Recommended methods, apparatus, readable storage media, and electronic devices

Abstract

This disclosure relates to a recommendation method, apparatus, readable storage medium, and electronic device, comprising: acquiring attribute information and the main business type of a network object; inputting the attribute information and the main business type into a ranking and classification model to obtain the ranking and classification result of the network object output by the ranking and classification model, wherein the ranking and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks, each business sub-network being used to extract at least a business feature vector of the network object under the corresponding business type, the weight generation sub-network being used to generate weights corresponding to each business type based on the generation parameters of the corresponding main business type, the feature fusion network being used to perform weighted fusion of the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain a target feature vector, and outputting the ranking and classification result based on the target feature vector; and recommending network objects based on the ranking and classification results of multiple network objects.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more specifically, to a recommended method, apparatus, readable storage medium 5, and electronic device. Background Technology

[0002] In recent years, the rapid development of technologies such as cloud computing, the Internet of Things, mobile internet, and artificial intelligence has brought many conveniences to people's work and life. Users can easily search for the information they want through the internet. However, when faced with...

[0003] Faced with the explosive growth of online information, users often struggle to make efficient choices. Recommender systems offer an effective solution to this information overload. A subset of information filtering systems, recommender systems aim to predict a user's preferences for online objects based on their likes, habits, personalized needs, and the characteristics of the online objects, thus recommending the most suitable options.

[0004] Appropriate online objects help users make quick decisions. Online objects can include videos, products, news, advertisements, etc.

[0005] Traditional recommender system models are mostly built based solely on users' historical behavior sequences. These models suffer from insufficient generalization ability and a lack of novelty in the recommendation results. In particular, for some new or niche network objects, due to the lack of historical interaction information or the limited number of training samples, there is a problem of sample scarcity, making it impossible to continue to accurately model and recommend, thus failing to achieve the recommendation effect required by users. Summary of the Invention

[0006] The purpose of this disclosure is to provide a recommended method, apparatus, readable storage medium, and electronic device to solve the aforementioned problems existing in the related art.

[0007] To achieve the above objectives, a first aspect of the disclosed embodiments provides a recommendation method, the method comprising: 5. obtaining attribute information of a network object and the main business type to which the network object belongs;

[0008] The attribute information and the main business type are input into a pre-trained ranking and classification model to obtain the ranking and classification result of the network object output by the ranking and classification model. The ranking and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate weights corresponding to each business type based on the pre-trained generation parameters corresponding to the main business type. The feature fusion network is used to perform weighted fusion of the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain a target feature vector, and output the ranking and classification result according to the target feature vector.

[0009] Based on the ranking and classification results of multiple network objects, network objects are recommended to the target user.

[0010] Optionally, the attribute information of the network object includes the user information of the target user;

[0011] The business sub-network is used to extract common feature vectors of different users under the corresponding business type and personalized feature vectors of the target user under the business type, and to fuse the common feature vectors and the personalized feature vectors to obtain the business feature vector.

[0012] Optionally, the training samples of the ranking classification model include training samples corresponding to the different business types, the labels of the training samples include ranking classification labels, and the training method of the ranking classification model includes:

[0013] For any training sample corresponding to any of the aforementioned business types, each of the business sub-networks of the ranking classification model outputs a business feature vector corresponding to the training sample.

[0014] The weights generated by the weight generation subnetwork based on the generation parameters for each business type are weighted and fused to the business feature vectors of the training samples under each business type, and the sorting and classification results are output based on the weighted fusion results.

[0015] The model loss value is calculated using a first loss function based on the sorting and classification results and the sorting and classification labels.

[0016] The model parameters of the ranking classification model are updated based on the model loss value, and the model parameters include the generated parameters.

[0017] Optionally, the training samples include sample data and the main business type label to which the sample data belongs, the weight generation subnetwork includes generation parameters corresponding to different business types, and the training method of the ranking classification model further includes:

[0018] Based on the main service type label, the corresponding generation parameters in the weight generation sub-network are determined so that the weight generation sub-network can generate weights corresponding to each service type based on the generation parameters.

[0019] The step of updating the model parameters of the ranking classification model based on the model loss value includes:

[0020] The generation parameters corresponding to the main business type label are updated based on the model loss value, so that the updated generation parameters can strengthen the weight of the business sub-network corresponding to the main business type label.

[0021] Optionally, each of the service sub-networks includes a bottom-level feature network, a user feature network, and a fusion network. The training samples of the ranking classification model include user data samples and network object data samples. The step of outputting a service feature vector corresponding to the training samples through each of the service sub-networks of the ranking classification model includes:

[0022] For each of the aforementioned service sub-networks, the training samples are input into the underlying feature network, and the user data samples are input into the user feature network. The output vectors of the underlying feature network and the user feature network are then fused through the fusion network to obtain the service feature vector of the training samples.

[0023] The step of updating the model parameters of the ranking classification model based on the model loss value includes:

[0024] The parameters of the user feature network are updated based on the model loss value.

[0025] Optionally, the underlying feature networks of each of the business sub-networks are connected through a residual structure, and updating the model parameters of the ranking classification model based on the model loss value includes:

[0026] The parameters of the underlying feature network in each of the service sub-networks are updated using the residual structure.

[0027] According to a second aspect of the present disclosure, a recommendation device is provided, the recommendation device comprising:

[0028] The acquisition module is used to acquire the attribute information of the network object and the main business type to which the network object belongs;

[0029] The ranking module is used to input the attribute information and the main business type into a pre-trained ranking and classification model to obtain the ranking and classification result of the network object output by the ranking and classification model. The ranking and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate weights corresponding to each business type based on the pre-trained generation parameters corresponding to the main business type. The feature fusion network is used to perform weighted fusion on the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain a target feature vector, and output the ranking and classification result according to the target feature vector.

[0030] The recommendation module is used to recommend network objects to the target user based on the ranking and classification results of multiple network objects.

[0031] Optionally, the attribute information of the network object includes the user information of the target user;

[0032] The business sub-network is used to extract common feature vectors of different users under the corresponding business type and personalized feature vectors of the target user under the business type, and to fuse the common feature vectors and the personalized feature vectors to obtain the business feature vector.

[0033] According to a third aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps of the recommended method provided in the first aspect of the present disclosure.

[0034] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising:

[0035] A memory on which computer programs are stored;

[0036] A processor for executing the computer program in the memory to implement the steps of the recommended method provided in the first aspect of this disclosure.

[0037] The above technical solution involves inputting the attribute information of network objects and the main business type to which the network objects belong into a pre-trained ranking and classification model to obtain ranking and classification results. These results are then used to categorize and classify multiple network objects.

[0038] The ranking and classification model recommends network objects to the target user based on the class results. Each business sub-network of the ranking and classification model is used to extract the business feature vector of the network object under the corresponding business type, and the weight generation sub-network of the ranking and classification model is used to generate the weight based on the corresponding main business.

[0039] The generation parameters for each business type generate weights corresponding to that type. The feature fusion network of the ranking and classification model is used to weight and fuse the business feature vectors extracted from each business sub-network according to the weights corresponding to each business type, to obtain the target feature vector, and then obtain the ranking and classification results based on the target feature vector. This allows for the classification of new network objects or niche markets.

[0040] The network objects can not only learn from the business sub-networks corresponding to the main business type, but also supplement their learning from the business sub-networks corresponding to other business types, generating a well-learned ranking and classification model to alleviate the sparsity of low-frequency services.

[0041] The goal is to achieve knowledge transfer, thereby obtaining reasonable ranking and classification results, recommending high-quality network objects to target users, and improving user recommendation effectiveness.

[0042] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0043] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings:

[0044] Figure 1 This is a flowchart illustrating a recommended method according to an exemplary embodiment;

[0045] Figure 2 This is a schematic diagram illustrating a recommendation classification model according to an exemplary embodiment;

[0046] Figure 3 This is a flowchart illustrating a training method for a recommendation classification model according to an exemplary embodiment;

[0047] Figure 4 This is a block diagram illustrating a recommended apparatus according to an exemplary embodiment;

[0048] Figure 5 This is a block diagram illustrating an electronic device according to an exemplary embodiment.

[0049] Explanation of reference numerals in the attached figures

[0050] 110 - Weight generation subnetwork; 120 - Feature fusion network; 130 - Service subnetwork; 400 - Recommendation device; 401 - Acquisition module; 402 - Ranking module; 403 - Recommendation module; 700 - Electronic device; 701 - Processor; 702 - Memory; 703 - Multimedia component; 704 - I / O interface; 705 - Communication component. Detailed Implementation

[0051] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0052] In the following description, the words "first" and "second" are used only to distinguish the purpose of the description and should not be interpreted as indicating or implying relative importance or order.

[0053] This disclosure provides a recommended method; please refer to [link / reference]. Figure 1 This recommended method can be applied to electronic devices, which can be implemented in various forms. For example, the electronic devices described in the embodiments of this disclosure may include mobile phones, tablets, laptops, desktop computers, handheld computers, personal digital assistants (PDAs), portable media players (PMPs), navigation devices, wearable devices, smart bracelets, etc. The recommended method includes:

[0054] Step S101: Obtain the attribute information of the network object and the main business type to which the network object belongs.

[0055] Online objects can be videos, products, news, advertisements, and other online recommended objects. For example, when a user clicks to enter an application, the application prepares to recommend online objects to the user, but the final online objects presented are those determined after sorting and filtering processes.

[0056] The attribute information of a network object can include the target user's user information, the network object's object information, and the interaction information between the target user and the network object. The object information of a network object can include video content, product functions, news content, advertising content and functions, etc. The interaction information between the target user and the network object can include the target user's historical clicks or browsing history.

[0057] The business type can be a classification based on the network object. For example, when the network object is a video, the business type can be, for example, comedy, food, fashion, travel, entertainment, lifestyle, news, parenting, knowledge, games, automobiles, finance, pets, sports, music, etc.; when the network object is a product, the business type can be, for example, hardware, chemicals, aquatic products, food, etc.; when the network object is news, the business type can be, for example, politics, economics, law, military, science and technology, culture and education, sports, society, etc.; when the network object is an advertisement, the business type can be, for example, product advertisement, corporate advertisement, service advertisement, concept advertisement, etc. The above are merely illustrative examples of business types. In other embodiments, other different classification methods may also be used, and this disclosure does not limit them.

[0058] Among all business types, the business type to which a network object most likely belongs is the main business type to which the network object belongs.

[0059] Step S102: Input the attribute information and main business type into the pre-trained ranking classification model to obtain the ranking classification result of the network object output by the ranking classification model. The ranking classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate the weight corresponding to each business type based on the generation parameters of the corresponding main business type obtained in pre-training. The feature fusion network is used to perform weighted fusion on the business feature vectors extracted by each business sub-network according to the weights of each business type to obtain the target feature vector, and output the ranking classification result according to the target feature vector.

[0060] The pre-trained ranking and classification model is obtained through machine learning. Specifically, a training dataset is first collected, and then the neural network is trained and modeled using the collected training dataset according to a preset algorithm. Based on the training dataset, the rules are summarized to obtain the trained ranking and classification model.

[0061] This can be understood as the ranking and classification model being able to be stored locally on the electronic device after training. See also... Figure 2 The ranking and classification model includes a weight generation sub-network 110, a feature fusion network 120, and multiple business sub-networks 130 corresponding to different business types. The weight generation sub-network 110 and the feature fusion network 120 are connected, and the multiple business sub-networks 130 are all connected to the feature fusion network 120. The business type can correspond one-to-one with the business sub-network.

[0062] The ranking and classification result can be the output obtained by sequentially performing feature extraction and Sigmoid function operations on the target feature vector. In other embodiments of this disclosure, a percentage calculation can be performed after the Sigmoid function operation to obtain the output ranking and classification result.

[0063] Step S103: Recommend network objects to the target user based on the sorting and classification results of multiple network objects.

[0064] For each network object, steps S101 and S102 are executed to obtain the sorting and classification results of each network object. The sorting and classification results of each network object among multiple network objects are compared to determine a preset number of network objects to recommend to the target user.

[0065] For example, there are a total of 10 network objects, with a preset quantity of 3. The corresponding sorting and classification results of the network objects are 91%, 49%, 73%, 27%, 68%, 23%, 51%, 89%, 24%, and 65%. The top three sorting and classification results of 91%, 89%, and 73% are selected from the sorting and classification results, and their corresponding network objects are recommended to the target user.

[0066] By inputting the attribute information of network objects and their respective main business types into a pre-trained ranking and classification model, ranking and classification results are obtained. Based on these results, network objects are recommended to the target user. Each business sub-network 130 of the ranking and classification model extracts the business feature vector of the network object under its corresponding business type. The weight generation sub-network generates weights for each business type based on the generation parameters corresponding to the main business type. The feature fusion network weights and fuses the business feature vectors extracted by each business sub-network according to the weights for each business type, obtaining the target feature vector. The ranking and classification results are then derived from this target feature vector. In this way, for new or niche network objects, learning can be achieved not only from the business sub-networks corresponding to the main business type but also from the business sub-networks corresponding to other business types, generating a well-learned ranking and classification model. This alleviates the sparsity problem of low-frequency business objects, achieves knowledge transfer, and ultimately yields reasonable ranking and classification results. This allows for the recommendation of high-quality network objects to the target user, improving the user recommendation effect.

[0067] In one possible implementation, the attribute information of the network object includes the user information of the target user.

[0068] The business sub-network is used to extract the common feature vectors of different users under the corresponding business type and the personalized feature vectors of the target user under the business type, and then fuse the common feature vectors and personalized feature vectors to obtain the business feature vector.

[0069] The target user's information can include user profiles and user location information. User profiles can include personal information such as gender, age, and occupation, while user location information can include current location, time, weather, and environment.

[0070] The business sub-network is used to extract common feature vectors of different users under the corresponding business type and personalized feature vectors of the target user under the corresponding business type. The common feature vectors and personalized feature vectors are then fused to obtain the business feature vector. This can be understood as extracting common information of different users under the corresponding business type of the business sub-network and personalized information of the target user under the corresponding business type of the business sub-network, and obtaining the business feature vector by fusing the common information and personalized information.

[0071] By fusing common information from different users with personalized information from target users, the final business feature vector is obtained. This fully considers the personalized differences of target users under this business type, thereby enabling personalized recommendations for target users.

[0072] In one possible implementation, the training samples for the ranking classification model include training samples corresponding to different business types, and the labels of the training samples include ranking and classification labels. (See [link to relevant documentation]). Figure 3 The training method for the ranking classification model may include steps S201 to S204:

[0073] Step S201: For any training sample corresponding to any business type, the business feature vector of the corresponding training sample is output by each business sub-network of the sorting classification model.

[0074] Step S202: Based on the weights generated by the sub-network according to the generation parameters, the weights corresponding to each business type are generated, and the business feature vectors of the corresponding training samples under each business type are weighted and fused. The sorting and classification results are then output based on the weighted fusion results.

[0075] Step S203: Calculate the model loss value using the first loss function based on the sorting and classification results and the sorting and classification labels.

[0076] The first loss function can be the cross-entropy loss function.

[0077] Step S204: Update the model parameters of the ranking classification model based on the model loss value. The model parameters include the generation parameters.

[0078] Iteratively execute steps S202 to S204 until the calculated model loss value is lower than the preset loss value.

[0079] In one possible implementation, the training samples include sample data and the main business type label to which the sample data belongs; the weight generation subnetwork includes generation parameters corresponding to different business types; and the training method for the ranking classification model also includes:

[0080] Based on the main business type label, determine the corresponding generation parameters in the weight generation sub-network so that the weight generation sub-network can generate weights for each business type according to the generation parameters.

[0081] The model parameters of the ranking classification model are updated based on the model loss value, including:

[0082] The generated parameters corresponding to the main business type label are updated based on the model loss value, so that the updated generated parameters can strengthen the weight of the business sub-network corresponding to the main business type label.

[0083] By determining the corresponding generation parameters in the weight generation sub-network based on the main business type label, the weight generation sub-network can generate the weight corresponding to each business type according to the generation parameters, thereby distinguishing the weight differences of each business sub-network and strengthening the weight of the business sub-network corresponding to the main business type.

[0084] In other embodiments, the main business type label can be calculated using an auxiliary loss function to obtain a gated loss value. The model loss value and the gated loss value are then propagated to the weight generation sub-network to notify the weight generation sub-network to update the generation parameters corresponding to the business sub-network indicated by the gated loss value based on the model loss value. This allows the updated generation parameters to strengthen the weights of the business sub-network corresponding to the main business type label.

[0085] For example, if the primary business type of the network object is food delivery, and the other business types are e-commerce, content, and comprehensive services, the auxiliary enhancement loss function uses one-hot coding to set the code corresponding to the primary business type to 1, and the codes corresponding to the other business types to 0. The order of the business types is food delivery, e-commerce, content, and comprehensive services, then the gate loss value is 1000.

[0086] In one possible implementation, each business sub-network includes a bottom-level feature network, a user feature network, and a fusion network. The training samples for the ranking classification model include user data samples and network object data samples. Each business sub-network of the ranking classification model outputs a business feature vector corresponding to the training samples, including:

[0087] For each business sub-network, the training samples are input into the bottom feature network, and the user data samples are input into the user feature network. The output vectors of the bottom feature network and the user feature network are fused through the fusion network to obtain the business feature vector of the training samples.

[0088] For each business sub-network, user sample data and network object data samples from the training samples are input into the bottom feature network, and user data samples are input into the user feature network. The fusion network fuses the output vectors of the bottom feature network and the user feature network to obtain the business feature vector of the training samples.

[0089] The model parameters of the ranking classification model are updated based on the model loss value, including:

[0090] The parameters of the user feature network are updated based on the model loss value.

[0091] By updating the parameters of the user feature network, the ability to extract personalized information in the business sub-network can be enhanced.

[0092] In one possible implementation, the underlying feature networks of each business sub-network are connected through a residual structure, and the model parameters of the ranking classification model are updated based on the model loss value, including:

[0093] The parameters of the underlying feature network in each service sub-network are updated using the residual structure.

[0094] By connecting the underlying feature networks of each business sub-network through a residual structure, data sharing between business sub-networks can be achieved, and common information of different users can be captured more effectively.

[0095] To implement the above-described method embodiments, this embodiment provides a recommended device 400, such as... Figure 4 As shown, Figure 4 This is a block diagram illustrating a recommendation device 400 according to an exemplary embodiment. The recommendation device 400 may include an acquisition module 401, a sorting module 402, and a recommendation module 403.

[0096] Module 401 is used to obtain the attribute information of the network object and the main business type to which the network object belongs;

[0097] The sorting module 402 is used to input attribute information and main business type into a pre-trained sorting and classification model to obtain the sorting and classification result of the network object output by the sorting and classification model. The sorting and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate the weight corresponding to each business type based on the generation parameters of the corresponding main business type obtained in pre-training. The feature fusion network is used to perform weighted fusion on the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain the target feature vector, and output the sorting and classification result according to the target feature vector.

[0098] Recommendation module 403 is used to recommend network objects to target users based on the ranking and classification results of multiple network objects.

[0099] Optionally, the attribute information of the network object includes the user information of the target user;

[0100] The business sub-network is used to extract the common feature vectors of different users under the corresponding business type and the personalized feature vectors of the target user under the business type, and then fuse the common feature vectors and personalized feature vectors to obtain the business feature vector.

[0101] Optionally, the training samples for the ranking and classification model include training samples corresponding to different business types, and the labels of the training samples include ranking and classification labels. The recommendation device 400 also includes:

[0102] The first training module is used to output the business feature vector of the corresponding training sample through each business sub-network of the sorting classification model for any business type.

[0103] The second training module is used to generate 5 weights for each business type based on the generation parameters of the sub-network, and to perform weighted fusion on the business feature vectors of the corresponding training samples under each business type, and then perform weighted fusion on the sub-network.

[0104] The output will be a sorted and categorized result.

[0105] The third training module is used to calculate the model loss value based on the sorting and classification results and the sorting and classification labels using the first loss function.

[0106] The fourth training module is used to update the model parameters of the ranking classification model based on the model loss value. The model parameters include the generated parameters.

[0107] Optionally, the training samples include sample data and the main business type label to which the sample data belongs; the weight generation subnetwork includes generation parameters corresponding to different business types; and the recommendation device further includes:

[0108] The fifth training module is used to determine the corresponding generation parameters in the weight generation sub-network based on the main business type label, so that the weight generation sub-network can generate weights for each business type according to the generation parameters.

[0109] 5. The fourth training module is specifically used for:

[0110] The generated parameters corresponding to the main business type label are updated based on the model loss value, so that the updated generated parameters can strengthen the weight of the business sub-network corresponding to the main business type label.

[0111] Optionally, each service sub-network includes an underlying feature network, a user feature network, and a fusion network, ordered as follows:

[0112] The training samples for the classification model include user data samples and network object data samples. The first training module is specifically used for: 0. For each business sub-network, inputting the training samples into the underlying feature network, and inputting the user data samples into the user...

[0113] The feature network is used, and the output vectors of the underlying feature network and the user feature network are fused through a fusion network to obtain the business feature vector of the training samples;

[0114] The fourth training module is specifically used for:

[0115] The parameters of the user feature network are updated based on the model loss value.

[0116] 5. Optionally, the underlying feature networks of each service sub-network are connected through a residual structure. The fourth module is specifically used for:

[0117] The parameters of the underlying feature network in each service sub-network are updated using the residual structure.

[0118] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0119] Figure 5 This is a block diagram illustrating an electronic device 700 according to an exemplary embodiment. Figure 5 As shown, the electronic device 700 may include a processor 701 and a memory 702. The electronic device 700 may also include one or more of a multimedia component 703, an input / output (I / O) interface 704, and a communication component 705.

[0120] The processor 701 controls the overall operation of the electronic device 700 to complete all or part of the steps in the recommended method described above. The memory 702 stores various types of data to support the operation of the electronic device 700. This data may include, for example, instructions for any application or method operating on the electronic device 700, and application-related data such as contact data, sent and received messages, pictures, audio, video, etc. The memory 702 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 703 may include a screen and audio components. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory 702 or transmitted via communication component 705. The audio component also includes at least one speaker for outputting audio signals. I / O interface 704 provides an interface between processor 701 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons may be virtual or physical buttons. Communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IoT, eMTC, or other 5G technologies, or combinations thereof, is not limited here. Therefore, the corresponding communication component 705 may include: a Wi-Fi module, a Bluetooth module, an NFC module, etc.

[0121] In an exemplary embodiment, the electronic device 700 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the recommended method described above.

[0122] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the recommended method described above. For example, the computer-readable storage medium may be the memory 702 including program instructions described above, which may be executed by the processor 701 of the electronic device 700 to perform the recommended method described above.

[0123] In another exemplary embodiment, a computer program product is also provided, the computer program product comprising the ability to

[0124] A computer program executed by a programmable device, the computer program having a code portion for performing the recommended method described above when executed by the programmable device.

[0125] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0126] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this disclosure provides for various...

[0127] Possible combinations will not be described separately.

[0128] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A recommendation method, characterized in that, The method includes: Obtain the attribute information of the network object and the main business type to which the network object belongs; The attribute information and the main business type are input into a pre-trained ranking and classification model to obtain the ranking and classification result of the network object output by the ranking and classification model. The ranking and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate weights corresponding to each business type based on the pre-trained generation parameters corresponding to the main business type. The feature fusion network is used to perform weighted fusion of the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain a target feature vector, and output the ranking and classification result according to the target feature vector. Based on the ranking and classification results of multiple network objects, network objects are recommended to the target user.

2. The method according to claim 1, characterized in that, The attribute information of the network object includes the user information of the target user; The business sub-network is used to extract common feature vectors of different users under the corresponding business type and personalized feature vectors of the target user under the business type, and to fuse the common feature vectors and the personalized feature vectors to obtain the business feature vector.

3. The method according to claim 1, characterized in that, The training samples of the ranking classification model include training samples corresponding to the different business types, and the labels of the training samples include ranking classification labels. The training method of the ranking classification model includes: For any training sample corresponding to any of the aforementioned business types, each of the business sub-networks of the ranking classification model outputs a business feature vector corresponding to the training sample. The weights generated by the weight generation subnetwork based on the generation parameters for each business type are weighted and fused to the business feature vectors of the training samples under each business type, and the sorting and classification results are output based on the weighted fusion results. The model loss value is calculated using a first loss function based on the sorting and classification results and the sorting and classification labels. The model parameters of the ranking classification model are updated based on the model loss value, and the model parameters include the generated parameters.

4. The method according to claim 3, characterized in that, The training samples include sample data and the main business type label to which the sample data belongs; the weight generation subnetwork includes generation parameters corresponding to different business types; and the training method of the ranking classification model further includes: Based on the main service type label, the corresponding generation parameters in the weight generation sub-network are determined so that the weight generation sub-network can generate weights corresponding to each service type based on the generation parameters. The step of updating the model parameters of the ranking classification model based on the model loss value includes: The generation parameters corresponding to the main business type label are updated based on the model loss value, so that the updated generation parameters can strengthen the weight of the business sub-network corresponding to the main business type label.

5. The method according to claim 3, characterized in that, Each of the aforementioned business sub-networks includes a bottom-level feature network, a user feature network, and a fusion network. The training samples of the ranking classification model include user data samples and network object data samples. The business feature vector corresponding to the training samples is output by each of the aforementioned business sub-networks of the ranking classification model, including: For each of the aforementioned service sub-networks, the training samples are input into the underlying feature network, and the user data samples are input into the user feature network. The output vectors of the underlying feature network and the user feature network are then fused through the fusion network to obtain the service feature vector of the training samples. The step of updating the model parameters of the ranking classification model based on the model loss value includes: The parameters of the user feature network are updated based on the model loss value.

6. The method according to claim 5, characterized in that, The underlying feature networks of each of the aforementioned business sub-networks are connected through a residual structure. Updating the model parameters of the ranking classification model based on the model loss value includes: The parameters of the underlying feature network in each of the service sub-networks are updated using the residual structure.

7. A recommended device, characterized in that, The recommendation device includes: The acquisition module is used to acquire the attribute information of the network object and the main business type to which the network object belongs; The ranking module is used to input the attribute information and the main business type into a pre-trained ranking and classification model to obtain the ranking and classification result of the network object output by the ranking and classification model. The ranking and classification model includes a weight generation sub-network, a feature fusion network, and multiple business sub-networks corresponding to different business types. Each business sub-network is used to extract the business feature vector of the network object under the corresponding business type. The weight generation sub-network is used to generate weights corresponding to each business type based on the pre-trained generation parameters corresponding to the main business type. The feature fusion network is used to perform weighted fusion on the business feature vectors extracted by each business sub-network according to the weights corresponding to each business type to obtain a target feature vector, and output the ranking and classification result according to the target feature vector. The recommendation module is used to recommend network objects to the target user based on the ranking and classification results of multiple network objects.

8. The recommended device according to claim 7, characterized in that, The attribute information of the network object includes the user information of the target user; The business sub-network is used to extract common feature vectors of different users under the corresponding business type and personalized feature vectors of the target user under the business type, and to fuse the common feature vectors and the personalized feature vectors to obtain the business feature vector.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1-6.

10. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method according to any one of claims 1-6.

Images