization model splices the three eigenvectors according to the row vector direction to obtain a long vector [34512768 ].

FIG. 8 is a diagram illustrating an embodiment of automatically updating a trained predictive model. As shown in fig. 8, the feature vectors of the sparse feature 1 and the sparse feature 2 are high-dimensional feature vectors, and require dimension reduction processing. The feature vector of the dense feature 3 is a low-dimensional feature vector, and dimension reduction is not required. And the factorization machine model in the prediction model obtains a feature vector corresponding to each feature by the extracted features of the user in a mode of one-hot coding. Since some features are sparse, the resulting feature vector is a high-dimensional discrete feature vector. High-dimensional discrete feature vectors are complex, the calculation process is more cumbersome, and errors are easily increased. In order to reduce the calculation error, the factorization model maps the discrete feature vectors with high dimension in the feature vectors into the continuous space with low dimension to realize dimension reduction, so as to obtain the continuous feature vectors with low dimension. w is a₀、w_i、w_j、v_iAnd v_jAnd the feature vectors are subjected to dimension reduction. Then, the factorization model reduces the dimension of a user to obtain low-dimensional continuous feature vectors and non-continuous feature vectorsAnd performing splicing processing on other eigenvectors needing dimension reduction according to the direction of the row vector. And outputting the spliced feature vectors to the integrated tree model by the factorization model. The integrated tree model comprises at least two trees, the integrated tree model combines the received feature vectors, each tree adopts different feature vector combination modes, and a complexity structure part and a leaf weight part of each tree are defined. And obtaining the optimal segmentation point through a specific algorithm to perform division until a certain threshold value is met or a pure node is obtained. And obtaining the final parameters of the integrated tree model through multiple times of training and parameter adjustment, thereby realizing the automatic updating training of the prediction model.

In an embodiment, the updating the target identifier set of the target information push service according to the candidate identifier set includes: determining different identifiers in the identifier set to be selected and the target identifier set; and replacing different identifiers in the target identifier set with different identifiers in the to-be-selected identifier set.

Specifically, the computer device compares each identifier in the candidate identifier set with each identifier in the target identifier set one by one, and determines different identifiers in the two identifier sets. Then, the computer device obtains the identifiers in the candidate identifier set which are different from the identifiers in the target identifier set, and replaces the different identifiers in the target identifier set with the different identifiers in the candidate identifier set, so as to obtain a new target identifier set. The updating of the pushing object of the target information pushing service is realized by determining the different identifications in the two identification sets and replacing the different identifications in the target identification set with the different identifications in the to-be-selected identification set. And only different identifications need to be replaced, so that incremental updating of data is realized, and the data processing efficiency is improved.

In an embodiment, as shown in fig. 9, the updating the target identifier set of the target information push service according to the candidate identifier set includes:

step 902, determining different identifiers in a candidate identifier set and a target identifier set, where the different identifiers in the target identifier set are first identifiers, and the different identifiers in the candidate identifier set are second identifiers.

Specifically, the computer device compares each identifier in the candidate identifier set with each identifier in the target identifier set one by one, and determines different identifiers in the two identifier sets. And taking each identifier which is different from the identifiers in the to-be-selected identifier set and exists in the target identifier set as a first identifier, and taking each identifier which is different from the identifiers in the target identifier set and exists in the to-be-selected identifier set as a second identifier.

At step 904, the number of first identifiers is determined.

Step 906, selecting the second identifications with the same number as the first identifications.

Step 908 replaces the first identifier with an equal number of second identifiers.

Specifically, the computer device determines the number of first identifiers and selects second identifiers having the same number as the first identifiers. Then, the computer device replaces the first identifiers in the target identifier set with the same number of second identifiers to update the target identifier set.

The information push object updating method updates the push object of the target information push service by determining the different identifications in the two identification sets, determining the number of the different identifications in the target identification set, and selecting the same number of identifications from the different identifications in the to-be-selected identification set to replace the different identifications in the target identification set. And only different identifiers are replaced, and not all the identifiers are required to be replaced, so that incremental updating of data is realized, and the number of push objects of the target information push service before and after updating is kept unchanged.

In one embodiment, the selecting the second identifiers with the same number as the first identifiers comprises: and selecting the second marks to replace the first marks according to the predicted values from high to low, wherein the number of the selected second marks is the same as that of the first marks.

Specifically, the computer device uses, as the first identifier, each identifier that is different from the identifier in the candidate identifier set and exists in the target identifier set, and uses, as the second identifier, each identifier that is different from the identifier in the target identifier set and exists in the candidate identifier set. And then, the computer equipment acquires the predicted numerical values of the users corresponding to the second identifications, and sorts the predicted numerical values corresponding to the second identifications. Further, the computer device may sort the predicted values corresponding to the respective second identifiers from high to low or from low to high. Then, the computer equipment determines the number of the first marks, and sequentially selects second marks from high to low according to the predicted numerical value, wherein the number of the selected second marks is the same as that of the first marks. Then, the computer device replaces the first identifier with the selected second identifier to achieve the update of the target identifier set. The second identifications in the identification set to be selected are sequenced according to the corresponding prediction numerical values, and the second identifications are sequentially selected from high to low to replace the first identifications, so that the possibility of exposing target information is increased. And the number of the second identifications is the same as that of the first identifications, and the number of the push objects of the target information push service before and after updating is kept unchanged.

In one embodiment, the information push object updating method further includes: acquiring the states of all the identifiers in the updated target identifier set, wherein the states comprise a login state and an unregistered state; and pushing the target information to the updated target identifier which is in the login state in the target identifier set.

The login state refers to that the user terminal corresponding to the identifier is in an online state. The non-login state refers to that the user terminal corresponding to the identifier is in an offline state.

Specifically, the computer device obtains the updated target identifier set and detects the status of each identifier in the updated target identifier set. And the computer equipment determines the identifier in the login state and the identifier in the non-login state in the updated target identifier set through detection. And then, the computer equipment pushes the target information to the user terminal corresponding to the identifier in the updated target identifier set in the login state. For an identifier in a non-logged-in state, the computer device does not push target information. And determining whether the user terminal corresponding to each identifier is in an online state or not by detecting the state of each identifier in the updated target identifier set, and pushing target information to the user terminal in the online state to increase the exposure of the target information. The mark in the non-login state does not push the target information, so that the waste exposure of the target information can be avoided.

In one embodiment, the information push object updating method further includes: and detecting the latest login time of each identifier in the updated target identifier set, and pushing target information when the latest login time of the identifier is within preset time.

The preset time may be a period of time obtained by calculating the time forward by a preset number of days from the time of updating the target identifier set, where the preset number of days may be three days, five days, seven days, or the like. For example, the target identification set is updated in the time of 2019-05-25, the preset number of days is 3 days, and the preset time is from 2019-05-22 to 2019-05-25.

Specifically, the computer device obtains an updated target identifier set, and detects the time when the target identifier set is updated. And then, the computer equipment detects the latest login time of each identifier in the updated target identifier set, and compares the latest login time of each identifier with preset time to determine whether the latest login time of each identifier is within the preset time. And when the latest login time of the identifier is within the preset time, pushing the target information to the user terminal corresponding to the identifier. And when the time of the last login of the identification is not within the preset time, not pushing the target information. And judging the possibility of login of the user terminal corresponding to the identifier by detecting whether the latest login time of each identifier in the updated target identifier set is within the preset time. When the latest login time of the identifier is within the preset time, the probability that the user corresponding to the identifier browses the target information is high, and the target information can be pushed to increase the exposure of the target information, so that the target information can be spread. When the last login time of the identifier is not within the preset time, the user corresponding to the identifier may not log in a short time, and the target information does not need to be pushed, so that invalid propagation of the target information is avoided.

FIG. 10 is a diagram illustrating automatic update of a push object, in one embodiment. Firstly, the computer device obtains initial training samples according to the population with the basic attributes or interests of the labels, the population using the application program, the population obtained by positioning information, the population of historical pushed advertisement information, the population corresponding to the keywords and the population obtained based on the similar population expansion algorithm, and trains and tests the constructed prediction model through the initial training samples. And predicting the users to be selected through the trained prediction model to obtain the prediction value of each user to be selected. And then, the computer equipment sequentially selects the identifications of the users to be selected from the high to the low according to the predicted numerical value to generate a target identification set of the target information pushing service, wherein the target identification set comprises the identifications of the users to be selected with the preset number, so that a pushing object of the target information is generated for the first time. The computer device then pushes the target information to the identifiers in the target identifier set. And then, after a period of time of delivery, the computer equipment regularly acquires a target identification set of the target information pushing service, and takes the delivered target identification set as feedback data. And taking the users with the target information clicked in the target identification set as positive samples, and taking the users without the target information clicked as negative samples. And then, the computer equipment selects the characteristics of the positive sample and the negative sample, performs characteristic transformation and characteristic splicing on the characteristics of the positive sample and the negative sample, and trains the prediction model according to the processed characteristics to update the prediction model. And then, the computer equipment predicts the user to be selected by using the updated prediction model to obtain a prediction value of the user to be selected. And sequentially selecting the identifications of the first N users to be selected from high to low according to the predicted numerical value to form an identification set to be selected, and updating the target identification set of the target information push service according to the identification set to be selected. And the target identification set is collected at regular time to be used as feedback data to automatically update the prediction model, so that the automatic update of the target information push object is realized.

In one embodiment, an information push object updating method is provided, and the method includes:

the computer device obtains an initial training sample based on at least one of an portrait indexing mode and a similar population expansion mode.

And then, the computer equipment extracts the characteristics of the initial training sample and trains according to the characteristics of the initial training sample to obtain a prediction model.

And then, the computer equipment acquires the identification and the characteristics of the user to be selected, and inputs the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected.

And then, the computer equipment sequentially selects the identification of the user to be selected from the high to the low according to the predicted numerical value, and generates a target identification set of the target information pushing service, wherein the target identification set comprises the identification of the user to be selected with the preset number.

Further, the computer device pushes the target information to the identifiers in the target identifier set.

The computer device then begins timing from pushing the target information to the identifiers in the target identifier set.

And then, when the preset time length is reached, the computer equipment acquires a target identification set of the target information push service.

And then, the computer equipment extracts the characteristics of the user corresponding to each identifier in the target identifier set, and respectively matches the characteristics of the user corresponding to each identifier in the target identifier set with preset characteristics through a factorization model.

Further, the computer device takes the feature vector of the preset features matched with the features of the user corresponding to each identifier as the feature vector corresponding to each identifier.

Then, aiming at any identifier in the target identifier set, the computer equipment obtains a first-dimension feature vector in the feature vectors corresponding to the identifier through a factorization machine model.

And then, the computer equipment performs dimensionality reduction processing on the first-dimensional feature vector through a factorization model to obtain a second-dimensional feature vector, wherein the dimensionality of the second-dimensional feature vector is smaller than that of the first-dimensional feature vector.

Further, the computer device carries out splicing processing on the second-dimension characteristic vector and the other characteristic vectors except the first-dimension characteristic vector in the characteristic vector corresponding to the identifier through a factorization machine model.

And then, inputting the output of the factorization model into the integrated tree model by the computer equipment for model training to obtain a prediction model.

And then, the computer equipment acquires the identification and the characteristics of the user to be selected, and determines the probability that the user to be selected belongs to each classification in the prediction model according to the characteristics of the user to be selected.

Next, the computer device obtains a weight corresponding to each classification in the predictive model.

Further, the computer device determines a prediction value of the user to be selected according to the probability and the weight.

And then, the computer equipment selects the identifiers of the users to be selected with preset quantity according to the predicted numerical value to obtain the identifier set to be selected.

Then, the computer device determines the different identifiers in the candidate identifier set and the target identifier set, the different identifiers in the target identifier set are first identifiers, and the different identifiers in the candidate identifier set are second identifiers.

Then, the computer device determines the number of the first identifications; and selecting the second identifications with the same number as the first identifications.

Further, the computer device selects second marks to replace the first marks according to the predicted values from high to low in sequence, and the number of the selected second marks is the same as that of the first marks.

Next, the computer device obtains the status of each identifier in the updated target identifier set, where the status includes a logged-in status and a logged-out status.

And then, the computer equipment pushes the target information to the mark in the login state in the updated target mark set.

According to the information push object updating method, the initial training sample is obtained in a mode based on portrait index and based on similar crowd expansion, the selection range of the initial training sample can be expanded, multiple choices are provided, and the requirements of different target information push services are met. And training according to the training samples to obtain a prediction model, predicting the user to be selected according to the prediction model, generating a target identification set, pushing the target information to the target identification set, and pushing the target information.

After the target information is pushed to the target identification set for a period of time, the target identification set which is being released can be collected at regular time so as to collect feedback data, and therefore the pushed object of the target information can be better adjusted.

By matching the extracted features with the preset features and taking the feature vectors of the preset features as the feature vectors corresponding to the user features successfully matched, the text features can be converted into corresponding numerical vectors, and therefore feature conversion processing is achieved.

And acquiring a first-dimension feature vector in the feature vector corresponding to the identifier by aiming at any identifier in the target identifier set, and performing dimension reduction processing on the first-dimension feature vector to obtain a second-dimension feature vector. The high-dimensional discrete feature vector can be converted into the low-dimensional continuous feature vector, the key information in the first-dimensional feature vector is reserved, and the non-key information is removed, so that the error caused by redundant information in the high-dimensional discrete feature vector is reduced. And splicing the second-dimension feature vector with the rest feature vectors except the first-dimension feature vector in the feature vectors corresponding to the identifiers, and integrating a plurality of feature vectors into one feature vector, so that a plurality of pieces of key information are integrated into one vector, and the accuracy of model identification is improved.

The probability that the user to be selected belongs to each classification in the prediction model is determined according to the characteristics of the user to be selected, the weight corresponding to each classification in the prediction model is obtained, and the prediction value of the user to be selected is determined according to the probability and the weight, so that the prediction value of each user to be selected can be obtained quickly and accurately.

The pushing object of the target information pushing service is updated by determining different identifications in the two identification sets, determining the number of the different identifications in the target identification set, and selecting the identifications with the same number from the different identifications in the to-be-selected identification set to replace the different identifications in the target identification set.

The second identifications in the identification set to be selected are sequenced according to the corresponding prediction numerical values, and the second identifications are sequentially selected from high to low to replace the first identifications, so that the possibility of exposing target information is increased. And only different identifiers are replaced, and not all the identifiers are required to be replaced, so that incremental updating of data is realized, and the number of push objects of the target information push service before and after updating is kept unchanged.

And determining whether the user terminal corresponding to each identifier is in an online state or not by detecting the state of each identifier in the updated target identifier set, and pushing target information to the user terminal in the online state to increase the exposure of the target information. The mark in the non-login state does not push the target information, so that the waste exposure of the target information can be avoided.

According to the scheme, the released target identification set is obtained at regular time, and the released target identification set is used as a sample to automatically train and update the prediction model so as to automatically update the target identification set without manual operation, so that the target information pushing object is automatically updated.

Fig. 11 is a schematic diagram illustrating a comparison between the present solution and a conventional information pushing and object updating method in an embodiment. Wherein, the left side is the traditional advertisement putting cycle and the flow chart of the optimization cycle, including: release-collect feedback-insight analysis-modify orientation-push-release again. The right is a flow chart of the pushing object based on automatic updating in the scheme, which comprises the following steps: release-collect feedback-generate new identity set-replace-release. The conventional scheme is that an advertiser can bind an advertisement to an identifier set at a delivery end for delivery after extracting the identifier set of a pushed object, and usually, the identifier set is static all the time in the whole delivery process, that is, a user group contained in the identifier set is unchanged. If the advertiser wants to optimize the process, the advertiser usually collects feedback data, namely the exposure click condition of the user, according to the effect of online delivery of the identifier set, then analyzes and insights the user with forward behavior to obtain the targeting condition which is more in line with the target population, so as to correct the original population targeting condition, extracts a new identifier set by using the new targeting condition, and then continuously pushes the new identifier set to a delivery end to bind the advertisement for delivery. The entire optimization process takes a long time and each step inside requires the participation of an advertiser or operator.

The method is characterized in that an advertiser uses the existing crowd targeting mode, and can perform label crowd extraction, application program crowd extraction, positioning information crowd extraction, advertisement crowd extraction, keyword crowd extraction and similar crowd extension algorithm extraction based on portrait index to generate a target identification set, the target identification set is pushed to a delivery end to be delivered after being bound with advertisements, an information pushing object updating system can continuously collect delivery effect feedback data of the target identification set, clicking users after exposure of the advertisements are used as positive samples, non-clicking users after exposure are used as negative samples, then the characteristics of the positive and negative sample users are respectively extracted, characteristic conversion and characteristic splicing are performed, and then the characteristics are input into a prediction model to train and update the prediction model. And then obtaining the users to be selected, scoring each user to be selected by using the trained prediction model, and selecting a specified number of users from high to low by using the scoring values to obtain an identifier set to be selected. The number of the identifiers in the candidate identifier set is the same as the number of the identifiers in the target identifier set generated for the first time. And pushing the identifier set to be selected to a releasing end to replace the original target identifier set to be released, namely generating a new identifier set with the same size to replace the original identifier set to be released. The entire automatic update process is then performed again after a period of delivery. Therefore, compared with the traditional scheme, the scheme can automatically collect feedback data and automatically train and update the prediction model according to the feedback data, so that the automatic update of the information push object is realized. In the whole releasing and updating process of the scheme, except that the initial population is an advertiser or an operator to extract and release, the subsequent processes are all automatically completed by the information pushing object updating system, manual participation is not needed, the updating period is short, and the quick releasing and updating method can be quickly effective.

In one embodiment, the scheme can also automatically update the push object of the information based on the portrait index. Because the crowd identification set extracted based on the image index mode depends on the construction of the user portrait, and the portrait data can be updated along with the update of the data source, the new identification set can be periodically extracted according to the same crowd oriented extraction condition to replace the existing cast identification set. For example, the target information is related advertisements recommending the XX game, and the set of identifiers being delivered is a user crowd labeled "XX game fans". During the time that the relevant advertisement for the XX game is being pushed to the user, some users may no longer be concerned with the XX game, some users may begin to be concerned with the XX game, and the user population labeled "XX game fan" changes. The computer equipment can regularly acquire the user crowd labeled as 'attention XX' to obtain a new identification set, and replace the identification set being released by the new identification set, so that the information can be automatically updated according to the portrait index.

Fig. 2 to fig. 9 are schematic flow charts of an information push object updating method in an embodiment. It should be understood that although the various steps in the flowcharts of fig. 2-9 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-9 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.

In one embodiment, as shown in fig. 12, there is provided an information push updating apparatus, including: an acquisition module 1202, a training module 1204, a prediction module 1206, a selection module 1208, and an update module 1210. Wherein,

an obtaining module 1202, configured to obtain a target identifier set of a target information push service.

And the training module 1204 is configured to extract features of the user corresponding to each identifier in the target identifier set, and train according to the features to obtain a prediction model.

The prediction module 1206 is configured to obtain the identifier and the feature of the user to be selected, and input the feature of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected.

The selecting module 1208 is configured to select, according to the predicted value, the identifiers of the preset number of users to be selected, so as to obtain a set of identifiers to be selected.

An updating module 1210, configured to update a target identifier set of the target information push service according to the identifier set to be selected.

The information push object updating device extracts the characteristics of the user corresponding to each identifier in the target identifier set by acquiring the target identifier set of the target information push service, and obtains the prediction model according to characteristic training. And after the target information is pushed to the users corresponding to the identifiers in the target identifier set, the feedback data are automatically recovered. And training the prediction model according to the characteristics of the extracted feedback data, so that the training model is adjusted and updated according to the feedback data recovered each time. And then, acquiring the identification and the characteristics of the user to be selected, and inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a prediction value of the user to be selected. The user to be selected is predicted by using the updated prediction model, so that the predicted result is more accurate. And then selecting the identifiers of the users to be selected with the preset number according to the predicted numerical value to obtain an identifier set to be selected, and updating the target identifier set of the target information push service according to the identifier set to be selected. Such that the updated target identification set increases the likelihood of exposure of the target information. According to the scheme, the released target identification set is obtained and used as a sample to automatically train and update the prediction model so as to automatically update the target identification set without manual operation, and therefore automatic updating of a target information pushing object is achieved.

In one embodiment, the obtaining module 1202 is further configured to: starting timing after the target identification set is updated; and when the preset time length is reached, acquiring a target identification set of the target information push service. By starting timing after updating the target identification set, when the preset duration is reached, the target identification set of the target information push service is obtained. After the target information is pushed to the target identification set for a period of time, the target identification set which is being released can be automatically collected so as to collect feedback data, and therefore the pushed object of the target information can be better adjusted.

In one embodiment, the prediction module 1206 is further to: determining the probability that the user to be selected belongs to each classification in the prediction model according to the characteristics of the user to be selected; acquiring the weight corresponding to each classification in the prediction model; and determining the prediction value of the user to be selected according to the probability and the weight. The information pushing object updating device determines the probability that the user to be selected belongs to each classification in the prediction model according to the characteristics of the user to be selected, obtains the weight corresponding to each classification in the prediction model, and determines the prediction value of the user to be selected according to the probability and the weight, so that the prediction value of each user to be selected can be obtained quickly and accurately.

In one embodiment, training module 1204 is further configured to: classifying users corresponding to the identifiers in the target identifier set into a positive sample and a negative sample; extracting features of a positive sample and features of a negative sample, the features of the positive sample and the features of the negative sample including at least one of attribute information of a user, personal interests and vertical industries; and training according to the characteristics of the positive sample and the characteristics of the negative sample to obtain a prediction model. The user corresponding to each identification in the target identification set is classified into a positive sample and a negative sample, the characteristics of the positive sample and the characteristics of the negative sample are extracted, and a prediction model is obtained through training according to the characteristics of the positive sample and the characteristics of the negative sample. And the model is trained according to the positive and negative samples, so that the discrimination of the trained prediction model is higher, and the prediction is more accurate.

In one embodiment, the obtaining module 1202 is further configured to: detecting the number of positive samples of users corresponding to each identifier in the target identifier set; comparing the number of positive samples with a preset number of samples; when the number of the positive samples is smaller than the number of the preset samples, acquiring the characteristics of a user corresponding to the information push service with the same type as the target information push service;

the training module 1204 is further configured to: extracting the characteristics of the user corresponding to each identifier in the target identifier set; and training according to the characteristics of the user corresponding to each identifier in the target identifier set and the characteristics of the user corresponding to the information push service with the same type to obtain a prediction model.

According to the information pushing object updating device, whether sufficient feedback data are collected or not is determined by detecting whether the number of positive samples of the user corresponding to each identification in the acquired target identification set reaches the preset sample number or not. When the number of the positive samples is smaller than the preset number of samples, the collected number of the positive samples is insufficient, and then a push object corresponding to the information push service of the same type as the target information push service can be used as supplementary data of the positive samples, so that training samples with sufficient number can be obtained.

In one embodiment, the predictive model includes a factorization model and an ensemble tree model; the training module 1204 is further configured to: performing feature conversion and feature splicing processing on the features through a factorization model; and inputting the output of the factorization machine model into the integrated tree model for model training to obtain a prediction model. And performing feature conversion and feature splicing processing on the features through the factorization model, inputting the output of the factorization model into the integrated tree model for model training to obtain a prediction model, so that the obtained prediction model is more accurate and has higher recognition degree.

In one embodiment, the training module 1204 is further configured to: respectively matching the characteristics of the user corresponding to each identifier in the target identifier set with preset characteristics; and taking the feature vector of the preset features matched with the features of the user corresponding to each identifier as the feature vector corresponding to each identifier. By matching the extracted features with the preset features and taking the feature vectors of the preset features as the feature vectors corresponding to the user features successfully matched, the text features can be converted into corresponding numerical vectors, and therefore feature conversion processing is achieved.

In one embodiment, the training module 1204 is further configured to: aiming at any identifier in the target identifier set, acquiring a first-dimension feature vector in the feature vector corresponding to the identifier; performing dimensionality reduction on the first-dimensional feature vector to obtain a second-dimensional feature vector, wherein the dimensionality of the second-dimensional feature vector is smaller than that of the first-dimensional feature vector; and splicing the second-dimension feature vector with the rest feature vectors except the first-dimension feature vector in the feature vector corresponding to the identifier. And acquiring a first-dimension feature vector in the feature vector corresponding to the identifier by aiming at any identifier in the target identifier set, and performing dimension reduction processing on the first-dimension feature vector to obtain a second-dimension feature vector. The high-dimensional discrete feature vector can be converted into the low-dimensional continuous feature vector, the key information in the first-dimensional feature vector is reserved, and the non-key information is removed, so that the error caused by redundant information in the high-dimensional discrete feature vector is reduced. And splicing the second-dimension feature vector with the rest feature vectors except the first-dimension feature vector in the feature vectors corresponding to the identifiers, and integrating a plurality of feature vectors into one feature vector, so that a plurality of pieces of key information are integrated into one vector, and the accuracy of model identification is improved.

In one embodiment, the update module 1210 is further configured to: determining different identifiers in the identifier set to be selected and the target identifier set; and replacing different identifiers in the target identifier set with different identifiers in the to-be-selected identifier set. The updating of the pushing object of the target information pushing service is realized by determining the different identifications in the two identification sets and replacing the different identifications in the target identification set with the different identifications in the to-be-selected identification set. And only different identifications need to be replaced, so that incremental updating of data is realized, and the data processing efficiency is improved.

In one embodiment, the update module 1210 is further configured to: determining different identifiers in a candidate identifier set and a target identifier set, wherein the different identifiers in the target identifier set are first identifiers, and the different identifiers in the candidate identifier set are second identifiers; determining the number of first identifications; selecting second identifications with the same number as the first identifications; the first identification is replaced by the same number of second identifications. The pushing object of the target information pushing service is updated by determining different identifications in the two identification sets, determining the number of the different identifications in the target identification set, and selecting the identifications with the same number from the different identifications in the to-be-selected identification set to replace the different identifications in the target identification set. And only different identifiers are replaced, and not all the identifiers are required to be replaced, so that incremental updating of data is realized, and the number of push objects of the target information push service before and after updating is kept unchanged.

In one embodiment, the update module 1210 is further configured to: and selecting the second marks to replace the first marks according to the predicted values from high to low, wherein the number of the selected second marks is the same as that of the first marks. The second identifications in the identification set to be selected are sequenced according to the corresponding prediction numerical values, and the second identifications are sequentially selected from high to low to replace the first identifications, so that the possibility of exposing target information is increased. And the number of the second identifications is the same as that of the first identifications, and the number of the push objects of the target information push service before and after updating is kept unchanged.

In one embodiment, the apparatus further comprises: and a pushing module. The push module is used for: acquiring the states of all the identifiers in the updated target identifier set, wherein the states comprise a login state and an unregistered state; and pushing the target information to the updated target identifier which is in the login state in the target identifier set. And determining whether the user terminal corresponding to each identifier is in an online state or not by detecting the state of each identifier in the updated target identifier set, and pushing target information to the user terminal in the online state to increase the exposure of the target information. The mark in the non-login state does not push the target information, so that the waste exposure of the target information can be avoided.

FIG. 13 is a diagram illustrating an internal structure of a computer device in one embodiment. The computer device may specifically be the terminal 110 (or the server 120) in fig. 1. As shown in fig. 13, the computer apparatus includes a processor, a memory, a network interface, an input device, and a display screen connected through a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program, which when executed by the processor, causes the processor to implement the information push object updating method. The internal memory may also store a computer program, and the computer program, when executed by the processor, may cause the processor to perform the information push object update method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.

Those skilled in the art will appreciate that the architecture shown in fig. 13 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, the information push object updating apparatus provided in the present application may be implemented in a form of a computer program, and the computer program may be run on a computer device as shown in fig. 13. The memory of the computer device may store various program modules constituting the information push object updating apparatus, such as the obtaining module 1202, the training module 1204, the predicting module 1206, the selecting module 1208 and the updating module 1210 shown in fig. 12. The computer program constituted by the respective program modules causes the processor to execute the steps in the information push object updating method according to the respective embodiments of the present application described in the present specification.

For example, the computer device shown in fig. 13 may perform the step of periodically acquiring the target identification set of the target information push service through the acquisition module 1202 in the information push object update apparatus shown in fig. 12. The computer device may perform the steps of extracting features of the user corresponding to each identifier in the target identifier set through the training module 1204, and obtaining a prediction model according to the feature training. The computer device can execute the steps of obtaining the identification and the characteristics of the user to be selected through the prediction module 1206, inputting the characteristics of the user to be selected into the prediction model for prediction, and obtaining the prediction value of the user to be selected. The computer device may execute, by the selection module 1208, a step of sequentially selecting, from high to low according to the predicted numerical value, identifiers of the users to be selected as target identifiers to obtain an identifier set to be selected, where the identifier set to be selected includes a preset number of target identifiers. The computer device may perform the step of updating the target identifier set of the target information push service according to the candidate identifier set through the updating module 1210.

In one embodiment, a computer device is provided, which includes a memory and a processor, the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the steps of the information push object updating method. Here, the steps of the information push object updating method may be the steps in the information push object updating method of each of the above embodiments.

In one embodiment, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the processor executes the steps of the information push object updating method. Here, the steps of the information push object updating method may be the steps in the information push object updating method of each of the above embodiments.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. An information push object updating method comprises the following steps:

acquiring a target identification set of a target information push service;

2. The method of claim 1, wherein the obtaining the target identity set of the target information push service comprises:

starting timing after the target identification set is updated;

and when the preset time length is reached, acquiring a target identification set of the target information push service.

3. The method of claim 1, wherein the inputting the characteristics of the user to be selected into the prediction model for prediction to obtain a predicted numerical value of the user to be selected comprises:

determining the probability that the user to be selected belongs to each classification in the prediction model according to the characteristics of the user to be selected;

acquiring weights corresponding to all classes in the prediction model;

and determining the prediction value of the user to be selected according to the probability and the weight.

4. The method according to claim 1, wherein the extracting features of the user corresponding to each identifier in the target identifier set and training according to the features to obtain a prediction model comprises:

classifying users corresponding to the identifiers in the target identifier set into a positive sample and a negative sample;

extracting features of the positive exemplar and features of the negative exemplar, the features of the positive exemplar and the features of the negative exemplar including at least one of attribute information of a user, personal interests and vertical industries;

and training according to the characteristics of the positive sample and the characteristics of the negative sample to obtain a prediction model.

5. The method according to claim 4, further comprising, after the periodically acquiring the target identity set of the target information push service:

detecting the number of positive samples of users corresponding to each identifier in the target identifier set;

comparing the number of the positive samples with a preset number of samples;

when the number of the positive samples is smaller than the preset number of samples, acquiring the characteristics of a user corresponding to an information push service with the same type as the target information push service;

the extracting of the characteristics of the user corresponding to each identifier in the target identifier set and the training according to the characteristics to obtain a prediction model comprise:

extracting the characteristics of the user corresponding to each identifier in the target identifier set;

and training according to the characteristics of the user corresponding to each identifier in the target identifier set and the characteristics of the user corresponding to the information push service with the same type to obtain a prediction model.

6. The method of claim 1, wherein the predictive models comprise a factorization model and an ensemble tree model; the training according to the features to obtain a prediction model comprises the following steps:

performing feature conversion and feature splicing processing on the features through the factorization machine model;

and inputting the output of the factorization machine model into the integrated tree model for model training to obtain a prediction model.

7. The method of claim 6, wherein the feature transforming the feature through a factorization model comprises:

respectively matching the characteristics of the user corresponding to each identifier in the target identifier set with preset characteristics;

and taking the feature vector of the preset feature matched with the feature of the user corresponding to each identifier as the feature vector corresponding to each identifier.

8. The method of claim 7, wherein performing a feature stitching process on the features through the factorizer model comprises:

aiming at any identifier in the target identifier set, acquiring a first-dimension feature vector in feature vectors corresponding to the identifier;

performing dimensionality reduction on the first-dimensional feature vector to obtain a second-dimensional feature vector, wherein the dimensionality of the second-dimensional feature vector is smaller than that of the first-dimensional feature vector;

and splicing the second-dimension feature vector with the rest feature vectors except the first-dimension feature vector in the feature vectors corresponding to the identifiers.

9. The method according to claim 1, wherein the updating the target identifier set of the target information push service according to the candidate identifier set comprises:

determining the identifiers which are different in the identifier set to be selected and the target identifier set;

and replacing the different identifiers in the target identifier set with the different identifiers in the to-be-selected identifier set.

10. The method according to claim 1, wherein the updating the target identifier set of the target information push service according to the candidate identifier set comprises:

determining different identifiers in the to-be-selected identifier set and the target identifier set, wherein the different identifiers in the target identifier set are first identifiers, and the different identifiers in the to-be-selected identifier set are second identifiers;

determining the number of the first identifications;

selecting second identifications with the same number as the first identifications;

and replacing the first identifiers with the second identifiers with the same number.

11. The method according to claim 10, wherein the selecting the second identifiers with the same number as the first identifiers comprises:

and selecting second marks from high to low in sequence according to the predicted numerical values to replace the first marks, wherein the number of the selected second marks is the same as that of the first marks.

12. The method according to any one of claims 1 to 11, further comprising:

acquiring the states of all the identifiers in the updated target identifier set, wherein the states comprise a login state and a non-login state;

and pushing the target information to the mark in the updated target mark set in the login state.

13. An information pushing and updating device, characterized in that the device comprises:

14. A computer-readable storage medium, storing a computer program which, when executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 12.

15. A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method of any one of claims 1 to 12.