CN114328995A - Content recommendation method, device, equipment and storage medium - Google Patents
Content recommendation method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114328995A CN114328995A CN202111591821.3A CN202111591821A CN114328995A CN 114328995 A CN114328995 A CN 114328995A CN 202111591821 A CN202111591821 A CN 202111591821A CN 114328995 A CN114328995 A CN 114328995A
- Authority
- CN
- China
- Prior art keywords
- resource
- media
- recommended
- account
- determining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000000605 extraction Methods 0.000 claims description 74
- 238000012549 training Methods 0.000 claims description 30
- 238000013528 artificial neural network Methods 0.000 claims description 20
- 230000000306 recurrent effect Effects 0.000 claims description 18
- 230000015654 memory Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 description 15
- 230000004927 fusion Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 239000013598 vector Substances 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 238000013500 data storage Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure relates to a content recommendation method, a content recommendation device and a storage medium, relates to the technical field of networks, and at least solves the problems that in the related art, the accuracy of recommended content is low, and the diversity of the recommended content cannot be improved. The method comprises the following steps: acquiring a plurality of media resources operated in a latest preset time period of an object account, and determining resource characteristics corresponding to each media resource; determining object characteristics of an object account according to resource characteristics corresponding to each media resource in a plurality of media resources; determining the matching degree between the object account and the media resource to be recommended according to the object characteristics of the object account to obtain a first matching degree result; and according to the first matching degree result, determining a first target media resource from the media resources to be recommended, and recommending the first target media resource to the object account.
Description
Technical Field
The present disclosure relates to the field of network technologies, and in particular, to a content recommendation method, apparatus, device, and storage medium.
Background
In the related art, when content information is recommended for an object account, a content to be recommended for the object account is generally determined according to a matching degree between object portrait information and content information of the object account. The object portrait information is usually determined based on data such as attribute information and hobbies and interests provided when the account id is registered in the object account. However, since the preference of the object account changes with time, if content recommendation is performed based on the object portrait information, on one hand, the accuracy of recommended content is lower and lower, and on the other hand, it is not beneficial to increase the diversity of the recommended content.
Disclosure of Invention
The present disclosure provides a content recommendation method, apparatus, device, and storage medium, to at least solve the problems of low accuracy of recommended content and inability to improve diversity of recommended content in the related art. The technical scheme of the disclosure is as follows:
according to a first aspect of the embodiments of the present disclosure, there is provided a content recommendation method, including: acquiring a plurality of media resources operated in a latest preset time period of an object account, and determining resource characteristics corresponding to each media resource; determining object characteristics of an object account according to resource characteristics corresponding to each media resource in a plurality of media resources; determining the matching degree between the object account and the media resource to be recommended according to the object characteristics of the object account to obtain a first matching degree result; and according to the first matching degree result, determining a first target media resource from the media resources to be recommended, and recommending the first target media resource to the object account.
In a possible implementation manner, determining an object feature of an object account according to a resource feature corresponding to each of a plurality of media resources includes: predicting a target resource characteristic corresponding to a next operated media resource after the target account operates the plurality of media resources based on the resource characteristic corresponding to each media resource and the operation time information of each media resource; and determining the target resource characteristics as the object characteristics of the object account.
In another possible implementation, determining the resource characteristics corresponding to each media resource includes: and determining the resource characteristics corresponding to each media resource based on at least one of the image information, the text information and the voice information of each media resource.
In another possible implementation manner, predicting, based on the resource characteristic corresponding to each media resource and the operation time information of each media resource, a target resource characteristic corresponding to a media resource that is operated next after the target account operates the plurality of media resources includes: and performing recursion processing on the resource characteristics corresponding to each media resource based on the pre-trained recurrent neural network and the operation time information of each media resource to obtain target resource characteristics.
In another possible implementation manner, when the media resource to be recommended includes a video resource to be recommended, determining a matching degree between the object account and the media resource to be recommended according to the object feature of the object account, and obtaining a first matching degree result, includes: acquiring resource characteristics corresponding to the video resources to be recommended based on image information, text information and voice information of the video resources to be recommended and a pre-trained characteristic extraction network; and determining the matching degree of the object account and the media resource to be recommended based on the object characteristics of the object account and the resource characteristics corresponding to the video resource to be recommended to obtain a first matching degree result.
In another possible embodiment, the method further comprises: determining a training sample for each of a plurality of video samples; the training samples comprise image information, text information and voice information of each video sample; and training the initial feature extraction network based on the image information, the text information and the voice information of each video sample to obtain a pre-trained feature extraction network.
In another possible implementation, the training samples further include a category identifier of each video sample, and the training of the initial feature extraction network based on the image information, the text information, and the speech information of each video sample to obtain a pre-trained feature extraction model includes: inputting the image information, the text information and the voice information of each video sample into an initial feature extraction network to obtain a prediction result of each video sample; determining a loss function value based on the prediction result of each video sample and the category identification of each video sample, and updating the parameters of the initial feature extraction network based on the loss function value; and iteratively executing the steps on the updated initial feature extraction network until the initial feature extraction network meets the model convergence condition, and determining the converged initial feature extraction network as a pre-trained feature extraction network.
In another possible implementation manner, in a case that a plurality of media resources operated within a last preset time period of the object account are not acquired, the method further includes: determining initial object characteristics of the object account according to the object portrait information of the object account; determining the matching degree between the object account and the media resource to be recommended based on the initial object characteristics of the object account to obtain a second matching degree result; and according to the second matching degree result, determining a second target media resource from the media resources to be recommended, and recommending the second target media resource to the object account.
In another possible implementation, determining the object characteristics of the object account according to the characteristics of each of the plurality of media resources includes: determining initial object characteristics of the object account based on the object image information of the object account; and determining the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources and the initial object characteristics of the object account.
According to a second aspect of the embodiments of the present disclosure, there is provided a content recommendation apparatus including: the acquisition module is configured to execute the operation of acquiring a plurality of media resources operated in the latest preset time period of the object account and determine the resource characteristics corresponding to each media resource; the determining module is configured to determine the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources; the determining module is further configured to determine the matching degree between the object account and the media resource to be recommended according to the object characteristics of the object account, and obtain a first matching degree result; and the recommending module is configured to determine a first target media resource from the media resources to be recommended according to the first matching degree result and recommend the first target media resource to the object account.
In a possible implementation, the determining module is specifically configured to perform: predicting a target resource characteristic corresponding to a next operated media resource after the target account operates the plurality of media resources based on the resource characteristic corresponding to each media resource and the operation time information of each media resource; and determining the target resource characteristics as the object characteristics of the object account.
In another possible implementation, the determining module is specifically configured to perform: and determining the resource characteristics corresponding to each media resource based on at least one of the image information, the text information and the voice information of each media resource.
In another possible implementation, the determining module is specifically configured to perform: and performing recursion processing on the resource characteristics corresponding to each media resource based on the pre-trained recurrent neural network and the operation time information of each media resource to obtain target resource characteristics.
In another possible embodiment, the recommendation module is specifically configured to perform: under the condition that the media resource to be recommended comprises the video resource to be recommended, acquiring a resource feature corresponding to the video resource to be recommended based on image information, text information and voice information of the video resource to be recommended and a pre-trained feature extraction network; and determining the matching degree of the object account and the media resource to be recommended based on the object characteristics of the object account and the resource characteristics corresponding to the video resource to be recommended to obtain a first matching degree result.
In another possible embodiment, the apparatus further includes a training module configured to perform: determining a training sample for each of a plurality of video samples; the training samples comprise image information, text information and voice information of each video sample; and training the feature extraction network based on the image information, the text information and the voice information of each video sample to obtain a pre-trained feature extraction network.
In another possible embodiment, the training samples further include a class identifier for each video sample, and the training module is specifically configured to perform: inputting the image information, the text information and the voice information of each video sample into an initial feature extraction network to obtain a prediction result of each video sample; determining a loss function value based on the prediction result of each video sample and the category identification of each video sample, and updating the parameters of the initial feature extraction network based on the loss function value; and iteratively executing the steps on the updated initial feature extraction network until the initial feature extraction network meets the model convergence condition, and determining the converged initial feature extraction network as a pre-trained feature extraction network.
In another possible embodiment, the determining module is further configured to perform: under the condition that a plurality of media resources operated in the latest preset time period of the object account are not acquired, determining initial object characteristics of the object account according to the object portrait information of the object account; determining the matching degree between the object account and the media resource to be recommended based on the initial object characteristics of the object account to obtain a second matching degree result; and according to the second matching degree result, determining a second target media resource from the media resources to be recommended, and recommending the second target media resource to the object account.
In another possible embodiment, the determining module is further configured to perform: determining initial object characteristics of the object account based on the object image information of the object account; and determining the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources and the initial object characteristics of the object account.
According to a third aspect of the embodiments of the present disclosure, there is provided an electronic apparatus including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the content recommendation method of the first aspect and any of its possible embodiments described above.
According to a fourth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, wherein instructions of the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the content recommendation method of any one of the above-mentioned first aspects and any one of its possible implementations.
According to a fifth aspect of embodiments of the present disclosure, there is provided a computer program product comprising computer instructions which, when run on an electronic device, cause the electronic device to perform the content recommendation method of the first aspect and any of its possible implementations.
The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects: the object characteristics of the object account are determined according to the characteristics of the media resources operated in the latest preset time period of the object account, on one hand, the characteristics of the media resources operated in the latest preset time period can well represent the interest of the object account, and on the other hand, the object characteristics of the object account are not constant but dynamically changed, so that the object characteristics can accurately express the recent interest of the object account, the accuracy of the recommended content pushed to the object account is improved, the recommended content can accord with the interest of the object account, and the use experience of the object account is improved. In addition, since the object feature is a dynamic feature that changes, it is possible to increase the diversity of recommended content each time recommended content of the object account is determined based on the dynamic object feature.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.
FIG. 1 is a flow diagram illustrating a method of content recommendation in accordance with an exemplary embodiment;
FIG. 2 is a flow diagram illustrating another method of content recommendation, according to an example embodiment;
FIG. 3 is a flow diagram illustrating a method of content recommendation, according to an example embodiment;
FIG. 4 is a schematic diagram illustrating pre-training of a feature extraction network in accordance with an illustrative embodiment;
FIG. 5 is a diagram illustrating one type of determining recommended content according to an exemplary embodiment;
FIG. 6 is a block diagram illustrating a content recommendation device according to an example embodiment;
FIG. 7 is a block diagram illustrating an electronic device in accordance with an example embodiment.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.
Before describing the information processing method provided by the present disclosure in detail, the application scenario and implementation environment related to the present disclosure are briefly described.
First, a brief description is given of an application scenario to which the present disclosure relates.
In the related art, when content information is recommended for an object account, a content to be recommended for the object account is generally determined according to a matching degree between object portrait information and content information of the object account. The object portrait information is usually determined based on data such as attribute information and hobbies and interests provided when the account id is registered in the object account. However, since the preference of the object account changes with time, if content recommendation is performed based on the object portrait information, on one hand, the accuracy of recommended content is lower and lower, and on the other hand, it is not beneficial to improve the diversity of the recommended content.
In view of the above problems, the present disclosure provides a content recommendation method, where an object feature of an object account is determined according to a feature of a media resource operated in a latest preset time period of the object account, on one hand, the feature of the media resource operated in the latest preset time period can well represent an interest of the object account, and on the other hand, the object feature of the object account is not constant but dynamically changed, so that the object feature can accurately express the interest of the object account, and thus accuracy of recommended content pushed to the object account is improved, the recommended content can conform to the interest of the object account, and experience of using the object account is improved. In addition, since the object feature is a dynamic feature that changes, it is possible to increase the diversity of recommended content each time recommended content of the object account is determined based on the dynamic object feature.
Next, the following briefly describes an implementation environment (implementation architecture) related to the present disclosure.
The content recommendation method provided by the embodiment of the disclosure can be applied to electronic equipment. The electronic device may be a terminal device or a server. The terminal device can be a smart phone, a tablet computer, a palm computer, a vehicle-mounted terminal, a desktop computer, a notebook computer and the like. The server may be any one server or server cluster, and the disclosure is not limited thereto.
In addition, the object information (including, but not limited to, object device information, object personal information, and the like) according to the present disclosure is information that is authorized by the object or sufficiently authorized by each party.
For the sake of understanding, the task processing method provided by the present disclosure is specifically described below with reference to the accompanying drawings.
Fig. 1 is a flow chart illustrating a content recommendation method for an electronic device according to an exemplary embodiment. As shown in fig. 1, the content recommendation method includes the steps of:
s101: the method comprises the steps of obtaining a plurality of media resources operated in a recent preset time period of an object account, and determining resource characteristics corresponding to each media resource.
Optionally, the operations include one or more of likes, forwards, concerns about authors, and views. For example, the author who approves the media asset 1, forwards the media asset 1, views the media asset 1, or pays attention to the publishing of the media asset 1, etc. are all operating the media asset 1.
Optionally, the media assets include image assets, video assets, text assets, and the like. The present disclosure is not limited to the presentation of media assets.
Optionally, the resource feature corresponding to the media resource includes a feature vector of the media resource.
In one embodiment, acquiring a plurality of media resources of the object account operation within the latest preset time period may be performed by the electronic device. For example, the electronic device obtains a plurality of media resources operated by the subject account at preset time intervals (e.g., every other week). Wherein, the last preset time period may be within 30 days of the last acquisition time. After the electronic device obtains the plurality of media resources, the feature extraction network may determine resource features corresponding to each media resource.
In another embodiment, the plurality of media resources that have acquired the object account operation within the last preset time period may be acquired by a pre-trained recurrent neural network. For example, the pre-trained recurrent neural network acquires a plurality of media assets operated by the subject account at a preset time interval (e.g., every other week). Wherein, the last preset time period may be within 30 days of the last acquisition time. After obtaining a plurality of media resources, the pre-trained recurrent neural network determines the resource characteristics corresponding to each media resource.
S102: and determining the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources.
In one embodiment, based on the characteristics of each media resource, the resource characteristics corresponding to the media resource operated next after the plurality of media resources are operated by the target account are predicted, and the resource characteristics corresponding to the next operated media resource are determined as the target characteristics.
In another embodiment, feature fusion is performed on resource features corresponding to each of a plurality of media resources, and the obtained fusion features are determined as object features. Among them, regarding feature fusion, Early fusion (Early fusion) or Late fusion (Late fusion) may be used, which is not limited by the present disclosure.
S103: and determining the matching degree between the object account and the media resource to be recommended according to the object characteristics of the object account to obtain a first matching degree result.
Optionally, the resource features corresponding to the media resources to be recommended are determined based on the pre-trained feature extraction network.
In one embodiment, the resource features corresponding to the media resources to be recommended include feature vectors of the media resources to be recommended.
Optionally, the first matching degree result comprises a similarity value.
In one embodiment, the electronic device calculates a similarity value of the object features of the object account and the resource features corresponding to the media resources to be recommended, and determines the similarity value as a first matching result. Wherein the similarity value may be obtained by a similarity algorithm, for example, a cosine similarity algorithm. It will be appreciated that a higher similarity value for two features indicates a higher degree of matching between the two features.
S104: and according to the first matching degree result, determining a first target media resource from the media resources to be recommended, and recommending the first target media resource to the object account.
Optionally, recommending the first target media resource to the object account includes: and sending the first target media resource to the client side where the object account is located so that the client side where the object account is located can display the first target media resource.
In one embodiment, when the first matching degree result of the media resource to be recommended and the object account is greater than or equal to a preset threshold value, the media resource to be recommended is determined as a first target media resource.
In another embodiment, the electronic device ranks the to-be-recommended media resources according to the first matching degree result from large to small to obtain a sequence, and then determines the pre-set number of the to-be-recommended media resources of the sequence as the first target media resources. Of course, the sequence may also sort the to-be-recommended media resources from small to large according to the first matching degree result, and determine the to-be-recommended media resources of a preset number after the sequence as the first target media resources.
In the above embodiment, the object characteristics of the object account are determined according to the resource characteristics corresponding to the media resources operated in the latest preset time period of the object account, on one hand, the resource characteristics corresponding to the media resources operated in the latest preset time period can well represent the interest of the object account, and on the other hand, the object characteristics of the object account are not constant but dynamically changed, so that the object characteristics can accurately express the recent interest of the object account, the accuracy of the recommended content (i.e., the first target media resource) pushed to the object account is improved, the recommended content can conform to the interest of the object account, and the use experience of the object account is improved. In addition, since the object feature is a dynamic feature that changes, it is possible to increase the diversity of recommended content each time recommended content of the object account is determined based on the dynamic object feature.
