CN117951383A

CN117951383A - Information intelligent recommendation method, device, equipment and storage medium

Info

Publication number: CN117951383A
Application number: CN202410196419.2A
Authority: CN
Inventors: 亚静
Original assignee: Northking Information Technology Co ltd
Current assignee: Northking Information Technology Co ltd
Priority date: 2024-02-22
Filing date: 2024-02-22
Publication date: 2024-04-30

Abstract

The invention discloses an information intelligent recommendation method, device, equipment and storage medium. The method comprises the following steps: inputting the target feature data and the feature data to be recommended into a target matching model to obtain an output target matching result; recommending the main body information to be recommended corresponding to the second main body to the first target main body under the condition that the target matching result meets the preset matching condition; the target feature data comprises target main body features of a first target main body and second interaction main body features of a second interaction main body with a history interaction relation with the first target main body, the feature data to be recommended comprises main body features to be recommended of the second main body to be recommended and first interaction main body features of the first interaction main body with the history interaction relation with the second main body to be recommended, and the target matching model improves accuracy and recommendation effect of personalized information recommendation by fusing main body features, global interaction features and local interaction features of double views.

Description

Information intelligent recommendation method, device, equipment and storage medium

Technical Field

The present invention relates to the field of network information technologies, and in particular, to an information intelligent recommendation method, apparatus, device, and storage medium.

Background

With the development of network information technology, service contents provided by an online service system are more and more abundant, such as post recruitment, financial product operation, wedding market and the like, and the quantity of service information under each service content is more and more. The online service system brings convenience for information resources to users and also brings difficulty in information selection to users.

The information recommendation system aims at realizing personalized information filtering by analyzing user characteristics. Conventional information recommendation systems generally use methods mainly including collaborative filtering-based recommendation algorithms and content-based recommendation algorithms. The recommendation algorithm based on collaborative filtering recommends service information matched with the user clusters to the user main body by finding the user clusters with similar information selection preferences of the user main body, and the recommendation algorithm based on the content is focused on analyzing user portraits of the user main body and operation behaviors aiming at the service information to realize information recommendation.

The traditional information recommendation method mostly realizes information recommendation from a single angle of a user main body, but certain business contents relate to bidirectional selection between the user main body and a business main body to which the business information belongs, such as post recruitment. Therefore, the traditional information recommendation method is not suitable for the service scene selected in two directions, and the recommendation effect is poor.

Disclosure of Invention

The embodiment of the invention provides an intelligent information recommendation method, device, equipment and storage medium, which are used for solving the problem that the traditional information recommendation method is used for recommending information from a single main body angle and improving the accuracy and recommendation effect of personalized information recommendation.

According to one embodiment of the invention, an intelligent information recommendation method is provided, and comprises the following steps:

Acquiring target feature data corresponding to a first target main body and feature data to be recommended corresponding to a second main body to be recommended;

inputting the target feature data and the feature data to be recommended into a target matching model which is trained in advance, and obtaining an output target matching result between the first target main body and the second target main body to be recommended;

recommending the main body information to be recommended corresponding to the second main body to the first target main body under the condition that the target matching result meets a preset matching condition;

The target feature data comprises target main body features of the first target main body and second interaction main body features respectively corresponding to at least one second interaction main body with a history interaction relationship with the first target main body, and the feature data comprises main body features to be recommended of the second main body to be recommended and first interaction main body features respectively corresponding to at least one first interaction main body with a history interaction relationship with the second main body to be recommended.

According to another embodiment of the present invention, there is provided an information intelligent recommendation apparatus, including:

The target feature data acquisition module is used for acquiring target feature data corresponding to the first target main body and feature data to be recommended corresponding to the second main body to be recommended;

The target matching result determining module is used for inputting the target characteristic data and the characteristic data to be recommended into a target matching model which is trained in advance to obtain an output target matching result between the first target main body and the second target main body to be recommended;

The main body information recommendation module to be recommended is used for recommending main body information to be recommended corresponding to the second main body to the first target main body under the condition that the target matching result meets a preset matching condition;

According to another embodiment of the present invention, there is provided an electronic apparatus including:

At least one processor and a memory communicatively coupled to the at least one processor; the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the information intelligent recommendation method according to any embodiment of the present invention.

According to another embodiment of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to implement the intelligent information recommendation method according to any of the embodiments of the present invention when executed.

According to the technical scheme, the target feature data corresponding to the first target main body and the feature data to be recommended corresponding to the second target main body are input into the target matching model which is trained in advance to obtain the output target matching result, and under the condition that the target matching result meets the preset matching condition, the main body information to be recommended corresponding to the second target main body is recommended to the first target main body, wherein the target feature data comprises target main body features of the first target main body and second interaction main body features respectively corresponding to at least one second interaction main body which has a historical interaction relationship with the first target main body, the feature data to be recommended comprises main body features to be recommended of the second main body to be recommended and first interaction main body features respectively corresponding to at least one first interaction main body which has a historical interaction relationship with the second main body to be recommended, so that the target matching model has the capability of learning the main body features of double views, and the recommendation effect of the personalized information recommendation is improved.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the invention or to delineate the scope of the invention. Other features of the present invention will become apparent from the description that follows.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flowchart of an intelligent information recommendation method according to an embodiment of the present invention;

FIG. 2 is a flowchart of a specific example of an intelligent information recommendation method according to an embodiment of the present invention;

FIG. 3 is a flowchart of another intelligent information recommendation method according to an embodiment of the present invention;

FIG. 4 is a diagram of a model architecture of a target matching model according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an intelligent information recommendation device according to an embodiment of the present invention;

fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.

It should be noted that the terms "first," "second," "to be recommended," "target," "reference," "preset," and the like in the description and the claims of the present invention and the above drawings are used for distinguishing similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Fig. 1 is a flowchart of an intelligent information recommendation method according to an embodiment of the present invention, where the method may be implemented by an intelligent information recommendation device, and the intelligent information recommendation device may be implemented in hardware and/or software, and the intelligent information recommendation device may be configured in a terminal device. As shown in fig. 1, the method includes:

S110, acquiring target feature data corresponding to the first target main body and feature data to be recommended corresponding to the second main body to be recommended.

In this embodiment, the pair of subjects for information recommendation includes two subjects, one of which is a first target subject and the other of which is a second subject to be recommended. Specifically, the types of the two bodies in the body pair corresponding to each other may be different, for example, the types of the two bodies in the body pair corresponding to each other are respectively [ user recruitment post ], [ user financial service ] or [ customer service personnel ], and the like, and of course, the types of the bodies in the body pair corresponding to each other may be the same, for example, in a wedding scene, the types of the two bodies in the body pair corresponding to each other are both users.

The types of the two bodies in the body pair are not limited, and it is understood that any body pair suitable for bidirectional selection is within the scope of the present application.

In this embodiment, the target feature data includes target features of a first target subject and second interaction subject features corresponding to at least one second interaction subject having a history interaction relationship with the first target subject, and the feature data to be recommended includes a feature of a second subject to be recommended and a first interaction subject feature corresponding to at least one first interaction subject having a history interaction relationship with the second subject to be recommended.

Specifically, the second interaction subject is a subject having a history interaction relationship with the first target subject, and the subject type of the second interaction subject is the same as the subject type of the second subject to be recommended; the first interaction subject is a subject which has a history interaction relationship with the second subject to be recommended, and the subject type of the first interaction subject is the same as the subject type of the first target subject. Taking a post recruitment scene as an example, a first target main body is a user A, each second interaction main body comprises a recruitment post 1, a recruitment post 2 and a recruitment post 3, a second main body to be recommended is a recruitment post 4, and each first interaction main body comprises a user B, a user C and a user D.

Specifically, the target main body features are used for representing feature data corresponding to the first target main body, the main body features to be recommended are used for representing feature data corresponding to the second main body to be recommended, and the feature dimensions of the target main body features and the feature dimensions of the main body features to be recommended have correlation with the information recommendation scene.

Taking a post recruitment scenario as an example, assuming that a first target subject is a user a and a second subject to be recommended is a recruitment post 4, the target subject features include, but are not limited to, a user portrait of the user a, a desired post, an recruitment resume, and the like, and the subject features to be recommended include, but are not limited to, a company name, an office address, recruitment requirements, and the like. Taking a financial product operation scene as an example, assuming that a first target subject is a user a and a second subject to be recommended is a financial service a, target subject characteristics include, but are not limited to, user portraits, financial demand levels, financial expense plans and the like of the user a, and subject characteristics to be recommended include, but are not limited to, financial service identifications, financial service acquisition costs, financial service contents and the like.

Specifically, the second interaction main body feature is used for representing feature data corresponding to the second interaction main body, the feature dimension of the second interaction main body feature is identical to the feature dimension of the main body feature to be recommended, the first interaction main body feature is used for representing feature data corresponding to the first interaction main body, and the feature dimension of the first interaction main body feature is identical to the feature dimension of the target main body feature.

Illustratively, the target feature data is represented by (u _i,[v₁,…,v_x ]), the feature data to be recommended is represented by (v _j,[u₁,...,u_y ]), wherein u _i represents the target subject feature, [ v ₁,…,v_x ] represents x second interaction subject features, v _j represents the subject feature to be recommended, and [ u ₁,...,u_y ] represents y first interaction subject features.

In particular, the historical interaction relationship may be used to characterize the presence of at least one arbitrary interaction activity or the presence of at least one specified interaction activity within the historical time period for the first target subject and the second interaction subject.

In an alternative embodiment, acquiring target feature data corresponding to a first target subject includes: acquiring historical service data corresponding to a first target main body; the historical service data comprise historical interaction data respectively corresponding to a first target main body and at least one second preset main body, wherein the historical interaction data comprise historical interaction moments respectively corresponding to at least one interaction behavior; screening at least one second interaction subject from the second preset subjects according to the historical time period and the historical interaction data; and adding target main body characteristics corresponding to the first target main body and second interaction main body characteristics corresponding to each second interaction main body in the historical service data into the target characteristic data.

