CN114969546A - Object classification method, and network model training method and device - Google Patents

Object classification method, and network model training method and device Download PDF

Info

Publication number
CN114969546A
CN114969546A CN202210720357.1A CN202210720357A CN114969546A CN 114969546 A CN114969546 A CN 114969546A CN 202210720357 A CN202210720357 A CN 202210720357A CN 114969546 A CN114969546 A CN 114969546A
Authority
CN
China
Prior art keywords
target
conversion rate
content
model
content push
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210720357.1A
Other languages
Chinese (zh)
Inventor
徐靖宇
刘昊骋
徐世界
王天祺
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Baidu Netcom Science and Technology Co Ltd
Original Assignee
Beijing Baidu Netcom Science and Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Baidu Netcom Science and Technology Co Ltd filed Critical Beijing Baidu Netcom Science and Technology Co Ltd
Priority to CN202210720357.1A priority Critical patent/CN114969546A/en
Publication of CN114969546A publication Critical patent/CN114969546A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The disclosure provides an object classification method, a network model training method, an object classification device, a network model training medium and a network model training product, and relates to the field of artificial intelligence, in particular to the technical field of content services. The specific implementation scheme comprises the following steps: dividing a target object set to be classified into a first object subset and a second object subset, wherein the portrait features of target objects in the target object set are distributed uniformly; determining a first conversion rate associated with the target objects in the first subset of objects and determining a second conversion rate associated with the target objects in the second subset of objects; and determining an attribute classification result for the target object set according to a first conversion rate and a second conversion rate, wherein the first conversion rate indicates the forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates the forward feedback probability of the corresponding target object based on the non-content push condition.

Description

Object classification method, and network model training method and device
Technical Field
The disclosure relates to the field of artificial intelligence, in particular to the technical field of content services, and can be applied to scenes such as object classification.
Background
The object attribute type can be determined through object classification, content pushing is carried out according to the object attribute type, and the content pushing effect can be effectively guaranteed. However, in some scenarios, the object classification process has the phenomena of poor accuracy and high cost consumption.
Disclosure of Invention
The disclosure provides an object classification method, a network model training method, an object classification device, a network model training medium and a network model product.
According to an aspect of the present disclosure, there is provided an object classification method including: dividing a target object set to be classified into a first object subset and a second object subset, wherein the portrait features of target objects in the target object set are distributed uniformly; determining a first conversion rate associated with a target object in the first subset of objects and determining a second conversion rate associated with a target object in the second subset of objects; and determining attribute classification results for the set of target objects according to the first conversion rate and the second conversion rate, wherein the first conversion rate indicates a forward feedback probability of the corresponding target object based on content push conditions, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on non-content push conditions.
According to another aspect of the present disclosure, there is provided a training method of a network model, including: determining a model to be trained matched with each sample object according to whether each sample object in the sample object set is a content pushing object; taking the object portrait data of each sample object as input data corresponding to a model to be trained to obtain a prediction conversion rate aiming at each sample object; and adjusting model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object to obtain a trained target network model.
According to another aspect of the present disclosure, there is provided an object classification apparatus including: the first processing module is used for dividing a target object set to be classified into a first object subset and a second object subset, and the portrait features of target objects in the target object set are distributed uniformly; a second processing module to determine a first conversion rate associated with a target object in the first subset of objects and to determine a second conversion rate associated with a target object in the second subset of objects; and a third processing module, configured to determine an attribute classification result for the set of target objects according to the first conversion rate and the second conversion rate, where the first conversion rate indicates a forward feedback probability of the corresponding target object based on a content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on a non-content push condition.
According to another aspect of the present disclosure, there is provided a training apparatus of a network model, including: the eighth processing module is configured to determine a model to be trained that is matched with each sample object in the sample object set according to whether each sample object is a content push object; a ninth processing module, configured to use object portrait data of each sample object as input data corresponding to a model to be trained, to obtain a predicted conversion rate for each sample object; and the tenth processing module is used for adjusting model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object to obtain the trained target network model.
According to another aspect of the present disclosure, there is provided an electronic device including: at least one processor and a memory communicatively coupled to the at least one processor. Wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the object classification method or the network model training method described above.
According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the object classification method or the training method of the network model described above.
According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the object classification method or the training method of a network model as described above.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.
Drawings
The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
fig. 1 schematically illustrates a system architecture of an object classification method and apparatus according to an embodiment of the present disclosure;
FIG. 2 schematically illustrates a flow diagram of an object classification method according to an embodiment of the present disclosure;
FIG. 3 schematically shows a flow chart of an object classification method according to a further embodiment of the present disclosure;
FIG. 4 schematically illustrates a flow chart of a method of training a network model according to an embodiment of the present disclosure;
FIG. 5 schematically shows a schematic diagram of a training process of a network model according to an embodiment of the present disclosure;
fig. 6 schematically shows a block diagram of an object classification apparatus according to an embodiment of the present disclosure;
FIG. 7 schematically illustrates a block diagram of a training apparatus for a network model according to an embodiment of the present disclosure;
fig. 8 schematically shows a block diagram of an electronic device for object classification according to an embodiment of the present disclosure.
Detailed Description
Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.
All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.
Where a convention analogous to "at least one of A, B and C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include but not be limited to systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).
The embodiment of the disclosure provides an object classification method. The method of the embodiment comprises the following steps: dividing a target object set to be classified into a first object subset and a second object subset, wherein the portrait features of target objects in the target object set are distributed uniformly; determining a first conversion rate associated with the target objects in the first subset of objects and determining a second conversion rate associated with the target objects in the second subset of objects; and determining an attribute classification result aiming at the target object set according to the first conversion rate and the second conversion rate. The first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
Fig. 1 schematically shows a system architecture of an object classification method and apparatus according to an embodiment of the present disclosure. It should be noted that fig. 1 is only an example of a system architecture to which the embodiments of the present disclosure may be applied to help those skilled in the art understand the technical content of the present disclosure, and does not mean that the embodiments of the present disclosure may not be applied to other devices, systems, environments or scenarios.
The system architecture 100 according to this embodiment may include a requesting terminal 101, a network 102, and a server 103. Network 102 is the medium used to provide communication links between requesting terminals 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others. The server 103 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud computing, network services, middleware services, and the like.
The requesting terminal 101 interacts with the server 103 through the network 102 to receive or transmit data or the like. The requesting terminal 101 is for example adapted to initiate an object classification request to the server 103, and the requesting terminal 101 is for example further adapted to send object representation data of a set of target objects to be classified to the server 103, which may for example comprise attribute representation data and behavior representation data of the target objects.
The server 103 may be a server that provides various services, and may be, for example, a background processing server (merely an example) that performs object classification processing according to an object classification request transmitted by the requesting terminal 101.
For example, the server 103, in response to an object classification request obtained from the request terminal 101, divides a target object set to be classified into a first object subset and a second object subset, and the portrait features of the target objects in the target object set are uniformly distributed; determining a first conversion rate associated with the target objects in the first subset of objects and determining a second conversion rate associated with the target objects in the second subset of objects; and determining an attribute classification result aiming at the target object set according to the first conversion rate and the second conversion rate. The first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
It should be noted that the object classification method provided by the embodiment of the present disclosure may be executed by the server 103. Accordingly, the object classification apparatus provided by the embodiment of the present disclosure may be disposed in the server 103. The object classification method provided by the embodiments of the present disclosure may also be performed by a server or a server cluster that is different from the server 103 and is capable of communicating with the requesting terminal 101 and/or the server 103. Accordingly, the object classification apparatus provided in the embodiments of the present disclosure may also be disposed in a server or a server cluster that is different from the server 103 and is capable of communicating with the requesting terminal 101 and/or the server 103.
It should be understood that the number of requesting terminals, networks, and servers in fig. 1 is merely illustrative. There may be any number of requesting terminals, networks, and servers, as desired for an implementation.
An object classification method according to an exemplary embodiment of the present disclosure is described below with reference to fig. 2 to 3 in conjunction with the system architecture of fig. 1. The object classification method of the embodiment of the present disclosure may be performed by the server 103 shown in fig. 1, for example.
Fig. 2 schematically shows a flow chart of an object classification method according to an embodiment of the present disclosure.
As shown in fig. 2, the object classification method 200 of the embodiment of the present disclosure may include, for example, operations S210 to S230.
In operation S210, the target object set to be classified is divided into a first object subset and a second object subset, and the portrait features of the target objects in the target object set are uniformly distributed.
In operation S220, a first conversion rate associated with the target objects in the first subset of objects is determined, and a second conversion rate associated with the target objects in the second subset of objects is determined.
In operation S230, an attribute classification result for the target object set is determined according to the first conversion rate and the second conversion rate.
The first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
The following exemplifies each operation example flow of the object classification method of the present embodiment.
For example, the target object set may be uploaded by an object classification requester, or may be reported by a third-party client, where the third-party client may be, for example, a client capable of processing a target service corresponding to the content to be pushed.
For a set of target objects to be classified, the set of target objects may be randomly divided into a first subset of objects and a second subset of objects. The target object set includes a plurality of target objects, and the image features of the plurality of target objects are uniformly distributed. The number of target objects in the first object subset and the second object subset may be the same or different, and this embodiment does not limit this.
A forward feedback probability of the target object based on the content push condition may be determined as the first conversion rate from the object representation data of the target object in the first subset of objects. A forward feedback probability of the target object based on the non-content-push condition is determined as a second conversion rate from the object representation data of the target object in the second subset of objects. The object representation data may include, for example, attribute representation data and behavior representation data of the target object.
The object representation data may be acquired in various public, legally compliant manners, such as from a public data set, or by a data collection facility after obtaining user authorization associated with the object representation data. The object picture data is not scene data for a specific user, and does not reflect personal information of a specific user. The scope of application of the object representation data is limited to the extent that the user has knowledge and authorization to use.
Illustratively, the pushed content may include, for example, pushed marketing content. A forward feedback probability that a target object in the first subset of objects performs a responsive action with respect to the marketing target when pushed marketing content is determined as the first conversion rate. Determining a positive feedback probability that the target objects in the second subset of objects still perform a responsive action against the marketing target without being pushed marketing content as a second conversion rate.
Marketing objectives may include, for example, products, services, events, locations, etc., products may include, for example, tangible products and intangible products, events may include, for example, real-time events to be promoted, etc., locations may include, for example, locations to be promoted, landmark locations, etc. The response behavior for the marketing objective may include, for example, different behaviors such as trading, downloading, clicking, installing, sharing, and so on.
And determining an attribute classification result aiming at the target object set according to the first conversion rate and the second conversion rate. The attribute classification results may include, for example, marketing-sensitive objects, natural conversion objects, immobility objects, reaction objects, and the like.
Illustratively, the marketing-sensitive object performs a positive feedback probability boost of the response behavior under the condition of pushed marketing content. The natural conversion object has a positive feedback probability of executing a response action higher than a target upper limit value regardless of whether marketing contents are pushed or not. The positive feedback probability of the unmoved object performing the response action is lower than the target lower limit value under the condition of whether the marketing content is pushed or not. The reaction type object reduces the probability of performing the forward feedback of the response action under the condition of being pushed the marketing content.
By the embodiment of the disclosure, the target object set to be classified is divided into the first object subset and the second object subset, a first conversion rate of the target objects in the first object subset based on the content push condition is determined, a second conversion rate of the target objects in the second object subset based on the non-content push condition is determined, and the attribute classification result for the target object set is determined according to the first conversion rate and the second conversion rate. And performing attribute classification on the target object set according to the first conversion rate and the second conversion rate, so that the accuracy of an object classification result can be effectively improved, the object classification efficiency can be improved, and the object classification cost can be reduced. By determining the attribute classification result of the target object set, the method is favorable for providing credible decision support for improving the content pushing effect and reducing the content pushing cost.
Fig. 3 schematically shows a flow chart of an object classification method according to another embodiment of the present disclosure.
As shown in fig. 3, the object classification method 300 of the embodiment of the present disclosure may include, for example, operation S210, operation S310, and operation S230.
In operation S210, the target object set to be classified is divided into a first object subset and a second object subset, and the portrait features of the target objects in the target object set are uniformly distributed.
In operation S310, a first conversion rate is output according to the object image data of the target object in the first object subset using the first prediction model, and a second conversion rate is output according to the object image data of the target object in the second object subset using the second prediction model.
In operation S230, an attribute classification result for the target object set is determined according to the first conversion rate and the second conversion rate.
The first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
An exemplary flow of each operation of the object classification method of the present embodiment is illustrated below.
For example, the object representation data of the target object in the first subset of objects may be used as input data of the trained first prediction model to obtain a forward feedback probability of the target object based on the content push condition as the first conversion rate. And taking the object portrait data of the target object in the second object subset as input data of the trained second prediction model to obtain the forward feedback probability of the target object based on the non-content push condition as a second conversion rate.
The first prediction model and the second prediction model may be, for example, a Logistic Regression (LR) model, a Deep Neural Network (DNN) model, a Gradient Boosting Decision Tree (GBDT) model, or a binary model, which is not limited in this embodiment.
The first predictive model may be trained based on object representation data for content push objects and the second predictive model may be trained based on object representation data for non-content push objects. The object portrait data may be, for example, tag attribute data in an important data dimension based on preset portrait tag matching. The object representation data may include attribute representation data, which may include different data such as base attributes, social attributes, interest preferences, APP preferences, and behavior representation data, which may include different data such as trading behaviors, download behaviors, install behaviors, share behaviors.
And determining an attribute classification result aiming at the target object set according to the first conversion rate and the second conversion rate. One example way, a conversion difference may be calculated based on a first conversion statistic associated with a first subset of objects and a second conversion statistic associated with a second subset of objects. And determining an attribute classification result aiming at the target object set according to the first conversion rate, the second conversion rate and the conversion rate difference value. By carefully classifying the target objects, credible decision support is provided for improving the content pushing effect and promoting the content pushing income.
The first conversion rate statistics may for example comprise a mean conversion rate, a median conversion rate, etc. of the target objects in the first subset of objects and the second conversion rate statistics may for example comprise a mean conversion rate, a median conversion rate, etc. of the target objects in the second subset of objects. The average conversion may include, for example, an arithmetic average conversion, a geometric average conversion, a squared average conversion, and the like.
And calculating the difference between the first conversion rate statistic and the second conversion rate statistic to obtain the conversion rate difference. For example, the attribute classification result for the target object set may be determined according to the first conversion rate statistic, the second conversion rate statistic, and the conversion rate difference.
The target objects in the target object set may be determined to be a first type object, which may be, for example, a marketing sensitive object, if the first conversion rate statistic is greater than a first preset threshold, the second conversion rate statistic is less than the first preset threshold, and the conversion rate difference is greater than a second preset threshold. The second preset threshold may take, for example, a value of zero.
The target objects in the target object set may be determined to be objects of a second type, where the first conversion rate statistic is greater than a first preset threshold and the second conversion rate statistic is greater than the first preset threshold, and the objects of the second type may be, for example, natural conversion type objects.
The target objects in the target object set may be determined to be objects of a third type, which may be, for example, unmovable objects, if the first statistical conversion rate is smaller than a third preset threshold and the second statistical conversion rate is smaller than the third preset threshold.
The target objects in the target object set may be determined to be fourth type objects in the case that the first conversion rate statistic is smaller than a first preset threshold, the second conversion rate statistic is larger than the first preset threshold, and the conversion rate difference is smaller than a second preset threshold, and the fourth type objects may be reaction type objects, for example.
One example approach may determine, for a first type of object, a first projected conversion rate for the first type of object based on price content push conditions and a second projected conversion rate for the first type of object based on non-price content push conditions. And determining whether the first type object is a price sensitive object according to the first estimated conversion rate and the second estimated conversion rate. The pushed price contents may include, for example, subsidy information contents, discount information contents, offer information contents, and the like. By carefully classifying the target objects, the content pushing effect can be effectively ensured, and the content flow distribution with obvious response effect can be realized.
By way of example, content push parameters for a set of target objects may be determined based on the attribute targeting results. The content push parameter may indicate at least one of the following information: whether to carry out content push, content to be pushed and a content push mode. The content push manner may indicate information such as a content push medium, content push time, and the like.
For example, content may be pushed to the first type of object in order to raise the probability of positive feedback of the first type of object based on the content push condition. For a price sensitive object in the first type of object, price content can be pushed to the price sensitive object so as to improve the positive feedback probability of the price sensitive object based on the price content pushing condition. For the price non-sensitive object in the first type of object, other content can be pushed to the price non-sensitive object so as to improve the multi-product cross recommendation probability of the price non-sensitive object based on other content pushing conditions, and the other content can comprise marketing reminding content, marketing interesting content and the like.
In one example, in the case that the content push parameter indicates content push, the portrait feature statistics of the target objects in the target object set may be input into a trained random forest model, and a plurality of decision trees may be generated with the portrait feature statistics as classification conditions. And obtaining a decision result aiming at the target object set according to the target leaf nodes matched with the portrait feature statistic values in each decision tree. The maximum depth of each decision tree is smaller than a preset depth threshold value, and the decision result indicates target content to be pushed.
The random forest model is a cluster classification model, and the random forest model can comprise a plurality of decision trees which are independent of each other. Each decision tree may include a plurality of determination nodes for performing attribute determination based on input data and a plurality of classification nodes for performing decision classification based on the attribute determination results of adjacent determination nodes.
The object portrait characteristics of the sample object can be used as a training sample for model training to obtain a random forest model. For example, model parameters of the random forest model to be trained may be obtained, and the model parameters may include, for example, a sample threshold and a depth threshold. And splitting by taking the object portrait characteristics of the sample object as a classification condition to generate a random forest model. The random forest model comprises a plurality of decision trees, the maximum depth of each decision tree is smaller than a depth threshold, leaf nodes of each decision tree can be candidate push contents, and the number of training samples corresponding to each candidate push content is larger than a sample threshold.
When the content push parameter indicates to push the content, the flow distribution value for pushing the content may be determined according to the profile feature distribution type of the target object in the target object set. And pushing the content to the target object in the target object set based on the flow distribution value. For example, in the case where the image feature is an age feature, the image feature distribution type may be, for example, a section to which the age belongs.
Illustratively, the flow distribution value for content push is also determined according to the target content type to be pushed to the target object set. The flow distribution value indicates the proportion of content push objects in the target object set and also indicates the exposure proportion distributed to the target content to be pushed.
By way of example, response behavior data of the content push object based on the pushed content may be obtained. And determining the conversion object ratio for forward feedback of the pushed content in the content pushing objects according to the response behavior data. Determining a content push gain according to the conversion object ratio and the content push cost, and adjusting the flow distribution value based on the content push gain. The positive feedback behavior for the pushed content may include, for example, an installation behavior, a download behavior, a transaction behavior, and the like, which is not limited in this embodiment.
In the case where the content push parameter indicates to perform content push, response behavior data of the target object in the target object set may be acquired. And determining the conversion object proportion which is used for carrying out forward feedback on the pushed content in the target object set according to the response behavior data. And adjusting the model parameters of the first prediction model according to the conversion object ratio to obtain the adjusted first prediction model.
In the case where the content push parameter indicates that content push is not performed, response behavior data of the target object in the target object set may be acquired. And determining the conversion object proportion which is fed back in a forward direction aiming at the target content in the target object set according to the response behavior data. And adjusting the model parameters of the second prediction model according to the conversion object ratio to obtain the adjusted second prediction model.
With the disclosed embodiments, a first conversion rate of the target object based on the content push condition is output from the object representation data of the target object in the first subset of objects using the first prediction model, and a second conversion rate of the target object based on the non-content push condition is output from the object representation data of the target object in the second subset of objects using the second prediction model, and the attribute classification result for the set of target objects is determined from the first conversion rate and the second conversion rate. The accuracy of the object classification result can be effectively guaranteed, the object classification efficiency can be favorably improved, and the cost consumption of object classification can be favorably reduced. The sensitivity of the target object to the content to be pushed can be effectively identified, the target object is accurately classified, so that the content pushing gain can be accurately estimated, and a more accurate and reasonable content pushing strategy can be realized.
FIG. 4 schematically shows a flow chart of a method of training a network model according to an embodiment of the present disclosure.
As shown in FIG. 4, the training method 400 may include operations S410-S430, for example.
In operation S410, a model to be trained that matches each sample object in the sample object set is determined according to whether each sample object is a content push object.
In operation S420, the object image data of each sample object is used as input data corresponding to the model to be trained, and a predicted conversion rate for each sample object is obtained.
In operation S430, the model parameters corresponding to the model to be trained are adjusted according to the predicted conversion rate and the preset conversion label of each sample object, so as to obtain a trained target network model.
An example flow of each operation of the model training method of the present embodiment is illustrated below.
Illustratively, it is determined whether each sample object in the set of sample objects is a content push object. And according to whether each sample object in the sample object set is a content pushing object or not, taking the object portrait data of each sample object as the input data of the corresponding model to be trained to obtain the predicted conversion rate aiming at each sample object.
The model to be trained can be utilized to perform feature extraction on object portrait data of the sample object, so as to obtain an initial feature vector based on at least one feature dimension. And screening a preset number of characteristic values with the maximum dimension weight from the initial characteristic vectors according to the dimension weight of each characteristic dimension to obtain target characteristic vectors. Based on the target feature vector, a predicted conversion rate of the sample object based on the content push condition or the non-content push condition is output. By screening the model training characteristics, the problem of overlarge calculated amount of the machine learning model can be effectively solved, the model training speed is favorably improved, and the prediction precision of marketing transformation groups is favorably improved.
Illustratively, in the model training process, a preset number of feature values with the largest dimension weight can be screened according to a multi-way recall strategy and a chi-square test algorithm to obtain a target feature vector. The idea of the chi-squared test algorithm is to determine whether the assumption of independence between variables holds based on the deviation of the actual values from the assumed values (theoretical values obtained by assuming independence between the variables).
For any target feature dimension in the at least one feature dimension, a precision gain coefficient of the model to be trained based on the target feature dimension may be determined. And determining the dimension weight matched with the target feature dimension according to the precision gain coefficient.
For example, the object portrait data corresponding to the target feature dimension may be input into a corresponding model to be trained, so as to obtain a first MSE (Mean Squared Error) score of the model to be trained. And determining the precision gain coefficient of the model to be trained based on the target feature dimension by combining the second MSE score when the object portrait data corresponding to the target feature dimension is not input into the model to be trained.
The MSE score is used for measuring the deviation between the predicted value and the actual value of the model, and the specific calculation formula can be as follows:
Figure BDA0003708649650000121
P i representing the predicted conversion, P, for the ith sample object i ' denotes an actual conversion rate of the i-th sample object, and the actual conversion rate may be 0 or 1, for example.
For example, the accuracy gain factor associated with the target feature dimension can be represented by equation (1),
Figure BDA0003708649650000122
MSE 1 representing a first MSE score, MSE, when object portrait data corresponding to the target feature dimensions are not input into the model to be trained 2 And representing a second MSE score when the object portrait data corresponding to the target characteristic dimension is input into the model to be trained. The precision gain coefficient g may be used as a dimension weight of the target feature dimension, the dimension weight indicates a degree of influence of the target feature dimension on the prediction conversion rate, and the target feature dimension may be any feature dimension of the at least one feature dimension.
For example, when the sample object is a content push object, the object image data of the sample object may be used as the input data of the first model to be trained, and the first predicted conversion rate for each sample object may be obtained. And adjusting model parameters of the first model to be trained according to the first prediction conversion rate and the first conversion label of the sample object to obtain a trained target network model to be used as a first prediction model. The first predicted conversion rate indicates a forward feedback probability of the corresponding sample object based on the content push condition, and the first conversion label indicates an actual conversion situation of the corresponding sample object based on the content push condition.
And when the sample object is a non-content pushing object, the object image data of the sample object is used as the input data of the second model to be trained, and a second prediction conversion rate for each sample object is obtained. And adjusting the model parameters of the second model to be trained according to the second prediction conversion rate and the second conversion label of the sample object to obtain a trained target network model to be used as a second prediction model. The second predicted conversion rate indicates a forward feedback probability of the sample object based on the non-content push condition, and the second conversion label indicates an actual conversion situation of the corresponding sample object based on the non-content push condition.
According to the method and the device, a first prediction model for predicting the first conversion rate of the target object based on the content pushing condition is trained according to the object portrait data of the content pushing object. And training to obtain a second prediction model for predicting a second conversion rate of the target object based on the non-content push condition according to the object portrait data of the non-content push object. The method is beneficial to assisting in analyzing the attribute classification result of the target object set, can effectively improve the object classification precision and effectively improve the object classification efficiency. The method is favorable for improving the prediction precision of marketing conversion groups and providing credible decision support for improving the content pushing effect.
Fig. 5 schematically shows a schematic diagram of a training process of a network model according to an embodiment of the present disclosure.
As shown in fig. 5, a content push object 502A and a non-content push object 502B in a sample object set 501 are determined. Content push objects 502A may be, for example, sample objects that have been pushed with marketing content, and non-content push objects 502B may be, for example, sample objects that have not been pushed with marketing content.
For content push object 502A, a first transformation tag 503A and first object representation data 504A associated with content push object 502A are determined. The first conversion label 503A indicates the actual conversion of the corresponding sample object based on the content push condition. The first object image data 504A is used as input data of the first model to be trained 505A to obtain a first predicted conversion rate for the content-pushed object 502A. According to the first prediction conversion rate and the first conversion label 503A, the model parameters of the first model to be trained 505A are adjusted to obtain a first prediction model 506A.
For non-content push object 502B, a second conversion label 503B and second object representation data 504B associated with non-content push object 502B are determined. The second conversion label 503B indicates the actual conversion of the corresponding sample object based on the non-content push condition. The second object representation data 504B is used as input data for the second model 505B to be trained, resulting in a second predicted conversion rate for the non-content-pushed object 502B. And adjusting the model parameters of the second model 505B to be trained according to the second predicted conversion rate and the second conversion label 503B to obtain a second prediction model 506B.
For the target object set to be classified, the target object set may be divided into a first object subset and a second object subset, and the portrait features of the target objects in the target object set are uniformly distributed. A first conversion rate 507A of the corresponding target object based on the content push conditions may be determined from the object representation data of the target objects in the first subset of objects using the first predictive model 506A. A second conversion rate 507B based on non-content push conditions for the target object is determined from the object representation data for the target object in the second subset of objects using a second predictive model 506B.
The attribute classification result 508 for the set of target objects may be determined according to a first conversion rate 507A based on content push conditions and a second conversion rate 507B based on non-content push conditions. From the attribute classification result 508, content push parameters 509 for the set of target objects are determined.
By introducing the prediction model constructed based on the object portrait data of the content push object and the non-content push object, the method is beneficial to realizing the detailed classification of the target object set, and can effectively ensure the accuracy of the object classification result. The method and the device are favorable for providing credible decision support for improving the content pushing effect and improving the content pushing efficiency, and are favorable for more accurately and pertinently pushing the content.
Fig. 6 schematically shows a block diagram of an object classification apparatus according to an embodiment of the present disclosure.
As shown in fig. 6, the object classification apparatus 600 of the embodiment of the present disclosure includes, for example, a first processing module 610, a second processing module 620, and a third processing module 630.
The first processing module 610 is configured to divide a target object set to be classified into a first object subset and a second object subset, where the portrait features of target objects in the target object set are uniformly distributed; a second processing module 620 for determining a first conversion rate associated with the target objects in the first subset of objects and determining a second conversion rate associated with the target objects in the second subset of objects; and a third processing module 630, configured to determine an attribute classification result for the set of target objects according to a first conversion rate and a second conversion rate, where the first conversion rate indicates a forward feedback probability that the corresponding target object is based on the content push condition, and the second conversion rate indicates a forward feedback probability that the corresponding target object is based on the non-content push condition.
By the embodiment of the disclosure, the target object set to be classified is divided into the first object subset and the second object subset, a first conversion rate of the target objects in the first object subset based on the content push condition is determined, a second conversion rate of the target objects in the second object subset based on the non-content push condition is determined, and the attribute classification result for the target object set is determined according to the first conversion rate and the second conversion rate. And attribute classification is carried out on the target object set according to the first conversion rate and the second conversion rate, so that the accuracy of an object classification result can be effectively improved, the object classification efficiency can be improved, and the object classification cost can be reduced. By determining the attribute classification result of the target object set, the method is favorable for providing credible decision support for improving the content pushing effect and reducing the content pushing cost.
According to an embodiment of the present disclosure, the second processing module includes: the first processing submodule is used for outputting a first conversion rate according to the object image data of the target object in the first object subset by utilizing a first prediction model, and the first prediction model is obtained by training the object image data of the content push object; and a second processing sub-module configured to output a second conversion rate according to the object image data of the target object in the second object subset by using a second prediction model, where the second prediction model is obtained by training according to the object image data of the non-content push object.
According to an embodiment of the present disclosure, the third processing module includes: a third processing sub-module for calculating a conversion probability difference based on the first conversion statistics associated with the first subset of objects and the second conversion statistics associated with the second subset of objects; and the fourth processing submodule is used for determining an attribute classification result aiming at the target object set according to the first conversion rate, the second conversion rate and the conversion probability difference value.
According to an embodiment of the present disclosure, the apparatus further includes a fourth processing module configured to: determining a content pushing parameter aiming at the target object set according to the attribute classification result, wherein the content pushing parameter indicates at least one of the following information: whether to push content, content to be pushed and a content pushing mode.
According to an embodiment of the present disclosure, the apparatus further includes a fifth processing module, which includes: the fifth processing submodule is used for determining a flow distribution value for content pushing according to the portrait feature distribution type of the target object in the target object set under the condition that the content pushing parameter indicates to carry out content pushing; and a sixth processing submodule, configured to perform content push on the target objects in the target object set based on a traffic distribution value, where the traffic distribution value indicates a proportion of content push objects in the target object set.
According to an embodiment of the present disclosure, the apparatus further includes a sixth processing module, which includes: the seventh processing submodule is used for acquiring response behavior data of the content push object based on the pushed content; the eighth processing submodule is used for determining the conversion object proportion which carries out forward feedback on the pushed content in the content pushing objects according to the response behavior data; a ninth processing submodule, configured to determine a content push gain according to the conversion object proportion and the content push cost; and a tenth processing submodule for adjusting the traffic distribution value based on the content push gain.
According to an embodiment of the present disclosure, the apparatus further includes a seventh processing module, which includes: an eleventh processing sub-module, configured to, when the content push parameter indicates that content push is performed, input the image feature statistics of the target object in the target object set into the trained random forest model to obtain multiple decision trees generated based on the image feature statistics, where a maximum depth of each decision tree is less than a preset depth threshold; and the twelfth processing submodule is used for obtaining a decision result aiming at the target object set according to the target leaf node matched with the portrait feature statistic value in each decision tree, and the decision result indicates target content to be pushed.
Fig. 7 schematically shows a block diagram of a training apparatus of a network model according to an embodiment of the present disclosure.
As shown in fig. 7, the training apparatus 700 for a network model according to the embodiment of the present disclosure includes, for example, an eighth processing module 710, a ninth processing module 720, and a tenth processing module 730.
An eighth processing module 710, configured to determine a model to be trained that is matched with each sample object in the sample object set according to whether each sample object is a content push object; a ninth processing module 720, configured to use the object portrait data of each sample object as input data of a corresponding model to be trained, to obtain a predicted conversion rate for each sample object; and a tenth processing module 730, configured to adjust model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object, so as to obtain a trained target network model.
According to the embodiment of the disclosure, a first prediction model for predicting the first conversion rate of the target object based on the content push condition is trained according to the object portrait data of the content push object. And training to obtain a second prediction model for predicting a second conversion rate of the target object based on the non-content push condition according to the object portrait data of the non-content push object. The method is beneficial to assisting in analyzing the attribute classification result of the target object set, can effectively improve the object classification precision and effectively improve the object classification efficiency. The method is favorable for improving the prediction precision of marketing conversion groups and providing credible decision support for improving the content push effect.
According to an embodiment of the present disclosure, the ninth processing module includes: the thirteenth processing submodule is used for extracting the characteristics of the object portrait data of the sample object by using the model to be trained to obtain an initial characteristic vector based on at least one characteristic dimension; a fourteenth processing submodule, configured to screen a preset number of feature values with the largest dimension weight from the initial feature vectors according to the dimension weight of each feature dimension, so as to obtain a target feature vector; and a fifteenth processing submodule for outputting a predicted conversion rate for the sample object based on the target feature vector, the dimension weight indicating a degree of influence of the corresponding feature dimension on the predicted conversion rate.
According to an embodiment of the present disclosure, the apparatus further includes an eleventh processing module, the eleventh processing module including: a sixteenth processing sub-module, configured to determine, for any target feature dimension in the at least one feature dimension, a precision gain coefficient of the model to be trained based on the target feature dimension; and a seventeenth processing submodule, configured to determine, according to the precision gain coefficient, a dimension weight matched with the target feature dimension.
According to an embodiment of the present disclosure, the tenth processing module includes: the eighteenth processing submodule is used for adjusting the model parameters of the first model to be trained according to the first prediction conversion rate and the first conversion label of the sample object under the condition that the sample object is a content pushing object, so as to obtain a trained target network model as a first prediction model; and the nineteenth processing submodule is used for adjusting model parameters of a second model to be trained according to a second prediction conversion rate and a second conversion label of the corresponding sample object under the condition that the sample object is a non-content push object to obtain a trained target network model as a second prediction model, wherein the first prediction conversion rate indicates the forward feedback probability of the corresponding sample object based on the content push condition, and the second prediction conversion rate indicates the forward feedback probability of the corresponding sample object based on the non-content push condition.
It should be noted that in the technical solutions of the present disclosure, the processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the related information are all in accordance with the regulations of the related laws and regulations, and do not violate the customs of the public order.
The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.
Fig. 8 schematically shows a block diagram of an electronic device for object classification according to an embodiment of the present disclosure.
FIG. 8 illustrates a schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure. The electronic device 800 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.
As shown in fig. 8, the apparatus 800 includes a computing unit 801 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM)802 or a computer program loaded from a storage unit 808 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The calculation unit 801, the ROM 802, and the RAM 803 are connected to each other by a bus 804. An input/output (I/O) interface 805 is also connected to bus 804.
A number of components in the device 800 are connected to the I/O interface 805, including: an input unit 806 such as a keyboard, a mouse, or the like; an output unit 807 such as various types of displays, speakers, and the like; a storage unit 808, such as a magnetic disk, optical disk, or the like; and a communication unit 809 such as a network card, modem, wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
Computing unit 801 may be a variety of general and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running deep learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and the like. The calculation unit 801 executes the respective methods and processes described above, such as the object classification method. For example, in some embodiments, the object classification method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and/or installed onto device 800 via ROM 802 and/or communications unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the object classification method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the object classification method in any other suitable manner (e.g., by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), Complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable model training apparatus, such that the program codes, when executed by the processor or controller, cause the functions/acts specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with an object, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to an object; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which objects can provide input to the computer. Other kinds of devices may also be used to provide for interaction with an object; for example, feedback provided to the subject can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the object can be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., an object computer having a graphical object interface or a web browser through which objects can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server with a combined blockchain.
It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel or sequentially or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.
The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims (25)

1. An object classification method, comprising:
dividing a target object set to be classified into a first object subset and a second object subset, wherein the portrait features of target objects in the target object set are distributed uniformly;
determining a first conversion rate associated with a target object in the first subset of objects and determining a second conversion rate associated with a target object in the second subset of objects; and
determining an attribute classification result for the set of target objects according to the first conversion rate and the second conversion rate,
wherein the first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
2. The method of claim 1, wherein the determining a first conversion rate associated with a target object in the first subset of objects and determining a second conversion rate associated with a target object in the second subset of objects comprises:
outputting the first conversion rate according to object portrait data of a target object in the first object subset by using a first prediction model, wherein the first prediction model is obtained by training according to the object portrait data of a content push object; and
and outputting the second conversion rate according to the object portrait data of the target object in the second object subset by using a second prediction model, wherein the second prediction model is obtained by training according to the object portrait data of the non-content push object.
3. The method of claim 1, wherein the determining a property classification result for the set of target objects from the first conversion rate and the second conversion rate comprises:
calculating a conversion probability difference based on a first conversion rate statistic associated with the first subset of objects and a second conversion rate statistic associated with the second subset of objects; and
and determining an attribute classification result aiming at the target object set according to the first conversion rate, the second conversion rate and the conversion probability difference value.
4. The method of any of claims 1 to 3, further comprising:
determining a content push parameter for the target object set according to the attribute classification result, wherein the content push parameter indicates at least one of the following information:
whether to carry out content push, content to be pushed and a content push mode.
5. The method of claim 4, further comprising:
under the condition that the content push parameters indicate content push, determining a flow distribution value for content push according to the portrait feature distribution type of the target object in the target object set; and
and performing content pushing on the target objects in the target object set based on the traffic distribution value, wherein the traffic distribution value indicates the proportion of the content pushing objects in the target object set.
6. The method of claim 5, further comprising:
acquiring response behavior data of the content push object based on pushed content;
determining a conversion object proportion for performing forward feedback on the pushed content in the content pushing object according to the response behavior data;
determining content push gain according to the conversion object ratio and the content push cost; and
adjusting the traffic allocation value based on the content push gain.
7. The method of claim 4, further comprising:
under the condition that the content push parameters indicate content push, inputting the portrait feature statistics of the target objects in the target object set into a trained random forest model to obtain a plurality of decision trees generated based on the portrait feature statistics, wherein the maximum depth of each decision tree is smaller than a preset depth threshold; and
and obtaining a decision result aiming at the target object set according to the target leaf nodes matched with the portrait feature statistic values in each decision tree, wherein the decision result indicates target content to be pushed.
8. A training method of a network model comprises the following steps:
determining a model to be trained matched with each sample object according to whether each sample object in the sample object set is a content pushing object;
taking the object portrait data of each sample object as input data corresponding to a model to be trained to obtain a prediction conversion rate aiming at each sample object; and
and adjusting model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object to obtain a trained target network model.
9. The method of claim 8, wherein said using the object representation data of each of the sample objects as input data corresponding to a model to be trained to derive a predicted conversion rate for each of the sample objects comprises:
performing feature extraction on the object portrait data of the sample object by using the model to be trained to obtain an initial feature vector based on at least one feature dimension;
screening a preset number of characteristic values with the maximum dimension weight from the initial characteristic vectors according to the dimension weight of each characteristic dimension to obtain target characteristic vectors; and
outputting the predicted conversion rate for the sample object based on the target feature vector,
wherein the dimension weight indicates a degree of influence of the corresponding feature dimension on the predicted conversion rate.
10. The method of claim 9, further comprising:
for any target feature dimension in the at least one feature dimension, determining a precision gain coefficient of the model to be trained based on the target feature dimension; and
and determining the dimension weight matched with the target feature dimension according to the precision gain coefficient.
11. The method of claim 8, wherein the adjusting model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object to obtain the trained target network model comprises:
under the condition that the sample object is a content pushing object, adjusting model parameters of a first model to be trained according to a first prediction conversion rate and a first conversion label of the sample object to obtain a trained target network model serving as a first prediction model; and
under the condition that the sample object is a non-content pushing object, adjusting model parameters of a second model to be trained according to a second prediction conversion rate and a second conversion label of the corresponding sample object to obtain a trained target network model serving as a second prediction model,
wherein the first predicted conversion rate indicates a forward feedback probability of the corresponding sample object based on content push conditions, and the second predicted conversion rate indicates a forward feedback probability of the corresponding sample object based on non-content push conditions.
12. An object classification apparatus comprising:
the system comprises a first processing module, a second processing module and a third processing module, wherein the first processing module is used for dividing a target object set to be classified into a first object subset and a second object subset, and the portrait features of target objects in the target object set are distributed uniformly;
a second processing module to determine a first conversion rate associated with a target object in the first subset of objects and to determine a second conversion rate associated with a target object in the second subset of objects; and
a third processing module for determining an attribute classification result for the set of target objects according to the first conversion rate and the second conversion rate,
wherein the first conversion rate indicates a forward feedback probability of the corresponding target object based on the content push condition, and the second conversion rate indicates a forward feedback probability of the corresponding target object based on the non-content push condition.
13. The apparatus of claim 12, wherein the second processing module comprises:
a first processing sub-module, configured to output the first conversion rate according to object portrait data of a target object in the first object subset by using a first prediction model, where the first prediction model is obtained by training object portrait data of a content-pushed object; and
and the second processing submodule is used for outputting the second conversion rate according to the object portrait data of the target object in the second object subset by using a second prediction model, wherein the second prediction model is obtained by training according to the object portrait data of the non-content push object.
14. The apparatus of claim 12, wherein the third processing module comprises:
a third processing sub-module for calculating a conversion probability difference based on the first conversion statistics associated with the first subset of objects and the second conversion statistics associated with the second subset of objects; and
and the fourth processing submodule is used for determining an attribute classification result aiming at the target object set according to the first conversion rate, the second conversion rate and the conversion probability difference value.
15. The apparatus according to any of claims 12 to 14, further comprising a fourth processing module for:
determining a content push parameter for the target object set according to the attribute classification result, wherein the content push parameter indicates at least one of the following information:
whether to carry out content push, content to be pushed and a content push mode.
16. The apparatus of claim 15, further comprising a fifth processing module,
the fifth processing module includes:
a fifth processing submodule, configured to determine, when the content push parameter indicates to perform content push, a flow distribution value for performing content push according to an image feature distribution type of a target object in the target object set; and
and a sixth processing sub-module, configured to perform content push on a target object in the target object set based on the traffic allocation value, where the traffic allocation value indicates a proportion of content push objects in the target object set.
17. The apparatus of claim 16, further comprising a sixth processing module,
the sixth processing module includes:
a seventh processing sub-module, configured to obtain response behavior data of the content push object based on the pushed content;
an eighth processing submodule, configured to determine, according to the response behavior data, a conversion object proportion for performing forward feedback on the pushed content in the content pushing object;
a ninth processing submodule, configured to determine a content push gain according to the conversion object proportion and the content push cost; and
a tenth processing submodule, configured to adjust the traffic distribution value based on the content push gain.
18. The apparatus of claim 15, further comprising a seventh processing module,
the seventh processing module includes:
an eleventh processing sub-module, configured to, when the content push parameter indicates to perform content push, input the portrait feature statistics of the target objects in the target object set into a trained random forest model to obtain a plurality of decision trees generated based on the portrait feature statistics, where a maximum depth of each decision tree is smaller than a preset depth threshold; and
and the twelfth processing submodule is used for obtaining a decision result aiming at the target object set according to the target leaf node matched with the portrait feature statistic value in each decision tree, wherein the decision result indicates target content to be pushed.
19. An apparatus for training a network model, comprising:
the eighth processing module is configured to determine a model to be trained, which is matched with each sample object in the sample object set, according to whether each sample object in the sample object set is a content push object;
a ninth processing module, configured to use object portrait data of each sample object as input data corresponding to a model to be trained, to obtain a predicted conversion rate for each sample object; and
and the tenth processing module is used for adjusting model parameters corresponding to the model to be trained according to the predicted conversion rate and the preset conversion label of each sample object to obtain the trained target network model.
20. The apparatus of claim 19, wherein the ninth processing module comprises:
a thirteenth processing submodule, configured to perform feature extraction on the object portrait data of the sample object by using the model to be trained, to obtain an initial feature vector based on at least one feature dimension;
a fourteenth processing submodule, configured to screen a preset number of feature values with the largest dimension weight from the initial feature vectors according to the dimension weight of each feature dimension, so as to obtain a target feature vector; and
a fifteenth processing sub-module for outputting the predicted conversion for the sample object based on the target feature vector,
wherein the dimension weight indicates a degree of influence of the corresponding feature dimension on the predicted conversion rate.
21. The apparatus of claim 20, further comprising an eleventh processing module,
the eleventh processing module, comprising:
a sixteenth processing submodule, configured to determine, for any target feature dimension in the at least one feature dimension, a precision gain coefficient of the model to be trained based on the target feature dimension; and
and the seventeenth processing submodule is used for determining the dimension weight matched with the target feature dimension according to the precision gain coefficient.
22. The apparatus of claim 19, wherein the tenth processing module comprises:
an eighteenth processing submodule, configured to, when the sample object is a content push object, adjust a model parameter of a first model to be trained according to a first prediction conversion rate and a first conversion label of the sample object, to obtain a trained target network model, which is used as a first prediction model; and
a nineteenth processing sub-module, configured to, when the sample object is a non-content push object, adjust a model parameter of a second model to be trained according to a second prediction conversion rate and a second conversion label of a corresponding sample object, to obtain a trained target network model as a second prediction model,
wherein the first predicted conversion rate indicates a forward feedback probability of the corresponding sample object based on the content push condition, and the second predicted conversion rate indicates a forward feedback probability of the corresponding sample object based on the non-content push condition.
23. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method for object classification of any one of claims 1 to 7 or the method for training a network model of any one of claims 8 to 11.
24. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the object classification method of any one of claims 1 to 7 or the training method of the network model of any one of claims 8 to 11.
25. A computer program product comprising a computer program which, when executed by a processor, implements an object classification method as claimed in any one of claims 1 to 7, or implements a training method for a network model as claimed in any one of claims 8 to 11.
CN202210720357.1A 2022-06-22 2022-06-22 Object classification method, and network model training method and device Pending CN114969546A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210720357.1A CN114969546A (en) 2022-06-22 2022-06-22 Object classification method, and network model training method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210720357.1A CN114969546A (en) 2022-06-22 2022-06-22 Object classification method, and network model training method and device

Publications (1)

Publication Number Publication Date
CN114969546A true CN114969546A (en) 2022-08-30

Family

ID=82966402

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210720357.1A Pending CN114969546A (en) 2022-06-22 2022-06-22 Object classification method, and network model training method and device

Country Status (1)

Country Link
CN (1) CN114969546A (en)

Similar Documents

Publication Publication Date Title
CN114265979B (en) Method for determining fusion parameters, information recommendation method and model training method
CN113159355A (en) Data prediction method, data prediction device, logistics cargo quantity prediction method, medium and equipment
CN113743971A (en) Data processing method and device
CN113392920B (en) Method, apparatus, device, medium, and program product for generating cheating prediction model
CN113205189B (en) Method for training prediction model, prediction method and device
CN113051911B (en) Method, apparatus, device, medium and program product for extracting sensitive words
CN113516185A (en) Model training method and device, electronic equipment and storage medium
CN113778979A (en) Method and device for determining live broadcast click rate
CN116342164A (en) Target user group positioning method and device, electronic equipment and storage medium
CN111768218A (en) Method and device for processing user interaction information
CN116228284A (en) Goods demand prediction method, training device, computer system and medium
CN114969546A (en) Object classification method, and network model training method and device
CN114610953A (en) Data classification method, device, equipment and storage medium
CN115062687A (en) Enterprise credit monitoring method, device, equipment and storage medium
CN114021642A (en) Data processing method and device, electronic equipment and storage medium
CN113807391A (en) Task model training method and device, electronic equipment and storage medium
CN113361621A (en) Method and apparatus for training a model
CN112434083A (en) Event processing method and device based on big data
CN114818843A (en) Data analysis method and device and computing equipment
CN113010782A (en) Demand amount acquisition method and device, electronic equipment and computer readable medium
CN115100461B (en) Image classification model training method and device, electronic equipment and medium
CN114065074A (en) Audience group acquisition method, and training method and device of user object matching model
CN115482422A (en) Deep learning model training method, image processing method and device
CN114037061A (en) Pre-training model generation method and device, electronic equipment and storage medium
CN116342183A (en) Information recommendation method, training device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination