CN111369315A

CN111369315A - Resource object recommendation method and device, and data prediction model training method and device

Info

Publication number: CN111369315A
Application number: CN202010125250.3A
Authority: CN
Inventors: 周瑜; 赵彬杰; 谢金锦
Original assignee: Rajax Network Technology Co Ltd
Current assignee: Rajax Network Technology Co Ltd
Priority date: 2020-02-27
Filing date: 2020-02-27
Publication date: 2020-07-03

Abstract

The application discloses a resource object recommendation method and device, a training method and device of a data prediction model, a computer storage medium and electronic equipment, wherein the recommendation method comprises the following steps: acquiring the selected resource object data and the target resource object data; inputting the selected resource object data and the target resource object data into a data prediction model, and obtaining the probability that the target resource object data can be added into the selected resource object data, wherein the data prediction model is used for predicting the probability that the target resource object data can be added into the selected resource object data; and determining the target resource object data with the probability meeting the recommendation requirement as recommended resource object data. The recommendation method can avoid the problems of recommendation weakness and cold start of new commodities caused by low commodity moving and selling rate, and avoid the problems of low recommendation data conversion rate, space waste and resource information waste.

Description

Resource object recommendation method and device, and data prediction model training method and device

Technical Field

The application relates to the technical field of computers, in particular to a resource object recommendation method and device, a data prediction model training method and device, a computer storage medium and electronic equipment.

Background

With the development of the internet, electronic application services based on the internet are widely applied in the aspects of life. In order to improve the utilization rate of internet resources and provide data based on electronic application services quickly and efficiently, data mining is carried out accordingly.

In the prior art, data similar to historical data is mined out in a machine learning mode according to existing historical data and is recommended as recommended data, which is usually adopted when data mining is performed based on electronic application services, however, the recommended data has at least two problems:

one is that, when providing recommended data for different users, recommendation of recommended data is usually performed only when the same data is purchased by both users, for example: the user A and the user B purchase the commodity G together, and the commodity a purchased by the user A is recommended to the user B or the commodity B purchased by the user B is recommended to the user A according to the purchasing relation, so that the recommendation of similar data is completed.

Secondly, no matter the data recommendation is performed for the same user or different users, in order to improve the accuracy of mining data, a large amount of historical data is required to be relied on, so that accurate recommendation data can be obtained on the premise of accumulating a large amount of data.

Disclosure of Invention

The application provides a resource object recommendation method, which aims to solve the problem that the accuracy rate of recommended data is low in the prior art.

The application provides a resource object recommendation method, which comprises the following steps:

acquiring the selected resource object data and the target resource object data;

inputting the selected resource object data and the target resource object data into a data prediction model, and obtaining the probability that the target resource object data can be added into the selected resource object data, wherein the data prediction model is used for predicting the probability that the target resource object data can be added into the selected resource object data;

and determining the target resource object data with the probability meeting the recommendation requirement as recommended resource object data.

Optionally, the inputting the selected resource object data and the target resource object data into a data prediction model to obtain a probability that the target resource object data can be added to the selected resource object data includes:

extracting sequence attribute features in the selected resource object data according to text information which is used for describing the content of the selected resource object in the selected resource object data;

according to text information used for describing the content of the target resource object data in the target resource object data, extracting the sequence attribute features in the target resource object data, wherein the sequence attribute features are the same as the sequence attribute features of the selected resource object data in feature type;

and inputting the sequence attribute characteristics of the selected resource object data and the sequence attribute characteristics of the target resource object data into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

Optionally, the feature types of the sequence data features of the selected resource object data include: the name characteristic type is used for describing the selected resource object data subject information, the price characteristic type is used for describing the selected resource object data payment information, and the specification characteristic type is used for describing the resource object data specification information.

Optionally, the method further includes:

extracting portrait information used for describing the target resource object data in the target resource object data;

inputting the selected resource object data and the target resource object data into a data prediction model to obtain a probability that the target resource object data can be added to the selected resource object data, including:

and inputting the extracted sequence attribute features of the resource object data, the sequence attribute features of the target resource object data and the portrait information of the target resource object data into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

Optionally, the extracting of the portrait information in the target resource object data, which is used to describe the target resource object data, includes:

performing data characteristic expansion processing on the image information of the target resource object data to obtain first expansion characteristic data of the target resource object data;

inputting the extracted sequence attribute features of the resource object data, the extracted sequence attribute features of the target resource object data and the portrait information of the target resource object data into the data prediction model, and obtaining the probability that the target resource object data can be added into the selected resource object data comprises:

and inputting the extracted sequence attribute feature of the resource object data, the sequence attribute feature of the target resource object data, the portrait information of the target resource object data and the first extended feature data into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

Optionally, the performing data feature expansion processing on the image information of the target resource object data to obtain first expansion feature data of the target resource object data includes:

and performing difference calculation on two adjacent preferential data describing the target resource object in the portrait information of the target resource object, and performing data feature expansion processing in a manner of square or logarithm of the difference to obtain first expansion feature data of the target resource object.

Optionally, the method further includes:

extracting portrait information for describing the resource object selector;

and inputting the extracted sequence attribute characteristics of the resource object data, the sequence attribute characteristics of the target resource object data and the portrait information of the resource object selector into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

Optionally, the extracting of the portrait information describing the resource object selector includes:

performing data characteristic expansion processing on the portrait information of the resource object selector to obtain second expansion characteristic data of the resource object selector;

inputting the extracted sequence attribute feature of the resource object data, the extracted sequence attribute feature of the target resource object data, and the portrait information of the resource object selector into the data prediction model to obtain the probability that the target resource object data can be added to the selected resource object data, including:

and inputting the extracted sequence attribute feature of the resource object data, the sequence attribute feature of the target resource object data, the portrait information of the resource object selector and the second extended feature data into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

Optionally, the performing data feature expansion processing on the portrait information of the resource object selector to obtain second expansion feature data of the resource object selector includes:

and calculating the difference of the two adjacent login durations of the resource object selector described in the portrait information of the resource object selector, and performing data characteristic expansion processing in a manner of square or logarithm of the difference to obtain second expansion characteristic data of the resource object selector.

Optionally, the acquiring the resource object data that has been selected includes:

and acquiring the resource object data meeting the acquisition requirement in the selected resource object data.

acquiring the selected resource object data added to the selection list; or acquiring historical resource object data which completes the confirmation operation aiming at the selected resource object data added to the selection list.

Optionally, the method further includes:

and outputting the determined recommended resource object data to an interface of the selected resource object data for output.

The present application further provides a resource object recommendation device, including:

an acquisition unit configured to acquire the selected resource object data and the target resource object data;

a prediction unit, configured to input the selected resource object data and the target resource object data into a data prediction model, and obtain a probability that the target resource object data can be added to the selected resource object data, where the data prediction model is used to predict the probability that the target resource object data can be added to the selected resource object data;

and the determining unit is used for determining the target resource object data with the probability meeting the recommendation requirement as recommended resource object data.

The application also provides a training method of the data prediction model, which comprises the following steps:

collecting sample data comprising the selected resource object data and the unselected resource object data;

inputting the sample data into a data prediction model for training;

according to the training result, obtaining the correlation degree between the selected resource object data in the sample data;

and determining the probability of being jointly selected among the resource object data according to the correlation.

Optionally, the inputting the sample data into a data prediction model for training includes:

extracting sequence attribute features in the unselected resource object data according to text information which is used for describing the unselected resource object content in the unselected resource object data;

and inputting the sequence attribute characteristics of the selected resource object data, the sequence attribute characteristics of the target resource object data and the sequence attribute characteristics of the unselected resource object data into the data prediction model for training.

Optionally, the feature types of the sequence data features of the selected resource object data include: the name characteristic type is used for describing the subject information of the selected resource object data, the price characteristic type is used for describing the payment information of the selected resource object data, and the specification characteristic type is used for describing the specification information of the selected resource object data.

Optionally, the inputting the sample data into a data prediction model for training further includes extracting at least one of the following information: extracting portrait information in the collected sample data for describing the selected resource object data, extracting portrait information in the collected sample data for describing a selecting party of the selected resource object data, and extracting portrait information in the collected sample data for describing the unselected resource object data;

and inputting the sequence attribute characteristics of the selected resource object data, the portrait information of the selecting party, the sequence attribute characteristics of the unselected resource object data and the portrait information of the unselected resource object data into the data prediction model for training.

Optionally, the extracting of the portrait information in the collected sample data for describing the selected resource object data and the extracting of the portrait information in the collected sample data for describing the selecting party of the selected resource object data include:

performing data feature expansion processing on the extracted image information of the selecting party to obtain first expansion feature data of the selecting party;

performing data characteristic expansion processing on the extracted image information of the selected resource object data to obtain second expansion characteristic data of the selected resource object data;

inputting the sequence attribute feature of the selected resource object data, the first extended feature data, the second extended feature data, the sequence attribute feature of the unselected resource object data and the portrait information of the unselected resource object data into the data prediction model for training.

Optionally, the performing data feature expansion processing on the extracted portrait information of the selecting party to obtain first expansion feature data of the selecting party includes:

calculating the difference of two adjacent login durations of the selecting party described in the image information of the selecting party, and performing data feature expansion processing in a square or logarithm mode of the difference to obtain first expansion feature data of the selecting party;

the performing data feature expansion processing on the extracted image information of the selected resource object data to obtain second expansion feature data of the selected resource object data includes:

and performing difference calculation on two adjacent preferential data describing the resource object data in the portrait information of the selected resource object data, and performing data feature expansion processing in a manner of square or logarithm of the difference to obtain second expansion feature data of the selected resource object data.

calculating the distance of the feature vectors between the sample data;

the obtaining the correlation between the selected resource object data in the sample data according to the training result includes:

according to the distance, determining the correlation degree between the selected resource object data in the sample data, and determining the correlation degree between the unselected resource object data in the sample data and the selected resource object data in the sample data;

the determining, according to the correlation, a probability of being selected jointly between the resource object data already selected in the sample data includes:

and determining the probability of common selection among the selected resource object data in the sample data according to the correlation among the selected resource object data in the sample data and the correlation among the unselected resource object data in the sample data and the selected resource object data in the sample data.

Optionally, the acquiring sample data including the selected resource object data includes:

collecting selected first resource object data, and collecting selected second resource object data meeting the selection time requirement according to the selection time of the first resource object data;

and taking the first resource object data and the second resource object data as sample data.

and determining whether the quantity of the collected selected resource object data meets the requirement of the selected quantity, if so, determining the selected resource object data as sample data.

Optionally, when the quantity of the selected resource object data determined to be collected does not meet the requirement of the selected quantity, the resource object data with the characteristic vector of zero is supplemented to the selected resource object data, and the supplemented resource object data meeting the requirement of the selected quantity is determined as sample data.

collecting a set of the resource object data which meets the requirement of selecting the number of sets and comprises the selected resource object data;

determining the set of resource object data as sample data.

Optionally, the acquiring sample data including unselected resource object data includes:

and collecting sample data of the unselected resource object data within the sample data selection range.

and sequencing the collected selected resource object data according to the selection time, and determining the sequenced resource object data as sample data.

The present application further provides a training apparatus for a data prediction model, comprising:

the sample data acquisition unit is used for acquiring sample data comprising the selected resource object data and the unselected resource object data;

the training unit is used for inputting the sample data into a data prediction model for training;

the relevancy obtaining unit is used for obtaining the relevancy between the selected resource object data in the sample data according to the training result;

and the probability determining unit is used for determining the probability of being jointly selected among the resource object data according to the correlation.

The present application provides a computer storage medium for storing a program;

the program, when read and executed, is capable of performing the resource object recommendation method as described above; or to perform the training method of the data prediction model as described above.

The present application further provides an electronic device, comprising:

a processor;

a memory for storing a program capable of executing the resource object recommendation method as described above when read and executed by the processor; or to perform the training method of the data prediction model as described above.

Compared with the prior art, the method has the following advantages:

the application provides a resource object recommendation method, which comprises the steps of inputting acquired selected resource object data and target resource object data into a data prediction model, acquiring the probability that the target resource object data can be added into the selected resource object data, and determining the target resource object data with the probability meeting recommendation requirements as recommended resource object data; the mining of the recommended resource object data can be carried out on the basis of the determined and selected resource object data, so that a refined basis is provided for data mining, the accuracy of the subsequent recommended resource object data is improved, and the selected resource object data is more suitable for the requirements of a selecting party. Under a specific commodity recommendation application scene, the recommendation method can avoid the problems of recommendation weakness and cold start of new commodities caused by low commodity sales rate, and the problems of low recommendation data conversion rate, space waste and resource information waste.

According to the training method of the data prediction model, sample data which comprises selected resource object data and unselected resource object data is input into the data prediction model for training, so that the correlation degree between the selected resource object data in the sample data can be obtained according to a training result, and the probability of common selection between the resource object data is determined according to the correlation degree. Similarly, because the sample data is derived from the resource object data selected by the user, the model training can be performed on the basis of the more refined requirement of the user, and the accuracy of the output data of the model training can be improved.

Drawings

FIG. 1 is a flowchart of an embodiment of a resource object recommendation method provided in the present application;

FIG. 2 is a schematic structural diagram of an embodiment of a resource object recommendation device provided in the present application;

fig. 3 is a schematic view of a first application scenario of selected resource object data in an embodiment of a resource object recommendation method provided in the present application;

fig. 4 is a schematic view of a second application scenario of resource object data that has been selected in an embodiment of a resource object recommendation method provided in the present application;

FIG. 5 is a flow chart of an embodiment of a method for training a data prediction model provided herein;

FIG. 6 is a schematic diagram of an architecture of a data prediction model in an embodiment of a training method of the data prediction model provided in the present application;

fig. 7 is a schematic structural diagram of an embodiment of a training apparatus for a data prediction model provided in the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is capable of implementation in many different ways than those herein set forth and of similar import by those skilled in the art without departing from the spirit of this application and is therefore not limited to the specific implementations disclosed below.

The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The description used in this application and in the appended claims is for example: the terms "a," "an," "first," and "second," etc., are not intended to be limiting in number or order, but rather are used to distinguish one type of information from another.

Before explaining an embodiment of a resource object recommendation method provided by the present application, a technical application background of the present application is briefly explained, so that the embodiment of the present application can be explained later to make it easier to understand the technical solution provided by the present application. The resource object recommendation method is original in technology, and mainly based on the fact that when data recommendation is conducted in the prior art, the recommended data are not needed by a user or the data recommended to the user are not fine enough, and the recommended data are similar to nominal data. Particularly, when the amount of accumulated data is small, not only is the recommended data hindered, but also the accuracy of the recommended data is lowered. The industries for which the recommendation data is applied are e-commerce, entertainment information, etc., wherein the recommendation data is widely applied to the following business models, namely: O2O (online to offline consumption experience), C2C (consumer to consumer: person-to-person), B2C (business to consumer: person-to-person). In the embodiment provided by the present application, the business model of B2B is mainly used as an application scenario for description, but the resource object recommendation method provided by the present application is not limited to the application scenario, and here, the method is only a preliminary description of a specific application scenario of the embodiment of the present application, and a technical solution and the application scenario will be described in detail later in combination.

Based on the above, the present application provides a resource object recommendation method, please refer to fig. 1, where fig. 1 is a flowchart of an embodiment of the resource object recommendation method provided in the present application, and the recommendation method includes:

step S101: and acquiring the selected resource object data and the target resource object data.

The selected resource object data in step S101 may be understood as information about a selected commodity, that is, the resource object may be a commodity, and the resource object data may be information describing the commodity. The target resource object data can be understood as commodity information to be selected. The behavior that has been selected may be understood as the behavior that the item has been added to the purchase interface or that the purchase has been completed, such as: items that have been added to the shopping cart, but for which payment has not yet been completed; or goods that have been added to the shopping cart and have completed payment.

The specific implementation process of step S101 may be to obtain the selected resource object data through the historical data that the resource object selecting party has completed purchasing, or obtain the resource object data that the resource object selecting party has currently selected. Namely: acquiring the selected resource object data added to the selection list; or acquiring historical resource object data which completes the confirmation operation aiming at the selected resource object data added to the selection list.

The selection of the target resource object data can also be obtained from the historical data of the resource object selector. The target resource object data may be the same as or different from the resource object data that has been selected by the resource object selector, that is, the source of the target resource object data may be the data in the resource object data that has been selected by the resource object selector, or may be another source.

In this embodiment, the acquiring the selected resource object data includes:

and acquiring the resource object data meeting the acquisition requirement in the selected resource object data. The acquisition requirements may include at least one of time requirements and quantity requirements. The time requirement may be selected resource object data within 10 days of the selected resource object data from the recommended time, and the quantity requirement may be 15 selected resource object data from the recommended time.

The time requirement and the quantity requirement can be randomly modified according to the actual recommendation requirement, and are not limited to a required mode.

Step S102: inputting the selected resource object data and the target resource object data into a data prediction model, and obtaining the probability that the target resource object data can be added into the selected resource object data, wherein the data prediction model is used for predicting the probability that the target resource object data can be added into the selected resource object data.

The purpose of step S102 is to know, through a data prediction model, the probability that there is a common choice between the data acquired in step S101. For example, the resource object data that the user has selected includes: A. b, C, D, the target object data is E, then the purpose of step S102 is to predict the probability of whether E can be added to the selected A, B, C, D by the data prediction model.

The data prediction model in step S102 will be described in detail later, and will not be described in detail herein.

The specific implementation process of step S102 may include:

step S102-1: extracting sequence attribute features in the selected resource object data according to text information which is used for describing the content of the selected resource object in the selected resource object data;

step S102-2: according to text information used for describing the content of the target resource object data in the target resource object data, extracting the sequence attribute features in the target resource object data, wherein the sequence attribute features are the same as the sequence attribute features of the selected resource object data in feature type;

step S102-2: and inputting the sequence attribute characteristics of the selected resource object data and the sequence attribute characteristics of the target resource object data into the data prediction model to obtain the probability that the target resource object data can be added into the selected resource object data.

The feature types in step S102-1 and step S102-2 may include: the name characteristic type is used for describing the selected resource object data subject information, the price characteristic type is used for describing the selected resource object data payment information, and the specification characteristic type is used for describing the resource object data specification information. Then, in step S102-1, the keyword (or core word) related to the name in the subject information of the selected resource object data may be extracted, the price number related to the payment price in the payment information may be extracted, the specification number related to the size, the weight, and the like in the specification information may be extracted, and the keyword (or core word), the price, and the specification may be used as the sequence attribute feature representing the selected resource object data. Similarly, the target resource object data may be a sequence attribute feature that extracts the above-mentioned key word (or core word), price, and rule as data representing the target resource object.

In this embodiment, the extraction of the sequence attribute features may be obtained based on the sequence prediction algorithm of Bert. The input part of the Bert sequence prediction (Bidirectional Encoder prediction from transforms) is a linear sequence, and sentences are divided by separators. The segmented word has three vectors (embedding):

a position information vector, wherein the word sequence is an important characteristic and needs to be coded; a word vector; sentence vectors, data sentences, then each sentence has a sentence-wide vector term for each word. The three vectors are overlapped to form the input of the Bert, and the sequence attribute features needing to be extracted are output through the Bert.

It should be noted that the above sequence attribute feature in this embodiment includes: the core words, the prices and the specifications, that is, for the resource object data being data describing information of the goods, the sequence attribute characteristics representing the goods are the name, the price and the specifications of the goods. The commodity information can also be completely different for different application scenarios, for example: a raw material product showing a restaurant, a product showing a living article, a product showing music information, a product showing video information, and the like. In this embodiment, in general, the resource object data may be text information describing the product, and the sequence attribute feature is a result of extracting the text information. The information that the user pays more attention to in the selected resource object data is obtained by extracting the attributes in the resource object data, so that the basic reference information which is comprehensive and more suitable for the user' S own requirements is provided when the probability is predicted in the subsequent step S102-3.

In order to further improve the accuracy and precision of recommendation, the embodiment further includes:

and extracting portrait information used for describing the target resource object data in the target resource object data. The portrait information of the target resource object data is mainly used for describing attribute information of the target resource object, and compared with sequence attribute information, the portrait information is not extracted from text information of the target resource object data, but is extracted based on some auxiliary information of the target resource object, for example: target resource object identification information (ID information), belonging category, preference information, original price, and the like.

Likewise, the present embodiment may further include:

and extracting portrait information for describing the resource object selector. The resource object selector can be understood as a user, and the portrait information of the resource object selector is a portrait of the user, and the resource object selector comprises the following steps: geographic location, cell phone model, price sensitivity, store size, store type, forecast store type, customer unit price, order size, last time log-in time, etc.

When the prediction is performed by the data prediction model, the sequence attribute feature of the selected resource object data, the sequence attribute feature of the target resource object data, the image information of the target resource object, and the image information of the selecting party may be input to the data prediction model as input feature data.

In order to further expand the input feature data, the present embodiment may further perform feature expansion on the portrait information of the target resource object and the portrait information of the resource object selecting party, and further include:

and performing data characteristic expansion processing on the portrait information of the resource object selector to obtain second expansion characteristic data of the resource object selector.

In this embodiment, the first augmented characteristic data may be obtained as follows:

In this embodiment, the second augmented characteristic data may be obtained as follows:

Based on the above, the specific implementation process of step S102 may be:

inputting the extracted sequence attribute feature of the resource object data, the sequence attribute feature of the target resource object data, the portrait information of the target resource object data, the first extended feature data, the portrait information of the resource object selector and the second extended feature data into the data prediction model, and obtaining the probability that the target resource object data can be added into the selected resource object data.

It will be appreciated that the input to the data prediction model may be any of the above features, or may be a plurality of features.

Step S103: and determining the target resource object data with the probability meeting the recommendation requirement as recommended resource object data.

The purpose of step S103 is to obtain output data according to the feature data input to the data prediction model in step S102, where the output data represents the probability that the target resource object data can be added to the selected resource object data, and determine whether the target resource object data can be recommended as recommended resource data according to the probability.

And when the probability value of the target resource object data relative to the selected resource object data meets the recommendation requirement, determining the target resource object data as recommended resource object data. The recommendation requirement may be a probability value of 80% or more.

After determining the recommended resource object data, the recommended resource object data may be output to an interface of the selected resource object data, for example: and outputting the recommended resource object data to the tail end of the selection list, wherein the specific application scene can be that the recommended resource object data is output to a display interface of the shopping cart as the recommended resource.

It will be appreciated that weights may be set for the sequence attribute features to highlight important ones of the sequence attribute features.

The above is a specific description of an embodiment of the resource object recommendation method provided by the application, and it can be known from the above contents that when resource object data is recommended to a user, the resource object data depends on resource object data that has been selected by the user, and prediction is performed through sequence attribute features of the resource object data and sequence attribute features of a target resource object, and the sequence attribute features can more accurately and accurately dig out the user's demand, so that a resource object closer to the user's demand can be recommended to the user, the conversion rate of the recommended resource object is further improved, and resource waste and storage space waste are avoided.

The above is a specific description of an embodiment of a resource object recommendation method provided in the present application, and corresponds to the foregoing provided embodiment of a resource object recommendation method, and the present application also discloses an embodiment of a resource object recommendation apparatus, please refer to fig. 2, since the apparatus embodiment is basically similar to the method embodiment, the description is simpler, and related points can be referred to the partial description of the method embodiment. The device embodiments described below are merely illustrative.

As shown in fig. 2, fig. 2 is a schematic structural diagram of an embodiment of a resource object recommendation apparatus provided in the present application, where the apparatus includes:

an obtaining unit 201, configured to obtain the selected resource object data and the target resource object data.

The obtaining unit 201 is specifically configured to obtain the selected resource object data added to the selection list; or acquiring historical resource object data which completes the confirmation operation aiming at the selected resource object data added to the selection list.

The obtaining unit 201 may be specifically configured to obtain resource object data that meets an obtaining requirement in the selected resource object data, where the obtaining requirement may include at least one of a time requirement and a quantity requirement.

A predicting unit 202, configured to input the selected resource object data and the target resource object data into a data prediction model, and obtain a probability that the target resource object data can be added to the selected resource object data, where the data prediction model is used to predict a probability that the target resource object data can be added to the selected resource object data.

The prediction unit 202 includes: a first extraction subunit, a second extraction subunit, and a prediction subunit.

The first extraction subunit is configured to extract, according to text information used to describe the content of the selected resource object in the selected resource object data, the sequence attribute feature in the selected resource object data.

And the second extraction subunit is configured to extract, according to text information used to describe content of the target resource object data in the target resource object data, a sequence attribute feature in the target resource object data, which is of the same feature type as the sequence attribute feature of the selected resource object data.

The predicting subunit is configured to input the sequence attribute feature of the selected resource object data and the sequence attribute feature of the target resource object data into the data prediction model, so as to obtain a probability that the target resource object data can be added to the selected resource object data.

The feature types of the sequence data features of the selected resource object data comprise: the name characteristic type is used for describing the selected resource object data subject information, the price characteristic type is used for describing the selected resource object data payment information, and the specification characteristic type is used for describing the resource object data specification information.

Further comprising: and the portrait extracting subunit is used for extracting portrait information which is used for describing the target resource object data in the target resource object data.

The prediction subunit is specifically configured to input the sequence attribute feature in the first extraction subunit, the sequence attribute feature in the second extraction subunit, and the portrait information of the target resource object data in the portrait extraction subunit into the data prediction model, so as to obtain a probability that the target resource object data can be added to the selected resource object data.

The portrait extraction subunit includes: and the expansion processing subunit is used for performing data characteristic expansion processing on the image information of the target resource object data to obtain first expansion characteristic data of the target resource object data. The prediction subunit is specifically configured to input the sequence attribute feature in the first extraction subunit, the sequence attribute feature in the second extraction subunit, the portrait information of the target resource object data in the portrait extraction subunit, and the first extended feature data in the extended processing subunit into the data prediction model, so as to obtain a probability that the target resource object data can be added to the selected resource object data.

The portrait extraction subunit may be further configured to extract portrait information describing the resource object selector. The prediction subunit is specifically configured to input the sequence attribute feature of the first extraction subunit, the sequence attribute feature of the second extraction subunit, and the portrait information of the resource object selector extracted in the portrait extraction subunit into the data prediction model, so as to obtain a probability that the target resource object data can be added to the selected resource object data.

Similarly, the expansion processing subunit may be further configured to perform data feature expansion processing on the image information of the resource object selector to obtain second expansion feature data of the resource object selector.

The prediction subunit may be specifically configured to input, to the data prediction model, the sequence attribute feature of the first extraction subunit, the sequence attribute feature of the second extraction subunit, the portrait information of the resource object selector extracted in the portrait extraction subunit, and the second extended feature data in the extended processing subunit, so as to obtain a probability that the target resource object data can be added to the selected resource object data.

In this embodiment, the prediction subunit may input the sequence attribute feature of the first extraction subunit, the sequence attribute feature of the second extraction subunit, the portrait information of the resource object selector extracted in the portrait extraction subunit, and the portrait information of the target resource object data, and the first extended feature data and the second extended feature data in the extended processing subunit into the data prediction model, so as to obtain the probability that the target resource object data can be added to the selected resource object data. In other words, the input data content may be at least one of the first extended feature data and the second extended feature data in the extension processing subunit, by the prediction subunit, from among the sequence attribute feature of the first extraction subunit, the sequence attribute feature of the second extraction subunit, the image information of the resource object selector extracted in the image extraction subunit, and the image information of the target resource object data.

The expansion processing subunit includes: and the first calculating subunit is used for performing difference calculation on two adjacent preferential data describing the target resource object in the portrait information of the target resource object, performing data feature expansion processing in a manner of square or logarithm of the difference, and obtaining first expansion feature data of the target resource object.

And the second calculating subunit is used for calculating the difference of two adjacent login durations of the resource object selector described in the portrait information of the resource object selector, and performing data feature expansion processing in a manner of square or logarithm of the difference to obtain second expansion feature data of the resource object selector.

A determining unit 203, configured to determine the target resource object data with the probability meeting the recommendation requirement as recommended resource object data.

Further comprising: and the output unit is used for outputting the determined recommended resource object data to the interface of the selected resource object data for output. After determining the recommended resource object data, the recommended resource object data may be output to an interface of the selected resource object data, for example: and outputting the recommended resource object data to the tail end of the selection list, wherein the specific application scene can be that the recommended resource object data is output to a display interface of the shopping cart as the recommended resource.

The above is a specific description of an embodiment of a resource object recommendation method and device provided by the present application, and in order to better understand the above contents, a process description of a recommendation principle of the above recommendation method in combination with a specific application scenario is described below. Referring to fig. 3 and fig. 4, fig. 3 is a schematic view of a first application scenario of selected resource object data in an embodiment of a resource object recommendation method provided in the present application; fig. 4 is a schematic view of a second application scenario of resource object data that has been selected in an embodiment of a resource object recommendation method provided in the present application.

The selected resource object may be a commodity that has been added to the shopping cart on the interface of the shopping cart of the application service, as shown in fig. 3, or may be a commodity that has completed a payment operation or a commodity that has not completed a payment operation after confirming the payment based on the commodity that has been selected in the shopping cart, as shown in fig. 4.

In this embodiment, after the target product is determined as the recommended product, the target product is output to the area of the recommended product in the shopping cart, or the area of the recommended product after payment is confirmed based on the completion of the shopping cart, for example: the recommended commodity is output on the interface after the payment operation is completed, or the recommended commodity is output on the interface without the payment operation, but of course, the recommended commodity may also be output according to the analysis result and the reason for the payment operation not being completed, for example: the reason for the incomplete payment operation is selection omission, wrong selection and the like, and the recommended commodities can be output on the interface of the shopping cart so as to improve the possibility of selection of the recommended commodities.

The above is an explanation of the selected product, and the specific implementation principle and process of the recommendation method in this embodiment may be as follows: firstly, the selected commodity information and the target commodity information of the user A are obtained, the target commodity information can be from the commodity information selected by all users on the basis of an application service platform, and a prediction result is obtained more accurately, so that the image information of the user A and the image information of the target commodity can be added to the data characteristics of the selected commodity, namely: the commodity type, the user category, the preference information, the login time, the address information and the like, so that when all the acquired data characteristics are input into the data prediction model for prediction, a prediction probability result can be better obtained. For example: the selected commodity information is potatoes, eggs and salt, and the data features extracted aiming at the commodity information are potatoes; 100 jin; the total price is 50 yuan; fast food restaurants, etc., eggs; 50 jin; unit price: 29.5 membered; salt; 20 bags of the Chinese herbal medicine; monovalent 1 membered, etc. The target commodity can be eggplant, tomato, potato, bean sprout, chicken essence and the like, the extracted data features are the same as the extracted feature types of the selected commodity information, the extracted data features are input into a data prediction model for prediction, the prediction results are that the eggplant purchase probability is 80%, the tomato is 85%, the potato is 90%, the bean sprout is 40%, and the chicken essence is 80%, then the eggplant, the tomato, the potato and the chicken essence can be recommended as the recommended commodity, and the target commodity can comprise preferential information when being the potato, so that the probability that the commodity is added as the recommended commodity is further predicted. Of course, since the commodity specification information is taken into consideration, since the recommended commodity can be recommended according to the sizes of different merchants. Therefore, in this embodiment, the commodity information is mainly analyzed from the perspective of the merchant demand, the corresponding characteristic data is obtained for input, and then the recommended commodity information is obtained to be more accurately fit with the merchant demand. The application scenario is mainly explained by using a food material raw material purchasing application service platform.

The above is merely a general introduction to specific application scenarios in the embodiments of the present application, and with reference to the above contents, the present application also provides a training method of a data prediction model, please refer to fig. 5 and fig. 6, where fig. 5 is a flowchart of an embodiment of the training method of the data prediction model provided by the present application; fig. 6 is a schematic diagram of an architecture of a data prediction model in an embodiment of a training method of the data prediction model provided in the present application.

As shown in fig. 5, the embodiment of the training method includes:

step S501: sample data including selected resource object data and unselected resource object data is collected.

The specific implementation process of step S501 may include:

collecting the selected first resource object data, and collecting the data satisfying the selection according to the selection time of the first resource object dataTime-required second resource object data that has been selected; namely: the selected resource object data may be selected according to a selection time, for example: first resource object data are determined, and then second resource object data are selected according to the selection time of the first resource object data, namely: the selection time of the second resource object data should be a short distance from the selection time of the first resource object data. The number of selections may be set, for example, 10 selections are selected, and then the resource object data that has been selected is used as a set, which includes 10 resource object data information. Resource object data set (item)₁，item₂，...，item_m) And m commodities are obtained in total, so that m groups of samples can be obtained.

If the selected quantity does not meet the requirement, the resource object data with the characteristic vector of zero can be supplemented into the selected resource object data, and the supplemented resource object data meeting the requirement of the selected quantity is determined as sample data.

In order to avoid bias caused by the fact that the selected resource objects all come from a certain selecting party during subsequent training, when sample data is collected, the resource object data set meeting the requirement of the number of the selected set, including the selected resource object data, can be collected, for example, the latest 15 pieces of data of the selecting party (user) are collected, and the latest can be understood as the time when the user selects or purchases the data or the recommended time.

In this embodiment, the collected sample data may be further sorted, that is, the selected resource object data is sorted according to the selection time, and the sorted resource object data is determined as the sample data.

The acquiring sample data of unselected resource object data in step S501 may specifically include:

and collecting sample data of the unselected resource object data within the sample data selection range. The selection range can be determined according to the position information of the selecting party. In this embodiment, the sampling manner of the unselected resource object data in the sample data may adopt a word2vec sample generation manner.

Step S502: inputting the sample data into a data prediction model for training;

the step S502 may specifically include:

and extracting sequence attribute features in the selected resource object data according to text information which is used for describing the content of the selected resource object in the selected resource object data.

And extracting the sequence attribute features in the target resource object data, which have the same feature type as the sequence attribute features of the selected resource object data, according to text information which is used for describing the content of the target resource object data in the target resource object data.

The feature types of the sequence data features of the selected resource object data comprise: the name characteristic type is used for describing the subject information of the selected resource object data, the price characteristic type is used for describing the payment information of the selected resource object data, and the specification characteristic type is used for describing the specification information of the selected resource object data.

Extracting the sample data may further include:

and extracting portrait information in the sample data for describing the selected resource object data and extracting portrait information in the sample data for describing a selecting party of the selected resource object data. Of course, as in the above-mentioned preferred embodiment, the extracted portrait information may be at least one of the portrait information and the portrait information.

In order to improve the refinement of sample data, the extracted image information of the selected resource object data is subjected to data characteristic expansion processing to obtain first expansion characteristic data of the selecting party;

and performing data characteristic expansion processing on the extracted portrait information of the selecting party to obtain second expansion characteristic data of the selected resource object data.

Wherein, the data feature expansion processing is performed on the extracted portrait information of the selecting party to obtain the first expansion feature data of the selecting party, and the method comprises the following steps:

Inputting the sample data into a data prediction model for training, wherein the training comprises the following steps:

and calculating the distance of the characteristic vectors between the sample data, namely the distance of the characteristic vectors between the selected resource object data and the unselected resource object data. That is to say, the content input into the data prediction model is a feature vector of the input feature, which can be obtained by Item2vec algorithm, and the feature vector embedding is input into the data prediction model and is subjected to continuous convolution operation to obtain a training result, as shown in fig. 6.

It should be noted that the sequence attribute features may be obtained by using the same algorithm as in the resource object recommendation method embodiment, and details are not described here. The sequence attribute features are based on the resource object and the user, and the content which is more fit with the real meaning expression of the resource object and the user requirement is finely excavated, so that a foundation is laid for subsequently improving the prediction accuracy.

Step S503: and acquiring the correlation between the selected resource object data in the sample data according to the training result.

Based on the above steps S501 to S502, the specific implementation process of the step S503 may be:

and determining the probability of common selection among the selected resource object data in the sample according to the correlation among the selected resource object data in the sample and the correlation among the unselected resource object data in the sample data and the selected resource object data in the sample data.

Step S504: and determining the probability of being jointly selected among the resource object data according to the correlation.

The specific implementation process of step S504 is to output a probability value at an output side of the data prediction model, where the output side of the data prediction model may use a sigmoid function, softplus function, softmax function, and the like as output values, and in this embodiment, the sigmoid function is mainly used, and an expression of the sigmoid function is as follows:

the value range of the sigmoid function is between 0 and 1, the sigmoid function has good symmetry, event probability is output by adopting the sigmoid function, and the sigmoid function can be well applied to the embodiment.

The above is a detailed description of an embodiment of a training method for a data prediction model provided in the present application, and corresponds to the aforementioned embodiment of the training method for a data prediction model provided in the present application, and the present application also discloses an embodiment of a training apparatus for a data prediction model, please refer to fig. 7, since the apparatus embodiment is basically similar to the method embodiment, the description is relatively simple, and related points can be referred to partial description of the method embodiment. The device embodiments described below are merely illustrative.

As shown in fig. 7, fig. 7 is a schematic structural diagram of an embodiment of a training apparatus for a data prediction model provided in the present application, where the embodiment of the apparatus includes:

the sample data acquisition unit 701 is configured to acquire sample data including the selected resource object data and unselected resource object data.

The sample data collecting unit 701 may be specifically configured to collect selected first resource object data, and collect selected second resource object data that meets a selection time requirement according to a selection time of the first resource object data; and taking the first resource object data and the second resource object data as sample data.

The sample data acquisition unit 701 may be specifically configured to determine whether the number of the acquired selected resource object data meets a selected number requirement, and if so, determine the selected resource object data as sample data. And when the quantity of the acquired selected resource object data does not meet the selection quantity requirement, supplementing the resource object data with the characteristic vector of zero into the selected resource object data, and determining the supplemented resource object data meeting the selection quantity requirement as sample data.

The sample data collecting unit 701 may be specifically configured to collect a set of the resource object data that meets a requirement of selecting a number of sets, where the set of the resource object data includes the selected resource object data; determining the set of resource object data as sample data.

The sample data collecting unit 701 may specifically collect sample data of unselected resource object data within a sample data selection range.

The sample data acquisition unit 701 includes a sorting subunit, configured to sort the acquired selected resource object data according to the selection time, and determine the sorted resource object data as sample data.

It is understood that, the sorting subunit may also be configured to sort the collected unselected resource object data according to the selection time, and determine the sorted resource object data as sample data. In this embodiment, the resource object data to be selected is not sorted, because the present embodiment mainly considers the correlation between the resource object data already selected in the sample data, but does not exclude the manner of predicting the relationship between the resource object data already selected in the sample data by considering the correlation between the resource object data not already selected in the sample data, and therefore, the resource object data not already selected in the sample data may be sorted according to the application requirements.

A training unit 702, configured to input the sample data into a data prediction model for training;

the training unit 702 comprises: an extraction subunit and a training subunit.

The extracting subunit is configured to extract, according to text information used to describe the content of the selected resource object in the selected resource object data, a sequence attribute feature in the selected resource object data;

the extraction subunit is configured to extract, according to text information used to describe content of the target resource object data in the target resource object data, a sequence attribute feature in the target resource object data, which is of the same feature type as the sequence attribute feature of the selected resource object data;

and the training subunit is configured to input the sequence attribute feature of the selected resource object data, the sequence attribute feature of the target resource object data, and the sequence attribute feature of the unselected resource object data into the data prediction model for training.

The extracting subunit may further be configured to extract at least one of portrait information describing the selected resource object data in the collected sample data and portrait information describing a selecting party of the selected resource object data in the collected sample data, and of course, may also extract portrait information describing the unselected resource object data in the collected sample data.

The training subunit is specifically configured to input the sequence attribute feature of the selected resource object data, the portrait information of the selecting party, the sequence attribute feature of the unselected resource object data, and the portrait information of the unselected resource object data into the data prediction model for training.

The extraction subunit includes: the first expansion subunit is used for carrying out data characteristic expansion processing on the extracted image information of the selecting party to obtain first expansion characteristic data of the selecting party; and the second expansion subunit is used for performing data characteristic expansion processing on the extracted image information of the selected resource object data to obtain second expansion characteristic data of the selected resource object data.

The training subunit is specifically configured to input the sequence attribute feature of the selected resource object data, the first extended feature data, the second extended feature data, the sequence attribute feature of the unselected resource object data, and the portrait information of the unselected resource object data into the data prediction model for training.

The first expansion subunit is specifically configured to perform difference calculation on two adjacent login durations of the selector described in the image information of the selector, and perform data feature expansion processing in a manner of square or logarithm of the difference to obtain first expansion feature data of the selector.

The second expansion subunit is specifically configured to perform difference calculation on two adjacent preferential data describing the resource object data in the portrait information of the selected resource object data, perform data feature expansion processing in a manner of taking a square or a logarithm of the difference, and obtain second expansion feature data of the selected resource object data.

The training unit comprises a calculating subunit, which is specifically configured to calculate a distance between feature vectors of selected resource object data in the sample data and a distance between feature vectors of selected resource object data and unselected resource object data in the sample data.

A correlation obtaining unit 703, configured to obtain, according to a training result, a correlation between the selected resource object data in the sample data;

the relevancy obtaining unit 703 is specifically configured to determine, according to the distance calculated by the calculating subunit, the relevancy between the resource object data already selected in the sample data, and determine the relevancy between the resource object data not already selected in the sample data and the resource object data already selected in the sample data.

A probability determining unit 704, configured to determine a probability that the resource object data is jointly selected according to the correlation.

The probability determining unit 704 is specifically configured to determine the probability of being jointly selected among the selected resource object data in the sample according to the correlation between the selected resource object data in the sample data, and the correlation between the unselected resource object data in the sample data and the selected resource object data in the sample data.

The above is an overview of the training method of the data prediction model provided in the present application, and specific contents may be combined with the contents of the resource object recommendation method embodiment.

Based on the above, the present application further provides a computer storage medium for storing a program;

Based on the above, the present application further provides an electronic device, including:

a processor;

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

1. Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory computer readable media (transient media), such as modulated data signals and carrier waves.

2. As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Although the present application has been described with reference to the preferred embodiments, it is not intended to limit the present application, and those skilled in the art can make variations and modifications without departing from the spirit and scope of the present application, therefore, the scope of the present application should be determined by the claims that follow.

Claims

1. A resource object recommendation method is characterized by comprising the following steps:

2. The method of claim 1, wherein the inputting the selected resource object data and the target resource object data into a data prediction model to obtain a probability that the target resource object data can be added to the selected resource object data comprises:

3. The resource object recommendation method of claim 2, wherein the feature types of the sequence data features of the selected resource object data comprise: the name characteristic type is used for describing the selected resource object data subject information, the price characteristic type is used for describing the selected resource object data payment information, and the specification characteristic type is used for describing the resource object data specification information.

4. The resource object recommendation method of claim 2, further comprising:

5. A resource object recommendation apparatus, comprising:

6. A method for training a data prediction model, comprising:

collecting sample data comprising selected resource object data and unselected resource object data;

inputting the sample data into a data prediction model for training;

7. The method for training a data prediction model according to claim 6, wherein the inputting the sample data into the data prediction model for training comprises:

8. An apparatus for training a data prediction model, comprising:

9. A computer storage medium for storing a program;

the program, when read and executed, is capable of performing the resource object recommendation method of any one of claims 1 to 4; or performing a method of training a data prediction model according to any of claims 6 to 7.

10. An electronic device, comprising:

a processor;

a memory for storing a program which, when read and executed by the processor, is capable of performing the resource object recommendation method of any one of claims 1 to 4; or performing a method of training a data prediction model according to any of claims 6 to 7.