In this embodiment, the history service data includes at least one history interaction data, a target subject feature corresponding to the first target subject, and second preset subject features corresponding to the second preset subjects, respectively.

Specifically, the second preset body is used for representing a body which has at least one random interaction behavior or at least one appointed interaction behavior with the first target body in the historical time period, and the type of the body of the second preset body is the same as the type of the body of the second body to be recommended.

Specifically, the interaction types corresponding to each interaction behavior in the historical interaction data may be one or more. Taking a post recruitment scene as an example, the interaction types corresponding to each interaction behavior in the historical interaction data comprise at least one of click behavior, browsing behavior, consultation behavior, application behavior, admission behavior and the like, taking a financial product operation scene as an example, the interaction types corresponding to each interaction behavior in the historical interaction data comprise at least one of click behavior, browsing behavior, consultation behavior, order placing behavior and the like, the interaction types corresponding to each interaction behavior in the historical interaction data are not limited, and the interaction types corresponding to each interaction behavior in the historical interaction data can be specifically set in a self-defined mode according to actual requirements.

For example, the historical time period may be 7 days or 15 days, and is not limited herein, and may be specifically set in a custom manner according to actual requirements.

In an optional embodiment, screening at least one second interaction entity from the second preset entities according to the historical time period and the historical interaction data includes: and for each historical interaction data, if the historical interaction data contains interaction time in a historical time period, taking a second preset main body corresponding to the historical interaction data as a second interaction main body.

In another optional embodiment, screening at least one second interaction entity from the second preset entities according to the historical time period and the historical interaction data includes: and for each piece of history interaction data, if the history interaction data comprises at least one interaction moment in a history time period and the interaction behaviors respectively corresponding to the interaction moments comprise appointed interaction behaviors, taking a second preset main body corresponding to the history interaction data as a second interaction main body.

The specific interaction behavior may be a clicking behavior or a consultation behavior, which is not limited herein, and may be specifically set in a user-defined manner according to actual requirements.

Specifically, target subject features corresponding to the first target subject in the historical service data are added to the target feature data, and second preset subject features corresponding to each second interaction subject in the historical service data are added to the target feature data as second interaction subject features.

On the basis of the above embodiment, optionally, the method further includes: and executing data cleaning operation on the historical service data. Exemplary data cleansing operations include, but are not limited to, duplicate data removal, missing value processing, outlier processing, etc., and specific embodiments of the data cleansing operations are not limited herein, and may be specifically customized according to actual needs.

The implementation manner of obtaining the feature data to be recommended in this embodiment is the same as or similar to the above-mentioned target feature data obtaining manner, and this embodiment is not described herein again.

S120, inputting the target feature data and the feature data to be recommended into a target matching model which is trained in advance, and obtaining an output target matching result between the first target main body and the second target main body to be recommended.

Exemplary model architectures for the target matching model include, but are not limited to, residual network (ResNet), transfomer network, CNN network (Convolutional Neural Networks, convolutional neural network), FCN network (Fully Convolutional Networks, full convolutional neural network), DNN network (Deep Neural Networks, deep neural network), RNN network (Recurrent Neural Network, cyclic neural network), etc., and the model architecture for the target matching model is not limited herein, and may be specifically customized according to actual requirements.

In an alternative embodiment, the method further comprises: acquiring first training data corresponding to a first training subject and second training data corresponding to a second training subject; inputting the first training data and the second training data into an initial matching model which is not trained, and obtaining an output prediction matching result; determining a loss function according to the predicted matching result and standard matching results corresponding to the first training main body and the second training main body; and adjusting model parameters of the initial matching model according to the loss function until the loss function converges, and taking the initial matching model as a target matching model after training.

In this embodiment, the first training data includes first training features of the first training subject and second reference subject features corresponding to at least one second reference subject having a history interaction relationship with the first training subject, and the second training data includes second training features of the second training subject and first reference subject features corresponding to at least one first reference subject having a history interaction relationship with the second training subject.

Exemplary types of functions of the loss function include, but are not limited to, square loss function, logarithmic loss function, exponential loss function, mean square error loss function, logistic regression loss function, huber loss function, cross entropy loss function, kullback-Leibler divergence loss function, etc., and the types of functions of the loss function are not limited herein, and can be specifically set in a customized manner according to practical requirements.

S130, recommending the main body information to be recommended corresponding to the second main body to be recommended to the first target main body under the condition that the target matching result meets the preset matching condition.

For example, the representation of the target matching result may be whether the target matching result is matched, the matching probability or the matching grade, etc., specifically, the matching probability and the matching grade may be used to represent the possibility of generating the interaction behavior between the first target subject and the second subject to be recommended in the future, and the higher the matching probability or the matching grade is, the greater the possibility of generating the interaction behavior between the first target subject and the second subject to be recommended in the future is, whereas the lower the matching probability or the matching grade is, the lower the possibility of generating the interaction behavior between the first target subject and the second subject to be recommended in the future is. The representation form of the target matching result is not limited, and can be specifically set in a self-defined manner according to actual requirements.

Correspondingly, the preset matching condition is matching, a preset probability range or a preset grade range. The preset probability range may be [80%100% ], the preset level range may be [ level II V ], and the preset probability range or the preset level range is not limited herein, and may be specifically set in a user-defined manner according to actual requirements.

Specifically, the main information to be recommended is used for characterizing main information corresponding to the second main to be recommended, and exemplary main information to be recommended includes, but is not limited to, at least one of a link address, a main recommendation diagram, a main icon, main detail information and the like of the second main to be recommended, where the main information to be recommended is not limited, and can be specifically set in a self-defined manner according to actual requirements.

Fig. 2 is a flowchart of a specific example of an information intelligent recommendation method according to an embodiment of the present invention, and in particular, a terminal system for executing the information intelligent recommendation method includes a service system terminal and a data processing terminal, where the data processing terminal collects historical service data from the service system terminal, and the historical service data includes a plurality of historical interaction data and main body features corresponding to a plurality of main bodies respectively. The data processing terminal executes data cleaning operation on the collected historical service data, and constructs a characteristic data pair according to the historical service data after data cleaning, wherein the characteristic data pair comprises first training data corresponding to a first training main body and second training data corresponding to a second training main body, and training is carried out on an initial matching model which is not trained according to the characteristic data pair to obtain a target matching model which is trained.

The business system terminal sends a real-time main body pair to the data processing terminal, the real-time main body pair comprises a first target main body and a second main body to be recommended, the data processing terminal inputs target feature data corresponding to the first target main body and the feature data to be recommended corresponding to the second main body to be recommended into a target matching model to obtain an output target matching result, the target matching result is sent to the business system terminal, and the business processing terminal pushes main body information to be recommended corresponding to the second main body to be recommended to the first target main body under the condition that the target matching result meets preset matching conditions.

According to the technical scheme, the target feature data corresponding to the first target main body and the feature data to be recommended corresponding to the second target main body are input into a pre-trained target matching model to obtain an output target matching result, and under the condition that the target matching result meets a preset matching condition, the main body information to be recommended corresponding to the second target main body is recommended to the first target main body, wherein the target feature data comprises target main body features of the first target main body and second interaction main body features respectively corresponding to at least one second interaction main body with a historical interaction relation with the first target main body, the feature data to be recommended comprises main body features to be recommended of the second main body to be recommended and first interaction main body features respectively corresponding to at least one first interaction main body with a historical interaction relation with the second main body to be recommended, so that the problem that the information recommendation is carried out by a traditional information recommendation method from a single main body angle is solved, the target matching model has the capability of learning the main body features of double views, and the accuracy and recommendation effect of personalized information recommendation are improved.

Fig. 3 is a flowchart of another intelligent information recommendation method according to an embodiment of the present invention, where the "target matching model" in the above embodiment is further refined. As shown in fig. 3, the method includes:

S210, acquiring target feature data corresponding to the first target main body and feature data to be recommended corresponding to the second main body to be recommended.

In this embodiment, S210 corresponds to S110 shown in fig. 1 in the above embodiment, which is the same or similar, and is not described herein.

S220, inputting the target feature data and the feature data to be recommended into a target matching model which is trained in advance.

In this embodiment, the object matching model includes a feature encoding network, a global feature network, a local feature network, and a fusion network.

S230, respectively carrying out feature coding on the input target feature data and the feature data to be recommended through a feature coding network to obtain a target feature vector and a feature vector to be recommended.

Exemplary coding modes adopted by the feature coding network include, but are not limited to, one-hot coding, TF-IDF (Term Frequency-inverse document Frequency) coding, word Embeddings (word embedding) coding, word bag coding, N-gram coding, neural network coding, etc., where the coding modes adopted by the feature coding network are not limited, and can be specifically set in a customized manner according to practical requirements.

In an alternative embodiment, the feature encoding network includes a feature encoding module. The feature coding module is used for carrying out feature coding on the input target feature data to obtain a target feature vector, and carrying out feature coding on the input feature data to be recommended to obtain the feature vector to be recommended.

In another alternative embodiment, the feature encoding network includes a first feature encoding module and a second feature encoding module. The first feature coding module is used for carrying out feature coding on input target feature data to obtain a target feature vector; and the second feature coding module is used for carrying out feature coding on the input feature data to be recommended to obtain feature vectors to be recommended.

Illustratively, the target feature vector X ⁱ and the feature vector to be recommended X ^j satisfy the formula:

Xⁱ＝Encode(u_i,[v₁,...,v_x])

X^j＝Encode(v_j,[u₁,...,u_y])

S240, respectively carrying out global feature extraction on the target feature vector and the feature vector to be recommended output by the feature coding network through the global feature network to obtain the target global feature and the global feature to be recommended.

Specifically, the target global feature is used for representing global interaction features between the first target main body and each second interaction main body in a historical time period, and the global feature to be recommended is used for representing global interaction features between the second main body to be recommended and each first interaction main body in the historical time period.

In an alternative embodiment, the network architecture of the global feature network is an LSTM (Long Short Term Memory, long-short-term memory) network. Exemplary, target global featuresAnd global features to be recommended/>The formula is satisfied:

In an alternative embodiment, the global feature network includes a global feature module. The global feature module is used for carrying out global feature extraction on the target feature vector to obtain a target global feature, and carrying out global feature extraction on the feature vector to be recommended to obtain the global feature to be recommended.

In another alternative embodiment, the global feature network includes a first global feature module and a second global feature module. The first global feature module is used for carrying out global feature extraction on the target feature vector to obtain a target global feature; and the second global feature module is used for carrying out global feature extraction on the feature vector to be recommended to obtain global features to be recommended.

On the basis of the above embodiment, optionally, the target feature vector includes a target feature vector corresponding to the target feature and second interaction vectors corresponding to the second interaction features, the first global feature module includes at least two feature extraction modules connected in series, input data of a first feature extraction module in the first global feature module is the target feature vector, input data of each feature extraction module except the first feature extraction module in the first global feature module is the second interaction vector, and an output result of a last feature extraction module connected in series with the feature extraction module.

Specifically, the output result of the last feature extraction module in the first global feature module is the target global feature.

On the basis of the above embodiment, optionally, the feature vector to be recommended includes a feature vector to be recommended corresponding to the feature of the subject to be recommended and first interaction vectors corresponding to the features of the first interaction subjects, the second global feature module includes at least two feature extraction modules connected in series, input data of a first feature extraction module in the second global feature module is the feature vector to be recommended, and input data of each feature extraction module except the first feature extraction module in the second global feature module is the first interaction vector and an output result of a last feature extraction module connected in series with the feature extraction module.

Specifically, the output result of the last feature extraction module in the second global feature module is the global feature to be recommended.

S250, through a local feature network, the method is used for extracting local features of the target feature vector and the feature vector to be recommended output by the feature coding network respectively to obtain the target local feature and the local feature to be recommended.

Specifically, the target local features are used for representing interaction strengths of the first target main body and the second interaction main bodies respectively corresponding to each other in the historical time period, and the global features to be recommended are used for representing interaction strengths of the second main bodies to be recommended and the first interaction main bodies respectively corresponding to each other in the historical time period.

In an alternative embodiment, the network architecture of the local feature network is an Attention (Attention) network or a pointer network. Exemplary, target local featuresAnd local features to be recommended/>The formula is satisfied:

In an alternative embodiment, the local feature network includes a local feature module. The local feature module is used for extracting local features of the target feature vector to obtain target local features, and extracting the local features of the feature vector to be recommended to obtain the local features to be recommended.

In another alternative embodiment, the local feature network includes a first local feature module and a second local feature module. The first local feature module is used for extracting local features of the target feature vector to obtain target local features; and the second local feature module is used for extracting local features of the feature vector to be recommended to obtain the local features to be recommended.

S260, outputting a target matching result between the first target main body and the second main body to be recommended according to the target global feature output by the global feature network, the global feature to be recommended, the target local feature output by the local feature network and the local feature to be recommended through the fusion network.

The fusion manner adopted by the fusion network includes, but is not limited to, an early fusion algorithm or a late fusion algorithm, wherein the early fusion algorithm includes, but is not limited to, a concat fusion algorithm, a parallel fusion algorithm and the like, the late fusion algorithm includes, but is not limited to, an FPN (Feature Pyramid Network, multi-scale object detection) algorithm, an adaptive feature fusion algorithm and the like, the fusion manner adopted by the fusion network is not limited, and the fusion manner can be specifically set in a self-defined manner according to actual requirements.

In an alternative embodiment, the converged network includes a converged layer and an output layer. The fusion layer is used for carrying out feature fusion on the target global features, the global features to be recommended, the target local features and the local features to be recommended to obtain main body fusion features, and the output layer is used for outputting a target matching result between the first target main body and the second main body to be recommended according to the main body fusion features output by the fusion layer.

In another alternative embodiment, the converged network includes a converged layer and an output layer, wherein the converged layer includes a first converged module, a second converged module, and a third converged module; the first fusion module is used for carrying out feature fusion on the target global features and the target local features to obtain target fusion features; the second fusion module is used for carrying out feature fusion on the global features to be recommended and the local features to be recommended to obtain fusion features to be recommended; the third fusion module is used for carrying out feature fusion on the target fusion feature output by the first fusion module and the fusion feature to be recommended output by the second fusion module to obtain a main fusion feature; and the output layer is used for outputting a target matching result between the first target main body and the second main body to be recommended according to the main body fusion characteristics output by the third fusion module.

In an alternative embodiment, the first fusion module includes two linear layers and an attention layer, input data of the two linear layers are a target global feature and a target local feature respectively, input data of the attention layer is an output result of the two linear layers, and an output result of the attention layer is a target fusion feature; and/or the second fusion module comprises two linear layers and an attention layer, wherein the input data of the two linear layers are the global feature to be recommended and the local feature to be recommended respectively, the input data of the attention layer is the output result of the two linear layers, and the output result of the attention layer is the fusion feature to be recommended.

In an alternative embodiment, the third fusion module employs a dot product algorithm. Illustratively, the target fusion feature F ⁱ, the fusion feature to be recommended Fj, and the subject fusion feature S satisfy the formula:

S＝F.F

in an alternative embodiment, the output layer employs a sigmoid activation function.

Fig. 4 is a schematic diagram of a target matching model according to an embodiment of the present invention. Specifically, the target matching model comprises a feature coding network, a global feature network, a local feature network and a fusion network. The feature encoding network comprises a first feature encoding module and a second feature encoding module, wherein the first feature encoding module is used for performing feature encoding on input target feature data (u _i,[v₁,...,v_x) to obtain a target feature vector X ⁱ, and the second feature encoding module is used for performing feature encoding (v _j,[u₁,...,u_y) on the input feature data to be recommended to obtain a feature vector X ^j to be recommended. In the example shown in fig. 4, a wider solid rectangle represents the target subject vector in the target feature vector X ⁱ, a wider open rectangle represents the second interaction vector in the target feature vector X ⁱ, a narrower solid rectangle represents the subject vector to be recommended in the feature vector to be recommended X ^j, and a narrower open rectangle represents the first interaction vector in the feature vector to be recommended X ^j.

The global feature network comprises a first global feature module formed by X LSTM networks and a second global feature module formed by y LSTM networks, wherein the first global feature module is used for extracting global features of a target feature vector X ⁱ output by the first feature coding module and outputting target global featuresThe second global feature module is used for extracting global features of the feature vector X ^j to be recommended output by the second feature coding module and outputting global features to be recommended

Wherein the local feature network comprises two attention modules, one of which is used for extracting local features of the target feature vector X ⁱ output by the first feature coding module and outputting target local featuresThe other attention module is used for extracting local features of the feature vector X ^j to be recommended, which is output by the second feature coding module, and outputting the local features/>

The fusion network comprises a fusion layer and an output layer, wherein the fusion layer comprises a first fusion module, a second fusion module and a third fusion module, and the first fusion module and the second fusion module are both composed of two linear layers and an attention layer.

S270, recommending the main body information to be recommended corresponding to the second main body to be recommended to the first target main body under the condition that the target matching result meets the preset matching condition.

In this embodiment, S270 corresponds to S130 shown in fig. 1 in the above embodiment, which is the same or similar, and is not described herein.

According to the technical scheme, the target feature data and the feature data to be recommended are input into a pre-trained target matching model, the input target feature data and the feature data to be recommended are subjected to feature coding respectively through a feature coding network to obtain a target feature vector and a feature vector to be recommended, the target feature vector and the feature vector to be recommended output by the feature coding network are subjected to global feature extraction respectively through a global feature network, the target global feature and the global feature to be recommended are output, the target feature vector and the feature vector to be recommended output by the feature coding network are subjected to local feature extraction respectively through a local feature network, the target local feature and the local feature to be recommended are output through a fusion network, and the target matching result between a first target main body and a second target main body to be recommended is output according to the target global feature, the global feature to be recommended, the target local feature and the local feature to be recommended output by the global feature network.

In the technical scheme of the invention, the related processes of collecting, using, storing, sharing, transferring and the like of the personal information of the user accord with the regulations of related laws and regulations, the user needs to be informed and obtain the consent or the authorization of the user, and when the personal information of the user is applicable, the technical processes of de-identification and/or anonymization and/or encryption are performed on the personal information of the user.

The following is an embodiment of the information intelligent recommendation device provided by the embodiment of the present invention, where the device and the information intelligent recommendation method of the embodiment belong to the same inventive concept, and details that are not described in detail in the embodiment of the information intelligent recommendation device may refer to the content of the information intelligent recommendation method in the embodiment.

Fig. 5 is a schematic structural diagram of an intelligent information recommendation device according to an embodiment of the present invention. As shown in fig. 5, the apparatus includes: the target feature data acquisition module 310, the target matching result determination module 320 and the subject information recommendation to be recommended module 330.

The target feature data obtaining module 310 is configured to obtain target feature data corresponding to the first target subject and feature data to be recommended corresponding to the second subject to be recommended;

The target matching result determining module 320 is configured to input target feature data and feature data to be recommended into a target matching model that is trained in advance, so as to obtain an output target matching result between the first target subject and the second target subject to be recommended;

the main to be recommended information recommending module 330 is configured to recommend main to be recommended information corresponding to the second main to be recommended to the first target main when the target matching result meets a preset matching condition;

The target feature data comprises target main body features of a first target main body and second interaction main body features respectively corresponding to at least one second interaction main body with a history interaction relation with the first target main body, and the feature data to be recommended comprises main body features to be recommended of the second main body to be recommended and first interaction main body features respectively corresponding to at least one first interaction main body with a history interaction relation with the second main body to be recommended.

In an alternative embodiment, the target feature data acquisition module 310 is specifically configured to:

Acquiring historical service data corresponding to a first target main body; the historical service data comprise historical interaction data respectively corresponding to a first target main body and at least one second preset main body, wherein the historical interaction data comprise historical interaction moments respectively corresponding to at least one interaction behavior;

screening at least one second interaction subject from the second preset subjects according to the historical time period and the historical interaction data;

And adding target main body characteristics corresponding to the first target main body and second interaction main body characteristics corresponding to each second interaction main body in the historical service data into the target characteristic data.

In an alternative embodiment, the object matching model comprises a feature coding network, a global feature network, a local feature network and a fusion network;

Correspondingly, the target matching result determining module 320 is specifically configured to:

inputting the target feature data and the feature data to be recommended into a target matching model which is trained in advance;

Respectively carrying out feature coding on input target feature data and feature data to be recommended through a feature coding network to obtain a target feature vector and a feature vector to be recommended;

respectively extracting global features from a target feature vector and a feature vector to be recommended output by a feature coding network through a global feature network to obtain a target global feature and a global feature to be recommended;

the local feature network is used for extracting local features of the target feature vector and the feature vector to be recommended output by the feature coding network respectively to obtain target local features and the local features to be recommended;

and outputting a target matching result between the first target main body and the second main body to be recommended according to the target global feature output by the global feature network, the global feature to be recommended, the target local feature output by the local feature network and the local feature to be recommended through the fusion network.

In an alternative embodiment, the fusion network comprises a fusion layer and an output layer, wherein the fusion layer comprises a first fusion module, a second fusion module and a third fusion module;

The first fusion module is used for carrying out feature fusion on the target global features and the target local features to obtain target fusion features; the second fusion module is used for carrying out feature fusion on the global features to be recommended and the local features to be recommended to obtain fusion features to be recommended; the third fusion module is used for carrying out feature fusion on the target fusion feature output by the first fusion module and the fusion feature to be recommended output by the second fusion module to obtain a main fusion feature; and the output layer is used for outputting a target matching result between the first target main body and the second main body to be recommended according to the main body fusion characteristics output by the third fusion module.

In an alternative embodiment, the global feature network includes a first global feature module and a second global feature module, and/or the local feature network includes a first local feature module and a second local feature module;

The first global feature module is used for carrying out global feature extraction on the target feature vector to obtain a target global feature; the second global feature module is used for carrying out global feature extraction on the feature vector to be recommended to obtain global features to be recommended; the first local feature module is used for extracting local features of the target feature vector to obtain target local features; and the second local feature module is used for extracting local features of the feature vector to be recommended to obtain the local features to be recommended.

In an alternative embodiment, the target feature vector includes a target feature vector corresponding to the target feature and second interaction vectors corresponding to the second interaction features, the first global feature module includes at least two feature extraction modules connected in series, input data of a first feature extraction module in the first global feature module is the target feature vector, input data of each feature extraction module except the first feature extraction module in the first global feature module is the second interaction vector, and an output result of a last feature extraction module connected in series with the feature extraction module.

In an alternative embodiment, the apparatus further comprises:

the target matching model training module is used for acquiring first training data corresponding to the first training main body and second training data corresponding to the second training main body;

Inputting the first training data and the second training data into an initial matching model which is not trained, and obtaining an output prediction matching result;

determining a loss function according to the predicted matching result and standard matching results corresponding to the first training main body and the second training main body;

adjusting model parameters of the initial matching model according to the loss function until the loss function converges, and taking the initial matching model as a target matching model after training is completed;

the first training data comprises first training features of a first training subject and second reference subject features respectively corresponding to at least one second reference subject with a history interaction relationship with the first training subject, and the second training data comprises second training features of the second training subject and first reference subject features respectively corresponding to at least one first reference subject with a history interaction relationship with the second training subject.

The information intelligent recommendation device provided by the embodiment of the invention can execute the information intelligent recommendation method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

Fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, servers, blade servers, mainframes, and other appropriate computers. Electronic equipment may also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices (e.g., helmets, eyeglasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.

As shown in fig. 6, the electronic device 10 includes at least one processor 11, and a Memory such as a Read-Only Memory (ROM) 12, a random access Memory (Random Access Memory, RAM) 13, etc. communicatively connected to the at least one processor 11, wherein the Memory stores a computer program executable by the at least one processor 11, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the Read-Only Memory (ROM) 12 or the computer program loaded from the storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 may also be stored. The processor 11, the ROM 12 and the RAM 13 are connected to each other via a bus 14. An Input/Output (I/O) interface 15 is also connected to bus 14.

Various components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, etc.; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, an optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information or data with other devices via a computer network, such as the internet, and/or various telecommunications networks.

The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central Processing unit (Central Processing Unit, CPU), a graphics Processing unit (Graphics Processing Unit, GPU), various specialized artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DIGITAL SIGNAL Processing, DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as the intelligent information recommendation method provided by the above embodiments.

In some embodiments, the information intelligent recommendation method provided in the above embodiments may be implemented as a computer program, which is tangibly embodied on a computer readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more of the steps of the information intelligent recommendation method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the information intelligent recommendation method in any other suitable way (e.g., by means of firmware).

Various embodiments of the systems and techniques described herein above may be implemented in the following systems or combinations thereof: digital electronic circuitry, integrated circuitry, field programmable gate array (Field Programmable GATE ARRAY, FPGA), application SPECIFIC INTEGRATED Circuit (ASIC), application SPECIFIC STANDARD PARTS, ASSP, system On Chip (SOC), complex programmable logic device (Complex Programmable Logic Device, CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

The computer program for implementing the information intelligent recommendation method of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be implemented. The computer program 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 the present application, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer readable storage 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. Alternatively, the computer readable storage medium may be a machine readable storage medium. Examples of machine-readable storage media may include at least one wire-based electrical connection, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-Only Memory (ROM), an erasable programmable read-Only Memory (EPROM), a 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 a user, the systems and techniques described here can be implemented on a terminal device having: a display device (e.g., cathode-Ray Tube (CRT) or Liquid crystal display (Liquid CRYSTAL DISPLAY, LCD) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the terminal device. Other kinds of devices may also provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background 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., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, 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 network (Local Area Network, LAN), wide area network (Wide Area Network, WAN), blockchain network, and the internet.

The computing system may include clients and servers. The client and server are typically 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 can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical host and Virtual special server (Virtual PRIVATE SERVER, VPS) service are overcome.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution of the present invention are achieved, and the present invention is not limited herein.

The above embodiments do not limit the scope of the present invention. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.

Claims

1. An intelligent information recommendation method is characterized by comprising the following steps:

2. The method of claim 1, wherein the acquiring target feature data corresponding to the first target subject comprises:

Acquiring historical service data corresponding to the first target main body; the historical service data comprise historical interaction data respectively corresponding to the first target main body and at least one second preset main body, and the historical interaction data comprise historical interaction moments respectively corresponding to at least one interaction behavior;

and adding target main body characteristics corresponding to the first target main body and second interaction main body characteristics corresponding to the second interaction main bodies in the historical service data into target characteristic data.

3. The method according to claim 1, wherein the object matching model comprises a feature coding network, a global feature network, a local feature network and a fusion network;

Correspondingly, the inputting the target feature data and the feature data to be recommended into a target matching model which is trained in advance to obtain an output target matching result between the first target main body and the second target main body to be recommended includes:

Respectively carrying out feature coding on input target feature data and feature data to be recommended through the feature coding network to obtain a target feature vector and a feature vector to be recommended;

respectively extracting global features of a target feature vector and a feature vector to be recommended output by the feature coding network through the global feature network to obtain a target global feature and a global feature to be recommended;

and outputting a target matching result between the first target main body and the second main body to be recommended according to the target global feature, the global feature to be recommended, the target local feature and the local feature to be recommended which are output by the global feature network through the fusion network.

4. The method of claim 3, wherein the converged network comprises a converged layer and an output layer, the converged layer comprising a first converged module, a second converged module, and a third converged module;

The first fusion module is used for carrying out feature fusion on the target global feature and the target local feature to obtain a target fusion feature; the second fusion module is used for carrying out feature fusion on the global feature to be recommended and the local feature to be recommended to obtain fusion features to be recommended; the third fusion module is used for carrying out feature fusion on the target fusion feature output by the first fusion module and the fusion feature to be recommended output by the second fusion module to obtain a main fusion feature; and the output layer is used for outputting a target matching result between the first target main body and the second main body to be recommended according to the main body fusion characteristics output by the third fusion module.

5. A method according to claim 3, wherein the global feature network comprises a first global feature module and a second global feature module, and/or the local feature network comprises a first local feature module and a second local feature module;

The first global feature module is used for extracting global features of the target feature vector to obtain target global features; the second global feature module is used for extracting global features of the feature vector to be recommended to obtain global features to be recommended; the first local feature module is used for extracting local features of the target feature vector to obtain target local features; and the second local feature module is used for extracting the local features of the feature vector to be recommended to obtain the local features to be recommended.

6. The method according to claim 5, wherein the target feature vector includes a target feature vector corresponding to the target feature and a second interaction vector corresponding to each of the second interaction features, the first global feature module includes at least two feature extraction modules connected in series, input data of a first feature extraction module in the first global feature module is the target feature vector, and input data of each feature extraction module in the first global feature module except the first feature extraction module is an output result of the second interaction vector and a last feature extraction module connected in series with the feature extraction module.

7. The method according to any one of claims 1-6, further comprising:

Acquiring first training data corresponding to a first training subject and second training data corresponding to a second training subject;

Determining a loss function according to the prediction matching result and standard matching results corresponding to the first training main body and the second training main body;

The first training data comprises first training features of the first training main body and second reference main body features respectively corresponding to at least one second reference main body with a historical interaction relationship with the first training main body, and the second training data comprises second training features of the second training main body and first reference main body features respectively corresponding to at least one first reference main body with a historical interaction relationship with the second training main body.

8. An intelligent information recommendation device, which is characterized by comprising:

9. An electronic device, the electronic device comprising:

At least one processor and a memory communicatively coupled to the at least one processor; wherein the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the intelligent information recommendation method of any of claims 1-7.

10. A computer readable storage medium, characterized in that the computer readable storage medium stores computer instructions for causing a processor to implement the intelligent information recommendation method according to any one of claims 1-7 when executed.