In one possible implementation, referring to fig. 1 and as shown in fig. 2, S102 includes the following steps:
S102A: and predicting the target resource characteristics corresponding to the media resources operated next after the target account operates the plurality of media resources based on the resource characteristics corresponding to each media resource and the operation time information of each media resource.
In one embodiment, the resource characteristics corresponding to each media resource are aggregated based on the operation time information of each media resource, and the target resource characteristics corresponding to the media resource operated next after the plurality of media resources are operated by the target account are predicted.
Optionally, determining the resource characteristics corresponding to each media resource includes: and determining the resource characteristics corresponding to each media resource based on at least one of the image information, the text information and the voice information of each media resource. The resource characteristics corresponding to each media resource are determined based on at least one of the image information, the text information and the voice information of each media resource, so that the resource characteristics corresponding to each media resource are representative, each media resource can be expressed more accurately, the object characteristics can be finally obtained, and the interest of the object account can be expressed more accurately.
Optionally, S102A includes: and performing recursion processing on the resource characteristics corresponding to each media resource based on the pre-trained recurrent neural network and the operation time information of each media resource to obtain target resource characteristics.
Optionally, the recurrent neural network comprises a Long Short-Term Memory (LSTM) network.
It should be noted that the recursive processing referred to in the present disclosure refers to a neural network implementation-based recursion, such as an LSTM network implementation. A number of loop units may be included in the LSTM network, each for implementing a one-step recursive process.
In one embodiment, the pre-trained recurrent neural network obtains, at preset time intervals (e.g., every other week), operation time information of each of a plurality of media resources operated by the subject account within a last preset time period and a plurality of media resources. Then, the pre-trained recurrent neural network determines the resource characteristics corresponding to each media resource, and performs recurrent processing on the resource characteristics corresponding to each media resource based on the operation time information of each media resource, so as to predict the target resource characteristics corresponding to the next operated media resource after the object account operates a plurality of media resources.
It should be noted that the Recurrent Neural network may also be other Neural Networks capable of recursively processing the feature vectors based on the time information, for example, a Recurrent Neural Network (RNN). The present disclosure is not so limited.
In one embodiment, the content recommendation method further includes: and training the recurrent neural network according to the training sample to obtain a pre-trained recurrent neural network. The training sample includes a plurality of media asset samples and an operating time for each media asset sample. Further, the training samples also include historical result samples, and the historical result samples are media resources of the next operation of the object account after the plurality of media resource samples are operated.
In the embodiment, the target resource characteristics corresponding to the media resources operated next after the plurality of media resources are operated by the object account are predicted through the pre-trained recurrent neural network, so that the obtained object characteristics are more robust, simple, convenient and high in efficiency.
S102B: and determining the target resource characteristics as the object characteristics of the object account.
In the embodiment, the target resource characteristics corresponding to the media resources operated next after the target account operates the plurality of media resources are predicted, and the target resource characteristics are determined as the target characteristics of the target account, so that the target characteristics can accurately express the preference of the target account, and therefore, based on the target characteristics of the target account, the matching degree of the content to be operated of the target account is higher for the first target media resource determined from the media resources to be recommended, the accuracy of the recommended content is further improved, and the use experience of the target account is improved. For example, predicting that the target resource feature represents a gourmet video, determining the target resource feature as an object feature, determining the matching degree of the object feature and the resource feature corresponding to the media resource to be recommended, and obtaining a first matching degree result.
In a possible implementation manner, in the case that the media resource to be recommended includes a video resource to be recommended, as shown in fig. 3 in conjunction with fig. 1, S103 includes the following steps:
S103A: and obtaining resource characteristics corresponding to the video resources to be recommended based on the image information, the text information and the voice information of the video resources to be recommended and the pre-trained characteristic extraction network.
In one embodiment, the video resource to be recommended is obtained, for example, the video resource to be recommended may be obtained from a content pool of a database. And then, extracting video frames of the video resource to be recommended, acquiring video frames with preset frame numbers, and determining the video frames as image information of the video resource to be recommended. Further, text information of the video resource to be recommended is obtained, for example, text information in the video resource to be recommended is identified through OCR character recognition software, and the text information includes subtitle information and/or bullet screen information in the video resource to be recommended. Further, voice information of the video resource to be recommended is acquired, for example, the voice information in the video resource to be recommended is identified through voice recognition software.
And further, inputting image information, text information and voice information of the video resource to be recommended into the pre-trained feature extraction network to obtain the resource features corresponding to the video resource to be recommended, which are output by the pre-trained feature extraction network. For example, the pre-trained feature extraction network may be a feature extraction network in a pre-trained category identification model, and the pre-trained category identification model may predict a category of the video resource to be recommended based on features extracted by the feature extraction network.
Optionally, the content recommendation method further includes:
the method comprises the following steps: a training sample is determined for each of a plurality of video samples. The training samples include image information, text information, and voice information for each video sample.
In one embodiment, a plurality of video samples are obtained, for example, from a pool of contents of a database. And aiming at each video sample, acquiring image information, text information and voice information in the video sample to obtain a training sample of each video sample. The method for acquiring the image information, the text information and the voice information may be the method described above, and is not described herein again.
In one embodiment, the voice information may be presented in textual form, for example, stored in a textual format.
Step two: and training the initial feature extraction network based on the image information, the text information and the voice information of each video sample to obtain a pre-trained feature extraction network.
Optionally, the initial feature extraction network comprises a residual network, e.g., a Resnet50 network.
In one embodiment, the image information, the text information and the voice information of each video sample are input into an initial feature extraction network, and the initial feature extraction network is trained, so that a pre-trained feature extraction network is obtained.
In another embodiment, the image information, the text information and the voice information of each video sample are input into an initial class recognition model, and the initial class recognition model is trained to predict the class of each video sample based on the image information, the text information and the voice information of each video sample, so as to obtain a pre-trained class recognition model.
It should be noted that the pre-trained category identification model at least includes a feature extraction network and a category identification network, the feature extraction network is configured to determine a feature of each video sample based on the image information, the text information and the voice information of each video sample, and the category identification network is configured to predict a category of each video sample based on the feature of each video sample. The feature extraction network of the pre-trained category identification model may be used as the pre-trained feature extraction network in the present disclosure. That is to say, the resource features corresponding to the media resources to be recommended are output by using the feature extraction network in the pre-trained category identification model.
In the embodiment, the feature extraction network is trained through multi-mode information, so that the feature extraction capability of the pre-trained feature extraction network is improved, the extracted features of the media resources to be recommended are more robust and representative, and the content of the media resources to be recommended can be better expressed.
Optionally, the training samples further include a category identifier of each video sample, and the initial feature extraction network is trained based on image information, text information, and speech information of each video sample to obtain a pre-trained feature extraction network, including the following steps:
the method comprises the following steps: and inputting the image information, the text information and the voice information of each video sample into an initial feature extraction network to obtain a prediction result of each video sample.
As shown in fig. 4, the image information, text information, and voice information of each video sample are input into an initial feature extraction network, for example, a Resnet50 network, to obtain a prediction result of each video sample.
Step two: and determining a loss function value based on the prediction result of each video sample and the class identification of each video sample, and updating the parameters of the feature extraction network.
In one embodiment, a loss function value, e.g., a cross entropy loss value, is calculated based on the prediction result of each video sample and the class identifier of each video sample, and a parameter of the initial feature extraction network, e.g., a learning rate, is updated based on the cross entropy loss value, resulting in an updated initial feature extraction network.
Step three: and iteratively executing the first step and the second step on the updated initial feature extraction network until the initial feature extraction network meets the model convergence condition, and determining the converged initial feature extraction network as a pre-trained feature extraction network.
In one embodiment, the image information, the text information and the voice information of each video sample in the step one are input into the updated initial feature extraction network, a loss function value is determined based on the obtained prediction result of each video sample and the category identification of each video sample, and the parameters of the initial feature extraction network are updated again.
In one embodiment, the model convergence condition includes that the initial feature extraction network is iterated for a preset number of times, or a difference between loss function values obtained from two adjacent iterations is smaller than or equal to a preset threshold.
In one embodiment, an SGD optimization algorithm may be employed to make the initial feature extraction network converge.
In the above embodiment, the deep neural network is trained by multimodal data, such as image information, text information, and voice information, to obtain a pre-trained feature extraction network capable of determining features based on the multimodal data.
S103B: and determining the matching degree of the object account and the media resource to be recommended based on the object characteristics of the object account and the resource characteristics corresponding to the video resource to be recommended to obtain a first matching degree result.
Optionally, according to the similarity value of the object features and the resource features corresponding to the media resources to be recommended, the matching degree of the object account and the features of the media resources to be recommended is determined, and a first matching degree result is obtained. It will be appreciated that a higher similarity value for two features indicates a higher degree of matching between the two features.
For example, as shown in fig. 5, an object feature, for example, an LSTM network, is determined based on a pre-trained recurrent neural network, and then a matching degree between the object feature and a resource feature corresponding to a media resource to be recommended is calculated, so as to obtain a first matching degree result. The resource characteristics corresponding to the media resource 1, … …, and the resource characteristics corresponding to the media resource n are the resource characteristics corresponding to a plurality of media resources operated in the latest preset time period of the object account, and n is a positive integer greater than 1.
In the embodiment, the resource characteristics corresponding to the video resources to be recommended are determined through the information of three modalities, namely the image information, the text information and the voice information, so that the resource characteristics corresponding to the video resources to be recommended are representative, the content of the video resources to be recommended can be expressed more accurately, the matching degree between the determined object account and the media resources to be recommended is higher based on the resource characteristics and the object characteristics corresponding to the video resources to be recommended, and the accuracy is higher and better accords with the interests and hobbies of the object account.
Optionally, in a case that a plurality of media resources operated within a last preset time period of the object account are not acquired, the content recommendation method further includes:
the method comprises the following steps: and determining initial object characteristics of the object account according to the object portrait information of the object account.
It can be understood that, when the object account of the application program is used for the first time, since no media resource is operated on the application program, the electronic device cannot acquire the media resources operated by the object account within the latest preset time period.
In one embodiment, for an object account which does not operate any media resource, the electronic equipment acquires attribute information and interest information of the object account, wherein the attribute information comprises information such as age, gender and academic calendar so as to construct object portrait information of the object account, and determines initial object characteristics of the object account according to the object portrait information. For example, the object image information is input to a feature extraction model, and feature extraction is performed to obtain an initial object feature of the object account.
Step two: and determining the matching degree between the object account and the media resource to be recommended based on the initial object characteristics of the object account to obtain a second matching degree result.
Optionally, the second matching degree result comprises a similarity value.
In one embodiment, the electronic device calculates a similarity value between the initial object feature and a resource feature corresponding to each media resource to be recommended, for example, a cosine similarity value obtained by a cosine similarity calculation method, and determines the similarity value as a second matching degree result.
Step three: and according to the second matching degree result, determining a second target media resource from the media resources to be recommended, and recommending the second target media resource to the object account.
In an embodiment, the electronic device sorts the media resources to be recommended according to the sequence of the second matching degree from large to small to obtain a sequence, and then determines the media resources to be recommended in the preset number before the sequence as the second target media resources. Of course, the sequence may also be arranged according to the order of the second matching degree from the small to the medium to be recommended to obtain a sequence, and determine the resource to be recommended in the sequence with the later preset number as the second target medium resource. In another embodiment, the electronic device determines a to-be-recommended media resource of which the cosine similarity value is greater than or equal to a preset threshold as a second target media resource.
In the above embodiment, the object account that has not acquired the plurality of media resources operated in the latest preset time period is usually a newly registered object account, and since the object portrait information of the part of the object accounts is determined based on the information of the object account in the latest time period, the correlation between the object portrait information and the preference of the object account at this time is relatively high, the preference of the object account can be accurately expressed based on the initial object feature determined by the object portrait information, so that the second target media resource obtained based on the initial object feature can be more accepted by the object account, the accuracy of the recommended content is higher, and furthermore, the problem that the content cannot be recommended for the object account due to the media resource that has not acquired the operation can be avoided.
In one possible implementation, S103 includes:
the method comprises the following steps: based on the object image information of the object account, the initial object characteristics of the object account are determined.
Optionally, the electronic device obtains attribute information and interest information of the subject account, where the attribute information includes information such as age, gender, and academic calendar, so as to construct subject portrait information of the subject account.
In one embodiment, the object image information of the object account is input into the feature extraction model to obtain a feature vector corresponding to the object image information, and then the feature vector corresponding to the object image information is determined as an initial object feature of the object account.
Step two: and determining the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources and the initial object characteristics of the object account.
Optionally, the electronic device determines an initial object feature of the object account according to the feature resource and the initial object feature corresponding to each media resource.
In one embodiment, the electronic device filters the resource features corresponding to each media resource based on the initial object features, eliminates the media resources with similarity smaller than or equal to a preset threshold with the initial object features, and determines the object features of the object account based on the resource features corresponding to the remaining media resources in the plurality of media resources.
For example, the category to which the remaining media resource belongs may be determined based on the resource characteristics corresponding to the remaining media resource, and the object characteristics of the object account may be determined based on the category containing the largest number of media resources. Or, feature fusion can be performed on resource features corresponding to the remaining media resources, and a feature vector obtained after feature fusion is determined as an object feature of the object account.
In another embodiment, the electronic device filters the resource features corresponding to each media resource based on the initial object features, eliminates the resource features corresponding to the media resources with the initial object feature similarity smaller than or equal to a preset threshold, predicts the resource features corresponding to the media resources operated next after the operation of the plurality of media resources by the object account based on the resource features corresponding to the remaining media resources and the time information of the media resource operation, and determines the resource features as the object features of the object account.
In another embodiment, the electronic device performs feature fusion on the initial object features and the resource features corresponding to each media resource to obtain the object features of the object account, so that the object features of the object account can represent both the features of the operated media resources and the basic attributes of the object account, and thus the object features of the object account can more accurately express the preference of the object account, and the accuracy of the target media resources determined based on the object features of the object account is improved.
In the embodiment, the object features of the object account are determined through the resource features corresponding to each media resource and the initial object features determined by the object portrait information, so that the media resources operated in the latest preset time period are screened based on the initial object features, the media resources with low similarity to the initial object features are eliminated, and the object features of the object account are determined based on the remaining media resources, so that the expression capability of the object features of the finally obtained object account is improved, the object features of the object account can more accurately express the preference of the object account, and the accuracy of the target media resources determined based on the object features of the object account is improved.
The scheme provided by the embodiment of the application is mainly introduced from the perspective of a method. To implement the above functions, it includes hardware structures and/or software modules for performing the respective functions. Those of skill in the art would readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is performed as hardware or computer software drives hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The embodiment of the disclosure also provides a content recommendation device.
Fig. 6 is a block diagram illustrating a content recommendation device according to an example embodiment. Referring to fig. 6, the content recommendation apparatus 600 includes an acquisition module 601, a determination module 602, and a recommendation module 603.
The obtaining module 601 is configured to perform obtaining of a plurality of media resources operated in a latest preset time period of the object account, and determine a resource feature corresponding to each media resource. For example, in conjunction with fig. 1, the obtaining module 601 may be configured to execute S101.
A determining module 602 configured to perform determining an object feature of the object account according to a resource feature corresponding to each of the plurality of media resources. For example, in conjunction with fig. 1, the determining module 602 may be configured to perform S102.
The determining module 602 is further configured to determine, according to the object features of the object account, a matching degree between the object account and the media resource to be recommended, and obtain a first matching degree result. For example, in conjunction with fig. 1, the recommendation module 603 may be configured to perform S103.
And the recommending module 603 is configured to determine a first target media resource from the media resources to be recommended according to the first matching degree result, and recommend the first target media resource to the object account. For example, in conjunction with fig. 1, the recommendation module 603 may be configured to perform S104.
In one possible implementation, the determining module 602 is specifically configured to perform: predicting a target resource characteristic corresponding to a next operated media resource after the target account operates the plurality of media resources based on the resource characteristic corresponding to each media resource and the operation time information of each media resource; and determining the target resource characteristics as the object characteristics of the object account.
In another possible implementation, the determining module 602 is specifically configured to perform: and determining the resource characteristics corresponding to each media resource based on at least one of the image information, the text information and the voice information of each media resource.
In another possible implementation, the determining module 602 is specifically configured to perform: and performing recursion processing on the resource characteristics corresponding to each media resource based on the pre-trained recurrent neural network and the operation time information of each media resource to obtain target resource characteristics.
In another possible implementation, the recommending module 603 is specifically configured to perform: under the condition that the media resource to be recommended comprises the video resource to be recommended, acquiring a resource feature corresponding to the video resource to be recommended based on image information, text information and voice information of the video resource to be recommended and a pre-trained feature extraction network; and determining the matching degree of the object account and the media resource to be recommended based on the object characteristics of the object account and the resource characteristics corresponding to the video resource to be recommended to obtain a first matching degree result.
In another possible embodiment, the apparatus further includes a training module configured to perform: determining a training sample for each of a plurality of video samples; the training samples comprise image information, text information and voice information of each video sample; and training the feature extraction network based on the image information, the text information and the voice information of each video sample to obtain a pre-trained feature extraction network.
In another possible embodiment, the training samples further include a class identifier for each video sample, and the training module is specifically configured to perform: inputting the image information, the text information and the voice information of each video sample into an initial feature extraction network to obtain a prediction result of each video sample; determining a loss function value based on the prediction result of each video sample and the category identification of each video sample, and updating the parameters of the initial feature extraction network based on the loss function value; and iteratively executing the steps on the updated initial feature extraction network until the initial feature extraction network meets the model convergence condition, and determining the converged initial feature extraction network as a pre-trained feature extraction network.
In another possible implementation, the determining module 602 is further configured to perform: under the condition that a plurality of media resources operated in the latest preset time period of the object account are not acquired, determining initial object characteristics of the object account according to the object portrait information of the object account; determining the matching degree between the object account and the media resource to be recommended based on the initial object characteristics of the object account to obtain a second matching degree result; and according to the second matching degree result, determining a second target media resource from the media resources to be recommended, and recommending the second target media resource to the object account.
In another possible implementation, the determining module 602 is further configured to perform: determining initial object characteristics of the object account based on the object image information of the object account; and determining the object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources and the initial object characteristics of the object account.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
FIG. 7 is a block diagram illustrating an electronic device in accordance with an example embodiment. As shown in fig. 7, electronic device 700 includes, but is not limited to: a processor 701 and a memory 702.
The memory 702 is used for storing the executable instructions of the processor 701. It is understood that the processor 701 is configured to execute instructions to implement the content recommendation method shown in any one of fig. 1-3 in the above embodiments.
It should be noted that the electronic device structure shown in fig. 7 does not constitute a limitation of the electronic device, and the electronic device may include more or less components than those shown in fig. 7, or combine some components, or arrange different components, as will be understood by those skilled in the art.
The processor 701 is a control center of the electronic device, connects various parts of the whole electronic device by using various interfaces and lines, and performs various functions of the electronic device and processes data by operating or executing software programs and/or modules stored in the memory 702 and calling data stored in the memory 702, thereby performing overall monitoring of the electronic device. Processor 701 may include one or more processing units; optionally, the processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly handles operating systems, user interfaces, application programs, and the like, and the modem processor mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 701.
The memory 702 may be used to store software programs as well as various data. The memory 702 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program (such as an acquisition module, a determination module, or a recommendation module) required by at least one functional module, and the like. Further, the memory 702 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
In an exemplary embodiment, the disclosed embodiment also provides a computer-readable storage medium including instructions, for example, a memory 702 including instructions, which are executable by a processor 701 of an electronic device 700 to perform the content recommendation method illustrated in any one of fig. 1-3 in the above-described embodiments.
In practical implementation, the processing functions of the obtaining module 601, the determining module 602 and the recommending module 603 can be implemented by the processor 701 shown in fig. 6 calling the program code in the memory 702. The specific implementation process may refer to the description of the push-in recommendation method portion shown in any one of fig. 1 to fig. 3, which is not described herein again.
Alternatively, the computer-readable storage medium may be a non-transitory computer-readable storage medium, which may be, for example, a Read-Only Memory (ROM), a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
In an exemplary embodiment, the disclosed embodiments also provide a computer program product comprising one or more instructions executable by the processor 701 of the electronic device 700 to perform the content recommendation method illustrated in any of fig. 1-3 in the above-described embodiments.
It should be noted that the instructions in the computer-readable storage medium or one or more instructions in the computer program product are executed by the processor 701 of the electronic device 700 to implement the processes of the embodiment of the task processing method, and the same technical effect as the content recommendation method shown in any one of fig. 1 to fig. 3 in the embodiment can be achieved, and in order to avoid repetition, no further description is provided here.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (10)
1. A content recommendation method, comprising:
the method comprises the steps of obtaining a plurality of media resources operated in a recent preset time period of an object account, and determining resource characteristics corresponding to each media resource;
determining object characteristics of the object account according to the resource characteristics corresponding to each media resource in the plurality of media resources;
according to the object characteristics of the object account, determining the matching degree between the object account and the media resource to be recommended to obtain a first matching degree result;
and according to the first matching degree result, determining a first target media resource from the media resources to be recommended, and recommending the first target media resource to the object account.
2. The method of claim 1, wherein the determining the object characteristics of the object account according to the resource characteristics corresponding to each of the plurality of media resources comprises:
predicting a target resource characteristic corresponding to a next operated media resource after the target account operates the plurality of media resources based on the resource characteristic corresponding to each media resource and the operation time information of each media resource;
and determining the target resource characteristics as the object characteristics of the object account.
3. The method of claim 2, wherein the determining the resource characteristic corresponding to each of the media resources comprises:
and determining the resource characteristics corresponding to each media resource based on at least one of the image information, the text information and the voice information of each media resource.
4. The method according to claim 2, wherein predicting, based on the resource characteristic corresponding to each media resource and the operation time information of each media resource, a target resource characteristic corresponding to a media resource that is operated next after the plurality of media resources are operated by the subject account comprises:
and performing recursive processing on the resource characteristics corresponding to each media resource based on the pre-trained recurrent neural network and the operation time information of each media resource to obtain the target resource characteristics.
5. The method according to claim 1, wherein when the media resource to be recommended includes a video resource to be recommended, the determining, according to the object feature of the object account, the matching degree between the object account and the media resource to be recommended, and obtaining a first matching degree result includes:
acquiring resource characteristics corresponding to the video resources to be recommended based on the image information, the text information and the voice information of the video resources to be recommended and a pre-trained characteristic extraction network;
and determining the matching degree of the object account and the media resource to be recommended based on the object characteristics of the object account and the resource characteristics corresponding to the video resource to be recommended to obtain a first matching degree result.
6. The method of claim 5, further comprising:
determining a training sample for each of a plurality of video samples; the training samples comprise image information, text information and voice information of each video sample;
and training an initial feature extraction network based on the image information, the text information and the voice information of each video sample to obtain the pre-trained feature extraction network.
7. A content recommendation apparatus characterized by comprising:
the acquisition module is configured to execute acquisition of a plurality of media resources operated in a latest preset time period of an object account and determine resource characteristics corresponding to each media resource;
a determining module configured to determine an object feature of the object account according to a resource feature corresponding to each of the plurality of media resources;
the determining module is further configured to determine a matching degree between the object account and the media resource to be recommended according to the object characteristics of the object account, so as to obtain a first matching degree result;
and the recommending module is configured to determine a first target media resource from the media resources to be recommended according to the first matching degree result, and recommend the first target media resource to the object account.
8. An electronic device, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the content recommendation method of any one of claims 1 to 6.
9. A computer-readable storage medium, wherein instructions in the computer-readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the content recommendation method of any of claims 1-6.
10. A computer program product, characterized in that it comprises computer instructions which, when run on an electronic device, cause the electronic device to perform the content recommendation method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111591821.3A CN114328995A (en) | 2021-12-23 | 2021-12-23 | Content recommendation method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111591821.3A CN114328995A (en) | 2021-12-23 | 2021-12-23 | Content recommendation method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114328995A true CN114328995A (en) | 2022-04-12 |
Family
ID=81055054
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111591821.3A Pending CN114328995A (en) | 2021-12-23 | 2021-12-23 | Content recommendation method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114328995A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114676341A (en) * | 2022-04-14 | 2022-06-28 | 杭州网易云音乐科技有限公司 | Method, medium, device and computing equipment for determining recommended object |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389168A (en) * | 2018-09-29 | 2019-02-26 | 国信优易数据有限公司 | Project recommendation model training method, item recommendation method and device |
| CN111666922A (en) * | 2020-07-02 | 2020-09-15 | 上海眼控科技股份有限公司 | Video matching method and device, computer equipment and storage medium |
| CN112148899A (en) * | 2020-10-15 | 2020-12-29 | 腾讯科技（深圳）有限公司 | Multimedia recommendation method, device, equipment and storage medium |
-
2021
- 2021-12-23 CN CN202111591821.3A patent/CN114328995A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389168A (en) * | 2018-09-29 | 2019-02-26 | 国信优易数据有限公司 | Project recommendation model training method, item recommendation method and device |
| CN111666922A (en) * | 2020-07-02 | 2020-09-15 | 上海眼控科技股份有限公司 | Video matching method and device, computer equipment and storage medium |
| CN112148899A (en) * | 2020-10-15 | 2020-12-29 | 腾讯科技（深圳）有限公司 | Multimedia recommendation method, device, equipment and storage medium |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114676341A (en) * | 2022-04-14 | 2022-06-28 | 杭州网易云音乐科技有限公司 | Method, medium, device and computing equipment for determining recommended object |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10824874B2 (en) | Method and apparatus for processing video | |
| US11941527B2 (en) | Population based training of neural networks | |
| CN110766142A (en) | Model generation method and device | |
| CN107463701B (en) | Method and device for pushing information stream based on artificial intelligence | |
| EP3893125A1 (en) | Method and apparatus for searching video segment, device, medium and computer program product | |
| CN106407381B (en) | A kind of method and apparatus of the pushed information based on artificial intelligence | |
| CN106227792B (en) | Method and apparatus for pushed information | |
| CN111160191A (en) | Video key frame extraction method and device and storage medium | |
| CN110597965B (en) | Emotion polarity analysis method and device for article, electronic equipment and storage medium | |
| CN110046571B (en) | Method and device for identifying age | |
| CN113792212B (en) | Multimedia resource recommendation method, device, equipment and storage medium | |
| CN114492601A (en) | Resource classification model training method and device, electronic equipment and storage medium | |
| CN111144567A (en) | Training method and device of neural network model | |
| CN116756576B (en) | Data processing method, model training method, electronic device and storage medium | |
| CN114328995A (en) | Content recommendation method, device, equipment and storage medium | |
| CN112269943B (en) | Information recommendation system and method | |
| CN117235371A (en) | Video recommendation method, model training method and device | |
| CN114119123A (en) | Information pushing method and device | |
| CN116956183A (en) | Multimedia resource recommendation method, model training method, device and storage medium | |
| CN114491093B (en) | Multimedia resource recommendation and object representation network generation method and device | |
| CN112269942B (en) | Method, device and system for recommending object and electronic equipment | |
| US9753745B1 (en) | System and method for system function-flow optimization utilizing application programming interface (API) profiling | |
| CN114357242A (en) | Training evaluation method and device based on recall model, equipment and storage medium | |
| CN113742593A (en) | Method and device for pushing information | |
| EP3683733A1 (en) | A method, an apparatus and a computer program product for neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |