Recommended methods, model training methods, devices, electronic equipment, and storage media

Abstract

The present disclosure provides a recommendation method, a model training method, an apparatus, a device and a storage medium, and relates to the field of artificial intelligence technology, especially to the field of deep learning technology and the field of data statistics technology. The specific implementation scheme is: determine the historical interaction resource sequence activated by the interest of the target object; determine the object characteristics of the target object with respect to the target interest based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object; determine multiple candidate resource characteristics with respect to the target interest; determine the resource to be recommended from the candidate resources based on the candidate resource characteristics and the object characteristics.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence technology, in particular to the fields of deep learning technology and data statistics technology, and specifically to recommendation methods, training methods, devices, electronic devices, storage media and program products for deep learning models. Background Art

[0002] With the rapid development of information technology and network technology, information expansion and redundancy have brought confusion in information selection to people's social activities and entertainment activities. It is extremely challenging to find the required information from the vast resources. The personalized information service of the Internet can provide different personalized information service strategies for different users. Automated information recommendation is performed based on the different characteristics and requirements of users. However, in the recommendation process, the satisfaction of the recommendation results in meeting the personalized needs of users needs to be improved. Summary of the invention

[0003] The present disclosure provides a recommendation method, a training method for a deep learning model, an apparatus, an electronic device, a storage medium, and a program product.

[0004] According to one aspect of the present disclosure, a recommendation method is provided, comprising:

[0005] Determine a historical interaction resource sequence activated by the interest of the target object, wherein the historical interaction resource sequence activated by the interest includes a plurality of historical interaction resources that the target object has interacted with in a historical period, and the plurality of historical interaction resources are all resources of the target interest;

[0006] Determining object features of the target object with respect to the target interest based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object;

[0007] Determine a plurality of candidate resource features about the target interest, wherein the plurality of candidate resource features correspond one-to-one to a plurality of candidate resources, the candidate resource features are used to characterize resource attribute features of the candidate resources corresponding to the candidate resource features, and the plurality of candidate resources are all resources of the target interest; and

[0008] Based on the multiple candidate resource characteristics and the object characteristics, a resource to be recommended is determined from the multiple candidate resources.

[0009] According to another aspect of the present disclosure, a method for training a deep learning model is provided, comprising:

[0010] Determine a sample history interaction resource sequence activated by the interest of the sample target object, wherein the sample history interaction resource sequence activated by the interest includes a plurality of sample history interaction resources interacted by the sample target object within the sample history period, and the plurality of sample history interaction resources are all resources of the sample target interest;

[0011] Determine the sample object features of the sample target object with respect to the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object;

[0012] Determine a target sample historical interaction resource from the sample historical interaction resource sequence activated by the interest of the sample target object;

[0013] Based on the sample resource attribute information of the target sample historical interaction resource, obtain the sample resource characteristics; and

[0014] Based on the above sample object characteristics and the above sample resource characteristics, the parameters of the above deep learning model are adjusted to complete the training.

[0015] According to another aspect of the present disclosure, there is provided a recommendation device, comprising:

[0016] A resource sequence determination module is used to determine a historical interaction resource sequence activated by the interest of the target object, wherein the historical interaction resource sequence activated by the interest includes a plurality of historical interaction resources that the target object has interacted with in a historical period, and the plurality of historical interaction resources are all resources of the target interest;

[0017] An object feature determination module, configured to determine an object feature of the target object with respect to the target interest based on a historical interaction resource sequence activated by the interest and object attribute information of the target object;

[0018] A candidate resource feature determination module, used to determine multiple candidate resource features about the target interest, wherein the multiple candidate resource features correspond to multiple candidate resources one by one, the candidate resource features are used to characterize resource attribute features of the candidate resources corresponding to the candidate resource features, and the multiple candidate resources are all resources of the target interest;

[0019] The module for determining resources to be recommended is used to determine resources to be recommended from the multiple candidate resources based on the multiple candidate resource characteristics and the object characteristics.

[0020] According to another aspect of the present disclosure, a training device for a deep learning model is provided, comprising:

[0021] A sample resource sequence determination module is used to determine a sample historical interaction resource sequence activated by the interest of the sample target object, wherein the sample historical interaction resource sequence activated by the interest includes a plurality of sample historical interaction resources that the sample target object has interacted with within a sample history period, and the plurality of sample historical interaction resources are all resources of the sample target interest;

[0022] A sample object feature determination module, configured to determine the sample object feature of the sample target object with respect to the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object;

[0023] A sample interaction resource determination module, configured to determine a target sample history interaction resource from the sample history interaction resource sequence activated by the interest of the sample target object;

[0024] A sample resource feature determination module, used to obtain sample resource features based on the sample resource attribute information of the target sample historical interaction resource;

[0025] The parameter adjustment module is used to adjust the parameters of the above-mentioned deep learning model based on the above-mentioned sample object characteristics and the above-mentioned sample resource characteristics to complete the training.

[0026] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method as disclosed in the present disclosure.

[0027] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method of the present disclosure.

[0028] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method of the present disclosure is implemented.

[0029] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0031] Figure 1 A schematic diagram schematically shows a recommended method according to a related example;

[0032] Figure 2 An exemplary system architecture to which the recommended method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown;

[0033] Figure 3 A flowchart of a recommendation method according to an embodiment of the present disclosure is schematically shown;

[0034] Figure 4 A schematic diagram of generating an offline database according to an embodiment of the present disclosure is schematically shown;

[0035] Figure 5 A schematic diagram schematically shows a method of generating interest features according to an embodiment of the present disclosure;

[0036] Figure 6 A schematic diagram schematically shows a method for training a graph neural network according to an embodiment of the present disclosure;

[0037] Figure 7 A schematic diagram schematically shows a recommended method according to another embodiment of the present disclosure;

[0038] Figure 8 A schematic diagram schematically shows a recommendation model according to an embodiment of the present disclosure;

[0039] Fig. 9 A flowchart of a deep learning model training method according to an embodiment of the present disclosure is schematically shown;

[0040] Fig.10 A block diagram of a recommendation device according to an embodiment of the present disclosure is schematically shown;

[0041] Fig.11 A block diagram schematically shows a training device for a deep learning model according to an embodiment of the present disclosure; and

[0042] Fig.12 A block diagram of an electronic device suitable for implementing a recommendation method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0043] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0044] According to relevant examples, in the recommendation field, deep neural networks (DNN) can be used as a recommendation model to recall resources to be recommended. For example, the advertising resources that the target object has interacted with, such as clicked, in the historical period and the relevant resource attribute information of multiple candidate advertising resources are input into the recommendation model, and the results are output. Based on the results, the resources to be recommended are determined from the multiple candidate advertising resources.

[0045] In the process of implementing the embodiments of the present disclosure, it is found that the above recommendation method focuses on discovering the similarity between resources by means of the propagation relationship between multiple historical interaction resources in the historical interaction resource sequence of the target object, and does not have the ability to express the long-term interests of the target object. Therefore, the recommendation method does not have the ability to recall the recommended resources that belong to the long-term interests with high precision.

[0046] According to another related example, in the recommendation field, the dual-tower model can also be used as a recommendation model to recall resources to be recommended. For example, the features of User (target object) and Item (resource) are extracted through dual deep neural networks to obtain object features and resource attribute features, and the degree of attention or interest of the target object in the resource is calculated through cosine similarity.

[0047] Figure 1 A schematic diagram of a recommended method according to a related example is schematically shown.

[0048] like Figure 1 As shown, the recommendation model may include a dual-tower model, including an embedding layer (Embedding layer) M111, an object processing module M112, a resource processing module M122 and a similarity calculation layer M131. The historical interaction resource sequence 111 of the target object can be input into the embedding layer M111 to obtain the historical interaction resource feature sequence of the target object, and the historical interaction resource feature sequence of the target object can be input into the object processing module M112 to obtain the object feature. The resource attribute information 121 of the resource is input into the embedding layer M111 to obtain the resource coding feature, and the resource coding feature is input into the resource processing module M122 to obtain the resource attribute feature. The object feature and the resource attribute feature are input into the similarity calculation layer M131 to determine the similarity 131 between the object feature and the resource attribute feature. Based on the similarity 131, it is determined whether to use the resource as a resource to be recommended.

[0049] In the process of implementing the embodiments of the present disclosure, it is found that the above-mentioned recommendation method can introduce the characteristics of the historical interaction resource sequence of the target object, and establish the historical interest characteristics of the target object through recurrent neural network (RNN), attention mechanism (Attention), etc. However, on the one hand, when the historical interaction resource sequence of the target object is long, more and more interests of different interest categories are involved, which will cause the weight differentiation of the interest characteristics of the target object output by the recommendation model in each interest category to become lower and lower, and then cause the preference in the entire interest space to become more and more vague, so that the recall accuracy of resources related to each interest is reduced. On the other hand, due to the difference in interest frequency, this recommendation method will still cause high-frequency interests to dominate the weight of interest characteristics, resulting in the resources belonging to long-term interests not being effectively recalled.

[0050] In the field of recommendation, positive feedback behavior can refer to the interactive behavior of the target object in the historical period. For example, ad clicks, form filling, order payment, credit application, etc. For different recommendation scenarios, the cycle of continuous positive feedback behavior generated by the target object varies. For example, in the daily commodity scenario, the positive feedback accounts for a high proportion; on the contrary, in the financial credit, paid training and other scenarios, the positive feedback accounts for a low proportion.

[0051] Using traditional recommendation methods, it is impossible to effectively discover the long-term interests of the target object during the recall phase. As a result, the recall results are concentrated in resources belonging to high-frequency interests, and resources related to long-term interests cannot be effectively recalled. In addition, even if a small number of resources belonging to long-term interests are recalled, the recall accuracy is relatively low. Ultimately, resources related to industries related to long-term interests cannot effectively target audiences, resulting in poor promotion effects and poor user experience.

[0052] The present disclosure provides a recommendation method, a training method for a deep learning model, an apparatus, an electronic device, a storage medium, and a program product.

[0053] According to an embodiment of the present disclosure, a recommendation method is provided, including: determining a historical interaction resource sequence activated by the interest of a target object, wherein the historical interaction resource sequence activated by the interest includes multiple historical interaction resources that the target object has interacted with within a historical period, and the multiple historical interaction resources all belong to resources of the target interest. Based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object, the object features of the target object with respect to the target interest are determined. Multiple candidate resource features with respect to the target interest are determined, wherein the multiple candidate resource features correspond one-to-one to the multiple candidate resources, the candidate resource features are used to characterize the resource attribute features of the candidate resources corresponding to the candidate resource features, and the multiple candidate resources all belong to resources of the target interest. Based on the multiple candidate resource features and the object features, a resource to be recommended is determined from the multiple candidate resources.

[0054] According to the embodiments of the present disclosure, a resource recall algorithm based on historical interaction resource serialization modeling of interest activation is provided, which performs serialization modeling of historical interaction resource sequences of target objects in historical periods, such as the entire life cycle, and uses interest activation to process historical interaction resource sequences, so that the historical interaction resource sequences activated by interest take into account both high-frequency and low-frequency interests, and realize high-precision recall of all candidate resources. It can significantly improve the accuracy of resource recommendations, improve user experience, and improve recommendation efficiency.

[0055] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure and application of user personal information involved comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0056] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0057] Figure 2 An exemplary system architecture to which the recommendation method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown.

[0058] It should be noted that Figure 2 The examples shown are only examples of system architectures to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. For example, in another embodiment, the exemplary system architecture to which the recommendation method and apparatus can be applied may include a terminal device, but the terminal device may implement the recommendation method and apparatus provided by the embodiments of the present disclosure without interacting with the server.

[0059] like Figure 2 As shown, the system architecture 200 according to this embodiment may include terminal devices 201, 202, 203, a network 204 and a server 205. The network 204 is used to provide a medium for communication links between the terminal devices 201, 202, 203 and the server 205. The network 204 may include various connection types, such as wired and / or wireless communication links, etc.

[0060] The user can use the terminal devices 201, 202, 203 to interact with the server 205 through the network 204 to receive or send messages, etc. Various communication client applications can be installed on the terminal devices 201, 202, 203, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (only as examples).

[0061] The terminal devices 201 , 202 , and 203 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0062] The server 205 may be a server that provides various services, such as a background management server (only an example) that provides support for the content browsed by the user using the terminal devices 201, 202, and 203. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to the user request) to the terminal device.

[0063] It should be noted that the recommendation method provided in the embodiment of the present disclosure can generally be executed by the terminal device 201, 202, or 203. Accordingly, the recommendation apparatus provided in the embodiment of the present disclosure can also be set in the terminal device 201, 202, or 203.

[0064] Alternatively, the recommendation method provided in the embodiment of the present disclosure may also be generally performed by the server 205. Accordingly, the recommendation device provided in the embodiment of the present disclosure may generally be arranged in the server 205. The recommendation method provided in the embodiment of the present disclosure may also be performed by a server or server cluster that is different from the server 205 and can communicate with the terminal devices 201, 202, 203 and / or the server 205. Accordingly, the recommendation device provided in the embodiment of the present disclosure may also be arranged in a server or server cluster that is different from the server 205 and can communicate with the terminal devices 201, 202, 203 and / or the server 205.

[0065] For example, when a user browses a video online, the terminal devices 201, 202, and 203 can obtain the user's login request or browsing request, and send the login request or browsing request to the server 205, and the server 205 responds to the login request or browsing request to determine the historical interactive resource sequence activated by the user's interest. Based on the historical interactive resource sequence activated by the interest and the user's object attribute information, determine the object characteristics of the user with respect to the target interest. Determine multiple candidate resource characteristics with respect to the target interest. Based on multiple candidate resource characteristics and object characteristics, determine the resource to be recommended from multiple candidate resources. And send the resource to be recommended to the terminal devices 201, 202, and 203 to achieve the purpose of recommending the resource to be recommended to the user. Or a server or server cluster that can communicate with the terminal devices 201, 202, 203 and / or the server 205 responds to the user's login request or browsing request, analyzes the user, and finally achieves the purpose of recommending the resource to be recommended to the user.

[0066] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0067] It should be noted that the sequence numbers of the operations in the following method are only used as representations of the operations for the purpose of description, and should not be regarded as representing the execution order of the operations. Unless explicitly stated, the method does not need to be executed completely in the order shown.

[0068] Figure 3 The flowchart of the recommendation method according to the embodiment of the present disclosure is schematically shown.

[0069] like Figure 3 As shown, the method 300 includes operations S310 to S340.

[0070] In operation S310 , a historical interaction resource sequence activated by the interest of a target object is determined.

[0071] In operation S320, an object feature of the target object with respect to the target interest is determined based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object.

[0072] In operation S330 , a plurality of candidate resource features related to the target interest are determined.

[0073] In operation S340 , a resource to be recommended is determined from the plurality of candidate resources based on the plurality of candidate resource features and the object features.

[0074] According to an embodiment of the present disclosure, the interest-activated historical interaction resource sequence includes a plurality of historical interaction resources with which the target object has interacted within a historical period, and the plurality of historical interaction resources all belong to resources of the target interest.

[0075] According to an embodiment of the present disclosure, the target object may refer to a user who uses a terminal device to obtain network resources.

[0076] According to an embodiment of the present disclosure, resources may be used to represent content displayed to a target object, including but not limited to: short videos, long videos, news, advertisements and other resources.

[0077] According to an embodiment of the present disclosure, the historical interaction resources that have been interacted with may refer to resources in which the target object has performed interaction behaviors in a historical period. The interaction behaviors may refer to behaviors such as browsing, collecting, clicking, purchasing, filling out a form, or applying for credit. Multiple historical interaction resources may be sorted according to the interaction time sequence to obtain a historical interaction resource sequence.

[0078] According to an embodiment of the present disclosure, the interest-activated historical interaction resource sequence may be obtained by interest-activating the historical interaction resource sequence of the target object. For example, based on the interest category of the interest, at least one target historical interaction resource belonging to the same interest is determined from the historical interaction resource sequence. Based on the at least one target historical interaction resource, the interest-activated historical interaction resource sequence is obtained.

[0079] For example, the historical interaction resource sequence of the target object may include {Ad1, Ad2, Ad3, Ad4, . . . , Ad n}, Ad n Indicates the nth advertising resource that the target object has interacted with in the historical period. Among them, historical interaction resources Ad1, Ad2, and Ad3 belong to advertising resources whose target interest is the electronic product interest category. Historical interaction resources Ad4 and Ad5 belong to advertising resources whose target interest is the daily necessities interest category. Based on the historical interaction resource sequence of the target object, the historical interaction resource sequence {Ad1, Ad2, Ad3} whose target interest is the electronic product interest category and the historical interaction resource sequence {Ad4, Ad5} whose target interest is the daily necessities interest category are obtained.

[0080] According to an embodiment of the present disclosure, a plurality of candidate resource features correspond one-to-one to a plurality of candidate resources, the candidate resource features are used to characterize resource attribute features of the candidate resources corresponding to the candidate resource features, and the plurality of candidate resources are all resources of target interest.

[0081] According to an embodiment of the present disclosure, object features are obtained based on a sequence of historical interactive resources activated by interest, and the features represented include resource features of historical interactive resources belonging to the target interest. As a result, the target object can be used as reference data such as object features for screening resources to be recommended in a long historical period, such as a full life cycle period, which are features that are distinguished in terms of interest, and features of other interests are shielded to avoid the mixing of features that are irrelevant to the target interest, thereby ignoring the difference in interest frequency and highlighting the interest preference in the interest space. In addition, multiple candidate resources also belong to resources of the target interest. In this way, the recommended resources and reference data belong to the same interest space, achieving the purpose of interest-based screening. By comparing multiple candidate resource features with object features and determining the resources to be recommended from multiple candidate resources, the resources to be recommended can be resources that belong to the same target interest and are close to the object features of the target object, meeting the personalized needs of the target object, while improving the personalized experience of the target object and improving the recall accuracy.

[0082] According to an embodiment of the present disclosure, when executing Figure 3In operation S310 shown, before determining the historical interaction resource sequence activated by the interest of the target object, the recommendation method may further include: generating a historical interaction resource sequence activated by the interest based on the historical interaction resource sequence.

[0083] According to an embodiment of the present disclosure, generating an interest-activated historical interaction resource sequence based on a historical interaction resource sequence may include: obtaining a historical interaction resource sequence of a target object in a historical period, performing interest activation on the historical interaction resource sequence according to a mapping relationship between interest and resource, and obtaining an interest-activated historical interaction resource sequence.

[0084] According to the embodiments of the present disclosure, the mapping relationship between interests and resources can be used to characterize whether a resource belongs to an interest. For example, advertising resource A about a computer and advertising resource B about a mouse both belong to the interest of the electronic product interest category. Then there is a mapping relationship between advertising resource A and advertising resource B and the interest of the electronic product interest category.

[0085] According to an embodiment of the present disclosure, the mapping relationship between interests and resources can be generated in advance based on statistics of open source data sets, such as generating a mapping relationship list between interests and resources. Based on the mapping relationship list, the mapping relationship between the target interest and the historical interaction resource can be determined. However, it is not limited to this. The mapping relationship between the interest and the resource can also be determined based on the similarity between the interest feature and the historical interaction resource feature.

[0086] According to the embodiments of the present disclosure, interest activation is performed on historical interaction resource sequences according to the mapping relationship between interests and resources to obtain interest-activated historical interaction resource sequences, which can ensure that the interest-activated historical interaction resource sequences ignore the interest frequency and the time period when the historical interaction behaviors occur. Multiple interest-activated historical interaction resource sequences are compatible with both historical interaction resources belonging to low-frequency, long-period interests and high-frequency, short-period interests, so that the interest-activated historical interaction resource sequences contain rich information and are highly accurate.

[0087] According to the embodiments of the present disclosure, Figure 3 Before the operation S310 shown, the recommendation method may further include an operation of obtaining a historical interaction resource sequence of the target object from the starting time to the current time.

[0088] According to the embodiments of the present disclosure, the start time is used to characterize the historical registration time of the target object. For example, the target object uses a terminal device to download an application and logs into the application. However, it is not limited to this. It can also refer to the time when relevant information is filled in and registration is completed. The period between the start time and the current time is taken as the historical period, which can be understood as the full life cycle period of the target object.

[0089] According to an embodiment of the present disclosure, the historical interaction resource sequence of the target object in the historical period is obtained, and based on the historical interaction resource sequence in the historical period, the historical interaction resource sequence of interest activation is determined. While being able to ensure that the historical interaction resource sequence of interest activation is a sequence after the historical interaction resources are differentiated by interest, multiple historical interaction resource sequences of interest activation are compatible with historical interaction resources belonging to low-frequency long-period interests and high-frequency short-period interests. This avoids the influence of interest frequency and the period of occurrence of historical interaction behaviors on the recall operation, and avoids the problem of low-frequency interests not being effectively represented while avoiding interference between multiple interests.

[0090] According to an embodiment of the present disclosure, according to a mapping relationship between interests and resources, interest activation is performed on a historical interaction resource sequence to obtain an interest-activated historical interaction resource sequence, which may include the following operations.

[0091] For example, according to the mapping relationship between interests and resources and according to preset activation weights, interest activation is performed on each historical interaction resource in the historical interaction resource sequence with respect to the target interest to obtain an interest-activated historical interaction resource sequence.

[0092] According to an embodiment of the present disclosure, the activation weight w is preset i The determination method of can be adopted as follows formula (1).

[0093]

[0094] Among them, f(a i ) represents the i-th historical interaction resource a i interest, f(a cand ) represents the target interest, in the i-th historical interaction resource a i If the interest belongs to the target interest, the activation weight w is preset i is 1, otherwise it is 0.

[0095] According to an embodiment of the present disclosure, interest activation is performed on each historical interaction resource in a historical interaction resource sequence regarding a target interest according to a preset activation weight to obtain an interest-activated historical interaction resource sequence, which may include: element-wise multiplying the preset activation weight by each historical interaction resource in the historical interaction resource sequence to obtain an interest-activated historical interaction resource sequence.

[0096] According to an embodiment of the present disclosure, the historical interaction resource sequence activated by interest can be obtained by the following formula (2).

[0097]

[0098] in, represents the historical interaction resource sequence of interest activation, and Au represents the historical interaction resource sequence.

[0099] According to an embodiment of the present disclosure, a preset activation weight is set, and the preset activation weight can be used to highlight the historical interaction resources related to the target interest, thereby improving the recall accuracy. In addition, the preset activation weight is set to a value of 0 or 1, so that the historical interaction resource sequence activated by the interest only includes at least one historical interaction resource belonging to the same interest, and the preset activation weight is equal to 0, indicating that no historical interaction resource of the target object is activated on this interest, and it has no recall significance. The purpose of interest-based analysis is thereby achieved, and interference between multiple historical interaction resources of multiple interests in the historical interaction resource sequence activated by interest is avoided, thereby avoiding the problem that low-frequency interests cannot be effectively represented.

[0100] According to an embodiment of the present disclosure, in response to a request of a target object, a historical interaction resource sequence of the target object can be determined based on identification information of the target object in the request. Based on a mapping relationship between interests and resources, interest activation is performed on the historical interaction resource sequence to obtain an interest-activated historical interaction resource sequence.

[0101] According to an embodiment of the present disclosure, in the process of the terminal device executing the recommendation method provided by the embodiment of the present disclosure, the historical interaction resource sequence of the target object is activated in real time, which will cause a large amount of calculation and be time-consuming, thereby failing to meet the response requirements of the target object.

[0102] According to an optional embodiment of the present disclosure, for example Figure 3 The operation S310 shown, determining a historical interaction resource sequence activated by the interest of the target object, may include: determining a historical interaction resource sequence activated by the interest of the target object from an offline database based on identification information of the target object and interest information of the target interest.

[0103] According to an embodiment of the present disclosure, the interest information of the target interest may include an interest tag or an interest identifier for representing the interest category, as long as it is information of unique attribute information for representing the target interest.

[0104] According to an embodiment of the present disclosure, the identification information of the target object may be information used to uniquely identify the target object and other objects, for example, the identity identification information of the target object.

[0105] According to an embodiment of the present disclosure, the offline database may be generated using multiple interest-activated historical interaction resource sequences. Based on the object identification, interest information and interest-activated historical interaction resource sequences of the object, a vocabulary is generated and stored in the offline database for online query.

[0106] Figure 4A schematic diagram of generating an offline database according to an embodiment of the present disclosure is schematically shown.

[0107] like Figure 4 As shown, for interest Interest1, the historical interaction resource sequence AS-400 of object A400 is activated to obtain the historical interaction resource sequence Interest1-A of interest activation of object A400. The historical interaction resource sequence BS-400 of object B400 is activated to obtain the historical interaction resource sequence Interest1-B of interest activation of object B400. The historical interaction resource sequence NS-400 of object N400 is activated to obtain the historical interaction resource sequence Interest1-N of interest activation of object N.

[0108] For the interest of the interest tag Interest2, the historical interaction resource sequence AS-400 of object A400 is activated to obtain the historical interaction resource sequence Interest2-A of the interest activation of object A400. The historical interaction resource sequence BS-400 of object B400 is activated to obtain the historical interaction resource sequence Interest2-B of the interest activation of object B. The historical interaction resource sequence NS-400 of object N400 is activated to obtain the historical interaction resource sequence Interest2-N of the interest activation of object N400.

[0109] An offline database DB400 can be generated based on the historical interaction resource sequence Interestl-A activated by the interest of object A400, the historical interaction resource sequence Interest1-B activated by the interest of object B400, the historical interaction resource sequence Interest1-N activated by the interest of object N400, the historical interaction resource sequence Interest2-A activated by the interest of object A400, the historical interaction resource sequence Interest2-B activated by the interest of object B400 and the historical interaction resource sequence Interest2-N activated by the interest of object N400.

[0110] According to the embodiments of the present disclosure, the offline database can be stored in the terminal device, and in response to the request of the target object, the historical interaction resource sequence activated by the target object's interest can be directly obtained from the offline database based on the target object's identification information and the target interest information in the request. This improves the processing efficiency, reduces the hardware energy consumption of the terminal device, improves the hardware processing performance of the terminal device, and improves the response processing efficiency.

[0111] According to an exemplary embodiment of the present disclosure, when executing Figure 3Before the operation S310 shown, the recommendation method may further include an operation of determining a mapping relationship between interests and resources.

[0112] According to an embodiment of the present disclosure, a mapping relationship between interests and resources may be determined based on interest features and historical interaction resource features.

[0113] According to an embodiment of the present disclosure, the historical interaction resource feature is used to characterize the resource attribute feature of the historical interaction resource. The interest feature is used to characterize the feature related to the interest information.

[0114] According to an embodiment of the present disclosure, determining the mapping relationship between interest and resource based on interest features and historical interaction resource features may include: determining the similarity between the interest features and historical interaction resource features. Based on the similarity, determining the mapping relationship between interest and resource.

[0115] According to an embodiment of the present disclosure, determining the mapping relationship between an interest and a resource based on similarity may include: when the similarity is greater than or equal to a predetermined similarity threshold, determining the mapping relationship between the interest and the resource as the historical interaction resource belonging to the interest. Conversely, when the similarity is less than the predetermined similarity threshold, determining the mapping relationship between the interest and the resource as the historical interaction resource not belonging to the interest.

[0116] According to the embodiments of the present disclosure, interest features and historical interaction resource features are used to determine the mapping relationship between interests and resources, so that the judgment result can be quantified, thereby improving the accuracy of determining the mapping relationship between interests and resources and improving the processing range, thereby avoiding the problem of being unable to judge due to the fact that the historical interaction resources do not exist in the preset mapping relationship list between interests and resources.

[0117] According to the embodiments of the present disclosure, Figure 3 Before the operation S310 shown, the recommendation method may further include: generating an interest feature of an interest.

[0118] According to an embodiment of the present disclosure, feature extraction may be performed on interest information to obtain interest features. For example, for interest category A, interest information of interest category A is extracted to obtain interest features of interest category A. However, this is not limited to this. Interest features of interest may also be generated, which may include the following operations.

[0119] For example, for each object interaction resource in a plurality of object interaction resources, an object interaction resource feature is obtained based on resource attribute information of the object interaction resource. The object interaction resource is used to characterize a historical interaction resource in which an object has interacted. The plurality of object interaction resource features are clustered to obtain at least one object interaction resource feature cluster. At least one object interaction resource feature cluster corresponds to at least one interest one-to-one. Based on each object interaction resource feature cluster, an interest feature of an interest corresponding to the object interaction resource feature cluster is obtained.

[0120] According to an embodiment of the present disclosure, the object interaction resource may be a historical interaction resource of an object in open source data. The object interaction resources of multiple objects may be collected from the open source data. The larger the amount of data, the richer the object interaction resource feature clusters obtained by clustering.

[0121] According to an embodiment of the present disclosure, obtaining the object interaction resource feature based on the resource attribute information of the object interaction resource may include: extracting the feature of the resource attribute information of the object interaction resource to obtain the object interaction resource feature. Exemplarily, the feature of the resource attribute information of the object interaction resource may be extracted using a graph neural network, but it is not limited to this, as long as it is a network that can extract features.

[0122] According to an embodiment of the present disclosure, a plurality of object interaction resource features can be clustered by a clustering algorithm to obtain at least one object interaction resource feature cluster. The size of each object interaction resource feature cluster can be between 10 5 The larger the amount of data, the more accurate the clustering effect of the object interaction resource feature cluster. Each object interaction resource feature cluster represents an interest.

[0123] According to an embodiment of the present disclosure, obtaining an interest feature of an interest corresponding to each object interaction resource feature cluster based on each object interaction resource feature cluster may include: calculating a mean of the object interaction resource feature cluster to obtain the interest feature of the corresponding interest.

[0124] According to the embodiments of the present disclosure, by performing vector representation on the resource attribute information of the object interaction resources of the object, and on this basis, clustering the object interaction resource features of multiple objects, obtaining the object interaction resource feature clusters after big data statistics, and using the vector mean of the object interaction resource feature clusters to represent the interest feature of the interest, the interest of the abstract concept can be concretized and quantified, thereby improving the accuracy and scope of application of the interest feature.

[0125] Figure 5 A schematic diagram of generating interest features according to an embodiment of the present disclosure is schematically shown.

[0126] like Figure 5As shown, an object interaction resource set 510 is collected, such as Ad1, Ad2, Ad3, ..., Adn. The resource attribute information of each of the multiple object interaction resources in the object interaction resource set 510 is input into the graph neural network M510 to obtain an object interaction resource feature set 520. The object interaction resource feature set 520 is clustered to obtain multiple object interaction resource feature subsets, and each object interaction resource feature subset is used as an object interaction resource feature cluster, that is, multiple object interaction resource feature clusters 530 are obtained. Each object interaction resource feature cluster belongs to an interest. For example, the object interaction resource feature clusters of Ad1, Ad4, ..., Adj belong to interest Interest1. The object interaction resource feature clusters of Ad2, Ad3, ..., Adi belong to interest Interestk. For each interest, the object interaction resource feature clusters are averaged to obtain the interest feature 540 of the interest. For example, interest feature 1 of interest Interest1 and interest feature k of interest Interestk are obtained.

[0127] Figure 6 A schematic diagram of a training method for a graph neural network according to an embodiment of the present disclosure is schematically shown.

[0128] like Figure 6 As shown, the object interaction resource sequences 610 of the multiple objects may be collected, for example, the object interaction resources of the objects that have had historical click behaviors in a historical period may be combined to obtain the object interaction resource sequence.

[0129] like Figure 6 As shown, an interactive resource map 620 can be generated based on multiple object interactive resource sequences 610. The interactive resource map 620 includes multiple nodes, each node represents an object interactive resource, and the edge between two nodes PA and PB represents that object interactive resource A and object interactive resource B have an adjacent relationship in the object interactive resource sequence. The edge is a directed edge, and the arrow of the edge between nodes PA and PB points from node PA to node PB, representing that the historical click behavior of the object is: first click on historical interactive resource A, then click on historical interactive resource B, and vice versa. The edge carries a weight, which indicates the frequency of the click behavior. There may be edges in two directions between two nodes at the same time, but the weights may be different.

[0130] like Figure 6 As shown, based on the directed edges in the interactive resource graph 620 , random walk sampling is performed according to frequency to obtain multiple interactive resource sequences as a training sample set 630 .

[0131] like Figure 6 As shown, the graph neural network M610 is trained using the training sample set 630 to obtain a graph neural network for extracting object interaction resource features of object interaction resources.

[0132] According to an embodiment of the present disclosure, based on the directed edges in the interaction resource graph, random walk sampling is performed according to frequency to obtain multiple interaction resource sequences. The property that multiple interaction resources in the interaction resource sequence belong to the same interest is utilized, and the multiple interaction resource sequences are used as training sample sets to train the graph neural network, which can enable the graph neural network to learn the characteristics of interaction resources of the same interest, and then use the trained graph neural network to extract the object interaction resource characteristics of the object interaction resource as a vector representation on the interest, thereby improving the accuracy of the graph neural network in extracting interaction resource features.

[0133] According to the embodiments of the present disclosure, Figure 3 Operation S320 shown, determining the object characteristics of the target object with respect to the target interest based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object, may include the following operations.

[0134] For example, the historical interaction resource sequence of interest activation is serialized and represented to obtain the interest activation interaction resource feature. Based on the object attribute information of the target object and the interest activation interaction resource feature, the object feature of the target object related to the target interest is obtained.

[0135] According to an embodiment of the present disclosure, serializing and characterizing the historical interaction resource sequence of interest activation to obtain the interest activation interaction resource feature may include: performing serialization feature extraction on the historical interaction resource sequence of interest activation to obtain the interest activation interaction resource feature.

[0136] According to an embodiment of the present disclosure, serialized representation may refer to serialized feature extraction. Specifically, it may refer to determining the cross-features between each historical interaction resource activated by interest in a historical interaction resource sequence activated by interest and other historical interaction resources activated by interest in the historical interaction resource sequence activated by interest, and obtaining a global representation vector of the historical interaction resource activated by interest, thereby strengthening the overall characterization of the target object.

[0137] According to an embodiment of the present disclosure, an embedding layer can be used to vectorize the historical interaction resource sequence activated by interest to obtain a historical interaction resource vector sequence activated by interest. A multi-layer stacked encoder-decoder (Transformer) can be used to serialize and represent the historical interaction resource vector sequence activated by interest to obtain an interest activation interaction resource feature. However, it is not limited to this. A multi-layer stacked decoder (transformer-decoder) can also be used to serialize and represent the historical interaction resource vector sequence activated by interest to obtain an interest activation interaction resource feature.

[0138] According to an embodiment of the present disclosure, a multi-layer stacked decoder is used to serialize and characterize the historical interaction resource vector sequence of interest activation, which can avoid the reverse order effect, thereby making the interest activation interaction resource features accurate and effective.

[0139] According to an embodiment of the present disclosure, the object attribute information (profile) of the target object may include object portrait information of the target object, for example, identity information such as gender, age, etc. or other attribute information of the target object.

[0140] According to an embodiment of the present disclosure, the object attribute information of the target object may be vectorized using an embedding layer to obtain an object attribute information vector of the target object.

[0141] According to an embodiment of the present disclosure, the object attribute information vector and the interest activation interaction resource feature of the target object may be input into the object processing layer to obtain the object feature of the target object related to the target interest.

[0142] According to an embodiment of the present disclosure, the object processing layer may include a DNN network, but is not limited thereto, as long as it is a network that can extract the object attribute information vector of the target object and the features of the interest-activated interaction resource features to obtain the object features of the target object about the target interest.

[0143] According to the embodiments of the present disclosure, the historical interactive resource sequence activated by interest is serialized and characterized to obtain the interactive resource feature activated by interest, which can make the interactive resource feature activated by interest accurate and effective. In addition, based on the object attribute information vector of the target object and the interactive resource feature activated by interest, the object feature of the target object about the target interest is obtained, which can make the feature information contained in the object feature rich and accurate, and improve the recall accuracy of the resource to be recommended obtained based on the object feature.

[0144] According to the embodiments of the present disclosure, Figure 3 The operation S330 shown, determining multiple candidate resource features about the target interest, may include: determining multiple candidate resources corresponding to the target interest. Based on the resource attribute information of each of the multiple candidate resources, determining multiple candidate resource features about the target interest. However, it is not limited to this. Multiple resource features can also be determined based on the resource attribute information of each of the multiple resources. Based on the similarity between the resource features and the interest features, multiple candidate resources corresponding to the target interest and the candidate resource features of each of the multiple candidate resources are determined from the multiple resources.

[0145] According to an embodiment of the present disclosure, resource attribute information of each of a plurality of resources may be input into an advertisement processing module to obtain a plurality of resource features.

[0146] According to an embodiment of the present disclosure, the interest of a resource may be determined based on resource characteristics using the following formula (3).

[0147]

[0148] Among them, in the resource and resource cluster i If the historical interaction resources in are the same, the interest i of the historical interaction resource is taken as the interest f(a) of the resource. If the resource is not in multiple resource clusters, calculate the resource feature W a e a and the interest feature W of the i-th interest i e i The similarity between them is calculated, and multiple similarities are obtained, and the interest with the greatest similarity is taken as the interest of the resource f(a). The resource cluster can represent the resource cluster corresponding to the object interaction resource feature cluster.

[0149] According to an embodiment of the present disclosure, in response to a request of a target object, multiple candidate resources corresponding to the target interest can be determined from multiple resources based on the interest of the resource. The resource can refer to content pre-stored in a database for display to the object.

[0150] According to an embodiment of the present disclosure, in the process of the terminal device executing the recommendation method provided by the embodiment of the present disclosure, feature extraction of candidate resources is performed in real time, and multiple candidate resource features regarding the target interest are determined, which requires large amount of calculation and is time-consuming, thus failing to meet the response requirements of the target object.

[0151] According to an optional embodiment of the present disclosure, for example Figure 3 The operation S330 shown, determining multiple candidate resource features about the target interest, may include: determining a target resource index corresponding to the target interest from multiple resource indexes. Based on the target resource index, determining multiple candidate resource features matching the interest feature of the target interest.

[0152] According to an embodiment of the present disclosure, the mapping relationship between the candidate resource features and the interests can be obtained in advance in an offline state. Based on the mapping relationship between the candidate resource features and the interests, a resource index is generated. Exemplarily, an ANN index (Approximate Nearest Neighbor) can be generated. Using the ANN index, it is convenient to search for feature vectors.

[0153] According to the embodiments of the present disclosure, it is found in the actual implementation process that the multiple candidate resource features corresponding to multiple interests of different interest categories are different from each other in feature vector representation. Therefore, based on the interest category of the interest, a resource index corresponding to the interest can be generated to improve the indexing efficiency, thereby improving the efficiency and accuracy of determining multiple candidate resource features about the target interest.

[0154] According to an embodiment of the present disclosure, when executing Figure 3 In operation S330 shown, before determining a plurality of candidate resource features related to the target interest, the recommendation method may further include the following operations.

[0155] For example, for each of the multiple interests, a resource feature set is determined based on the resource attribute information of each of the multiple resources. Based on the multiple resource feature sets, multiple resource indexes are generated. The multiple resource indexes correspond to the multiple interests one by one.

[0156] According to an embodiment of the present disclosure, a resource may refer to content pre-stored in a database for displaying or recommending to an object. Resource attribute information of the resource may be input into an embedding layer to obtain a resource attribute information vector. The resource attribute information vector is input into a resource processing layer to obtain resource features. The resource processing layer may include a deep neural network.

[0157] According to the embodiments of the present disclosure, by using an offline processing method, a resource index can be quickly generated in advance, which is convenient for responding to object requests during the online process, improving the user experience while reducing the hardware requirements of the terminal device and improving the hardware operating performance.

[0158] According to the embodiments of the present disclosure, Figure 3 The operation S340 shown, determining the resource to be recommended from the multiple candidate resources based on the multiple candidate resource features and the object features, may include: determining the similarity between the object feature and each candidate resource feature to obtain multiple similarities. Based on the multiple similarities, determining the resource to be recommended from the multiple candidate resources.

[0159] According to an embodiment of the present disclosure, a cosine similarity calculation method may be used to determine the similarity between an object feature and each candidate resource feature, but this is not limited thereto. A Euclidean distance calculation method may also be used, as long as the calculation method can be used to evaluate the similarity between two features.

[0160] According to an embodiment of the present disclosure, determining a resource to be recommended from a plurality of candidate resources based on a plurality of similarities may include: determining K candidate resources with the highest similarity as the resource to be recommended. K may be an integer of 1 or greater than 1.

[0161] According to the embodiments of the present disclosure, based on similarity, a resource to be recommended is determined from a plurality of candidate resources, which can improve the recall accuracy while simplifying the calculation method and improving the processing efficiency.

[0162] According to an embodiment of the present disclosure, the resource to be recommended may include multiple resources. The interests of the multiple resources to be recommended are different, but it is not limited thereto, and the interests of the multiple resources to be recommended may be the same.

[0163] According to an embodiment of the present disclosure, when executing Figure 3 After operation S340, the recommendation method may further include operations of: sorting the multiple resources to be recommended according to the similarity to obtain a sorting result, and recommending the multiple resources to be recommended according to the sorting result.

[0164] According to an embodiment of the present disclosure, similarity can be used to characterize the target object's attention or interest in the recommended resource. By sorting and recommending multiple recommended resources according to similarity, the priority display order of the recommended resources can be closer to the target object's expectations or personalized needs.

[0165] According to an embodiment of the present disclosure, the recommendation method can be adapted to the display interface of the terminal device, and multiple resources to be recommended can be displayed in the form of a drop-down list. The multiple resources to be recommended can be sorted and displayed according to similarity. The speed at which the target object preferentially obtains the resources that the target object expects or is interested in can be increased, thereby improving processing efficiency.

[0166] Figure 7 The figure schematically shows a recommended method according to another embodiment of the present disclosure.

[0167] like Figure 7 As shown, in online processing, the terminal device responds to a request of the target object 710, such as a login request, and based on the object identifier of the target object, obtains a historical interaction resource sequence 730 of multiple interest activations of the target object from an offline database 720 obtained through offline processing. The historical interaction resource sequences 730 of multiple interest activations correspond one-to-one to the multiple target interests. The interest categories of the multiple target interests are different.

[0168] like Figure 7 As shown, based on the multiple target interests, multiple target resource indexes corresponding to the multiple target interests can be determined from the multiple resource indexes 740.

[0169] like Figure 7 As shown, a plurality of candidate resource features 750 corresponding to the target interest are determined from each target resource index.

[0170] like Figure 7As shown, based on the historical interaction resource sequence 730 activated by the interest of the target interest and the object attribute information 760 of the target object, the object feature 770 of the target object with respect to the target interest is determined.

[0171] like Figure 7 As shown, based on multiple candidate resource features 750 and object features 770 of the target interest, a resource to be recommended of the target interest is determined from multiple candidate resources.

[0172] According to an embodiment of the present disclosure, a plurality of resources to be recommended corresponding to a plurality of target interests may be sorted, and the plurality of resources to be recommended may be displayed in the sorted order.

[0173] According to the embodiments of the present disclosure, the recommendation method provided by the embodiments of the present disclosure can effectively capture the interest distribution of the target object throughout its life cycle, improve the recommendation effect of resources, and activate the interest of historical interactive resource sequences for interests of different interest categories, improve the diversity of recall results, and increase the browsing time and forwarding quantity of resources by the target object. In addition, pre-generating some data in an offline processing manner can ensure the operating efficiency of the online processing process of the terminal device, and improve the user experience in terms of processing efficiency and response speed.

[0174] Figure 8 The figure schematically shows a recommendation model according to an embodiment of the present disclosure.

[0175] like Figure 8 As shown, the recommendation model can be used to perform the recommendation method to obtain the resources to be recommended. The recommendation model can include: interest activation layer M810, embedding layer M820, serialization representation layer M830, object processing layer M840, resource processing layer M850 and similarity calculation layer M860.

[0176] like Figure 8 As shown, the historical interaction resource sequence 810 of the target object and the target interest of the candidate resource 840 can be input into the interest activation layer M810 to obtain the interest-activated historical interaction resource sequence 820. The interest-activated historical interaction resource sequence 820 is input into the embedding layer M820 to obtain the interest-activated historical interaction resource vector sequence.

[0177] like Figure 8 As shown, the historical interaction resource vector sequence of interest activation can be input into the serialization representation layer M830 to obtain the interest activation interaction resource features.

[0178] like Figure 8 As shown, the object attribute information 830 of the target object can be input into the embedding layer M820 for vectorization processing to obtain the object attribute information vector of the target object.

[0179] like Figure 8 As shown, the object attribute information vector of the target object and the interest activation interaction resource feature can be input into the object processing layer M840 to obtain the object feature of the target object related to the target interest.

[0180] like Figure 8 As shown, the resource attribute information of the candidate resource 840 can be input into the embedding layer M820 to obtain a resource attribute information vector.

[0181] like Figure 8 As shown, the resource attribute information vector can be input into the resource processing layer M850 to obtain resource characteristics.

[0182] like Figure 8 As shown, the object feature and each candidate resource feature can be input into the similarity calculation layer M860 to obtain the similarity. Based on the similarity, it is determined whether the candidate resource is a resource to be recommended.

[0183] According to the embodiments of the present disclosure, the following can be used: Fig. 9 The training method of the deep learning model shown in the figure is used to train the deep learning model to obtain Figure 8 The recommended model shown. For example Figure 8 The recommendation model shown executes the recommendation method provided by the embodiment of the present disclosure.

[0184] Fig. 9 A flowchart of a deep learning model training method according to an embodiment of the present disclosure is schematically shown.

[0185] like Fig. 9 As shown, the method includes operations S910 to S950.

[0186] In operation S910 , a sample historical interaction resource sequence activated by the interest of a sample target object is determined.

[0187] In operation S920, based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object, a sample object feature of the sample target object with respect to the sample target interest is determined.

[0188] In operation S930 , a target sample history interaction resource is determined from a sequence of sample history interaction resources activated by the interest of the sample target object.

[0189] In operation S940, sample resource characteristics are obtained based on sample resource attribute information of the target sample historical interaction resource.

[0190] In operation S950, based on the sample object features and the sample resource features, the parameters of the deep learning model are adjusted to complete the training.

[0191] According to an embodiment of the present disclosure, the interest-activated sample history interaction resource sequence includes a plurality of sample history interaction resources that the sample target object has interacted with within a sample history period, and the plurality of sample history interaction resources are all resources of interest to the sample target.

[0192] According to the embodiments of the present disclosure, Fig. 9 The technical terms involved in the training method of the deep learning model shown in the figure are as follows: Figure 3 The technical terms involved in the recommendation method shown, such as the historical interaction resource sequence of interest activation, only differ in the addition of samples. The main purpose is to distinguish the differences in technical solutions, but the interpretations are similar and will not be repeated here.

[0193] According to an embodiment of the present disclosure, adjusting the parameters of a deep learning model based on sample object features and sample resource features may include: inputting the sample object features and sample resource features into a loss function to obtain a loss value. Adjusting the parameters of the deep learning model based on the loss value until the loss value converges or reaches a predetermined parameter adjustment round threshold. The deep learning model that has stopped training is used as the target deep learning model to complete the training.

[0194] According to an embodiment of the present disclosure, the loss function may be a cross entropy loss function, but is not limited thereto, as long as it is a loss function that can use sample object features and sample resource features as references to adjust the parameters of the deep learning model.

[0195] According to an embodiment of the present disclosure, the sample history interaction resources of the sample target object are activated according to interests, so that the interest-activated sample history interaction resource sequence is compatible with the sample short-term interests (e.g., high-frequency interests) and sample long-term interests (e.g., low-frequency interests) of the sample target object. In this way, the sample target object can have sample object characteristics that are distinguished in terms of interests during a sample history period, such as a full life cycle period. In addition, the reference object sample object characteristics for adjusting the parameters of the deep learning model are standard reference objects that ignore interest frequency differences and highlight interest preferences in the sample interest space. In this way, the training accuracy of the model can be improved while improving the training efficiency of the model.

[0196] According to an embodiment of the present disclosure, when executing Fig. 9 In the operation S910 shown, before determining the sample historical interaction resource sequence activated by interest of the sample target object, the training method of the deep learning model may further include: generating a sample historical interaction resource sequence activated by interest based on the sample historical interaction resource sequence.

[0197] According to an embodiment of the present disclosure, generating an interest-activated sample history interaction resource sequence based on a sample history interaction resource sequence may include: obtaining a sample history interaction resource sequence of a sample target object within a sample history period. According to a mapping relationship between sample interests and resources, interest activation is performed on the sample history interaction resource sequence to obtain an interest-activated sample history interaction resource sequence.

[0198] According to the embodiments of the present disclosure, interest activation is performed on the sample historical interaction resource sequence according to the mapping relationship between sample interests and resources to obtain the interest-activated sample historical interaction resource sequence, which can ensure that the interest-activated sample historical interaction resource sequence ignores the interest frequency and the time period of occurrence of the sample historical interaction behavior, and is compatible with the sample long-term interest and the sample short-term interest, so that the interest-activated sample historical interaction resource sequence contains rich information and high accuracy.

[0199] According to an embodiment of the present disclosure, according to a mapping relationship between sample interests and resources, interest activation is performed on a sample history interaction resource sequence to obtain an interest-activated sample history interaction resource sequence, which may include the following operations.

[0200] For example, according to the mapping relationship between sample interests and resources and according to the preset activation weight of the sample, each sample historical interaction resource in the sample historical interaction resource sequence is activated with respect to the sample target interest to obtain an interest-activated sample historical interaction resource sequence.

[0201] According to the embodiments of the present disclosure, by setting the sample preset activation weight, the sample preset activation weight can be used to highlight the sample history interaction resources related to the sample target interest, thereby improving the recall accuracy. In addition, the sample preset activation weight is set to a value of 0 or 1, so that the interest-activated sample history interaction resource sequence only includes at least one sample history interaction resource belonging to the same sample interest, and the sample preset activation weight is equal to 0, indicating that no sample history interaction resource is activated for the sample target object on the sample interest, and it has no recall significance. The purpose of sample interest analysis is thereby achieved, and interference between multiple sample history interaction resources of multiple sample interests in the interest-activated sample history interaction resource sequence is avoided, thereby avoiding the problem that low-frequency interests cannot be effectively represented.

[0202] According to the embodiments of the present disclosure, Fig. 9 The operation S910 shown, determining a sample historical interaction resource sequence activated by the interest of the sample target object, may include: determining a sample historical interaction resource sequence activated by the interest of the sample target object from a sample offline database based on sample identification information of the sample target object and sample interest information of the sample target interest. The sample offline database is generated using a plurality of sample historical interaction resource sequences activated by interest.

[0203] According to the embodiments of the present disclosure, the sample history interaction resource sequence activated by the sample target object's interest is directly obtained from the sample offline database, thereby improving the processing efficiency, reducing the hardware energy consumption, improving the hardware processing performance, and improving the training efficiency.

[0204] According to an embodiment of the present disclosure, when executing Fig. 9 In the operation S910 shown, before determining the sample historical interaction resource sequence activated by the interest of the sample target object, the training method of the deep learning model may further include the following operations.

[0205] For example, based on the sample interest feature and the sample historical interaction resource feature, the mapping relationship between the sample interest and the resource is determined. The sample historical interaction resource feature is used to characterize the sample resource attribute feature of the sample historical interaction resource, and the sample interest feature is used to characterize the feature related to the sample interest information.

[0206] According to the embodiments of the present disclosure, sample interest features and sample history interaction resource features are used to determine the mapping relationship between sample interests and resources, so that the judgment result can be quantified, thereby improving the accuracy of determining the mapping relationship between sample interests and resources while increasing the processing range, thereby avoiding the problem of being unable to judge due to the fact that the sample history interaction resources do not exist in the preset mapping relationship list between sample interests and resources.

[0207] According to the embodiments of the present disclosure, Fig. 9 Before the operation S910 shown, the training method of the deep learning model may further include: generating sample interest features of the sample interest.

[0208] For example, for each sample object interaction resource in the plurality of sample object interaction resources, a sample object interaction resource feature is obtained based on the sample resource attribute information of the sample object interaction resource. The sample object interaction resource is used to characterize the sample historical interaction resource with which the sample object has interacted.

[0209] Clustering multiple sample object interaction resource features to obtain at least one sample object interaction resource feature cluster. The at least one sample object interaction resource feature cluster corresponds to the at least one sample interest in a one-to-one manner.

[0210] Based on each sample object interaction resource feature cluster, a sample interest feature of the sample interest corresponding to the sample object interaction resource feature cluster is obtained.

[0211] According to the embodiments of the present disclosure, by performing vector representation on the sample resource attribute information of the sample object interaction resources of the sample object, and on this basis, clustering the sample object interaction resource features of multiple sample objects, the sample object interaction resource feature clusters after big data statistics are obtained, and the vector mean of the sample object interaction resource feature clusters is used to represent the interest feature of the interest, which can concretize the interest of the abstract concept and quantify it, thereby improving the accuracy and scope of application of the sample interest feature.

[0212] According to the embodiments of the present disclosure, Fig. 9 Operation S920 shown, determining the sample object features of the sample target object for the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object, may include:

[0213] The sample historical interaction resource sequence of interest activation is serialized and represented to obtain the sample interest activation interaction resource features.

[0214] Based on the sample object attribute information of the sample target object and the sample interest activation interaction resource feature, the sample object feature of the sample target object related to the sample target interest is obtained.

[0215] According to the embodiments of the present disclosure, the sample historical interaction resource sequence activated by interest is serialized and characterized to obtain the sample interest activation interaction resource feature, which can make the sample interest activation interaction resource feature accurate and effective. In addition, based on the sample object attribute information vector of the sample target object and the sample interest activation interaction resource feature, the sample object feature about the sample target interest of the sample target object is obtained, which can make the feature information contained in the sample object feature rich and accurate, and improve the training accuracy and efficiency based on the sample object feature.

[0216] Fig.10 The block diagram of a recommendation device according to an embodiment of the present disclosure is schematically shown.

[0217] like Fig.10 As shown, the recommendation device 1000 of this embodiment includes a resource sequence determination module 1010 , an object feature determination module 1020 , a candidate resource feature determination module 1030 and a to-be-recommended resource determination module 1040 .

[0218] The resource sequence determination module 1010 is used to determine the historical interaction resource sequence activated by the interest of the target object. The historical interaction resource sequence activated by the interest includes multiple historical interaction resources that the target object has interacted with in the historical period, and the multiple historical interaction resources are all resources of the target interest.

[0219] The object feature determination module 1020 is used to determine the object feature of the target object with respect to the target interest based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object.

[0220] The candidate resource feature determination module 1030 is used to determine multiple candidate resource features about the target interest. The multiple candidate resource features correspond to multiple candidate resources one by one, and the candidate resource features are used to characterize the resource attribute features of the candidate resources corresponding to the candidate resource features, and the multiple candidate resources are all resources of the target interest.

[0221] The to-be-recommended resource determination module 1040 is used to determine a to-be-recommended resource from a plurality of candidate resources based on a plurality of candidate resource features and object features.

[0222] According to an embodiment of the present disclosure, the resource sequence determination module 1010 includes: a resource sequence determination unit.

[0223] The resource sequence determination unit is used to determine the historical interaction resource sequence activated by the interest of the target object from the offline database based on the identification information of the target object and the interest information of the target interest. The offline database is generated using multiple historical interaction resource sequences activated by the interest.

[0224] According to an embodiment of the present disclosure, the recommendation device further includes: a resource sequence acquisition module and an interest activation module.

[0225] The resource sequence acquisition module is used to obtain the historical interaction resource sequence of the target object within the historical period.

[0226] The interest activation module is used to activate the interest of the historical interaction resource sequence according to the mapping relationship between the interest and the resource, and obtain the interest-activated historical interaction resource sequence.

[0227] According to an embodiment of the present disclosure, the interest activation module includes: an interest activation unit.

[0228] The interest activation unit is used to activate the interest of each historical interaction resource in the historical interaction resource sequence about the target interest according to the mapping relationship between the interest and the resource and the preset activation weight, so as to obtain the interest-activated historical interaction resource sequence.

[0229] According to an embodiment of the present disclosure, the recommendation device further includes: a mapping relationship determination module.

[0230] The mapping relationship determination module is used to determine the mapping relationship between interest and resource based on interest features and historical interaction resource features. The historical interaction resource features are used to characterize resource attribute features of historical interaction resources, and the interest features are used to characterize features related to interest information.

[0231] According to an embodiment of the present disclosure, the recommendation device further includes: an interactive resource feature determination module, an interactive resource feature clustering module, and an interest feature determination module.

[0232] The interactive resource feature determination module is used to obtain the object interaction resource feature for each object interaction resource among the multiple object interaction resources based on the resource attribute information of the object interaction resource. The object interaction resource is used to represent the historical interaction resource that the object has interacted with.

[0233] The interactive resource feature clustering module is used to cluster multiple object interactive resource features to obtain at least one object interactive resource feature cluster. The at least one object interactive resource feature cluster corresponds to at least one interest.

[0234] The interest feature determination module is used to obtain the interest feature of the interest corresponding to the object interaction resource feature cluster based on each object interaction resource feature cluster.

[0235] According to an embodiment of the present disclosure, the object feature determination module 1020 includes: an interactive resource sequence representation unit and an object feature determination unit.

[0236] The interactive resource sequence representation unit is used to serialize and represent the historical interactive resource sequence of interest activation to obtain the interest activation interactive resource features.

[0237] The object feature determination unit is used to obtain the object feature of the target object related to the target interest based on the object attribute information of the target object and the interest activation interaction resource feature.

[0238] According to an embodiment of the present disclosure, the candidate resource feature determination module 1030 includes: a target resource index determination unit and a candidate resource feature determination unit.

[0239] The target resource index determining unit is used to determine a target resource index corresponding to a target interest from a plurality of resource indexes.

[0240] The candidate resource feature determination unit is used to determine a plurality of candidate resource features that match the interest feature of the target interest based on the target resource index.

[0241] According to an embodiment of the present disclosure, the recommendation device further includes: a resource feature set determination module and a resource index generation module.

[0242] The resource feature set determination module is used to determine a resource feature set for each of the multiple interests based on the resource attribute information of each of the multiple resources.

[0243] The resource index generation module is used to generate multiple resource indexes based on multiple resource feature sets. The multiple resource indexes correspond to the multiple interests one by one.

[0244] According to an embodiment of the present disclosure, the to-be-recommended resource determination module 1040 includes: a similarity determination unit and a to-be-recommended resource determination unit.

[0245] The similarity determination unit is used to determine the similarity between the object feature and each candidate resource feature to obtain multiple similarities.

[0246] The to-be-recommended resource determining unit is used to determine the to-be-recommended resource from a plurality of candidate resources based on a plurality of similarities.

[0247] According to an embodiment of the present disclosure, the resources to be recommended include multiple resources.

[0248] According to an embodiment of the present disclosure, the recommendation device further includes: a sorting module and a recommendation module.

[0249] The sorting module is used to sort multiple resources to be recommended according to similarity to obtain a sorting result.

[0250] The recommendation module is used to recommend multiple resources to be recommended according to the sorting results.

[0251] According to an embodiment of the present disclosure, the recommendation device further includes: a resource sequence acquisition module.

[0252] The resource sequence acquisition module is used to acquire the historical interaction resource sequence of the target object from the start time to the current time. The start time is used to represent the historical registration time of the target object.

[0253] Fig.11 A block diagram of a deep learning model training device according to an embodiment of the present disclosure is schematically shown.

[0254] like Fig.11 As shown, the training device 1100 of the deep learning model includes: a sample resource sequence determination module 1110, a sample object feature determination module 1120, a sample interaction resource determination module 1130, a sample resource feature determination module 1140 and a parameter adjustment module 1150.

[0255] The sample resource sequence determination module 1110 is used to determine the sample historical interaction resource sequence activated by the sample target object's interest. The sample historical interaction resource sequence activated by the interest includes multiple sample historical interaction resources that the sample target object has interacted with in the sample history period, and the multiple sample historical interaction resources are all resources of the sample target's interest.

[0256] The sample object feature determination module 1120 is used to determine the sample object features of the sample target object with respect to the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object.

[0257] The sample interaction resource determination module 1130 is used to determine a target sample history interaction resource from a sequence of sample history interaction resources activated by the interest of the sample target object.

[0258] The sample resource feature determination module 1140 is used to obtain the sample resource feature based on the sample resource attribute information of the target sample historical interaction resource.

[0259] The parameter adjustment module 11 50 is used to adjust the parameters of the deep learning model based on the sample object characteristics and the sample resource characteristics to complete the training.

[0260] According to an embodiment of the present disclosure, the sample resource sequence determination module includes: a sample resource sequence determination unit.

[0261] The sample resource sequence determination unit is used to determine the sample historical interaction resource sequence activated by the sample target object's interest from the sample offline database based on the sample identification information of the sample target object and the sample interest information of the sample target interest. The sample offline database is generated using multiple sample historical interaction resource sequences activated by interest.

[0262] According to an embodiment of the present disclosure, the training device of the deep learning model also includes: a sample resource sequence acquisition module and a sample interest activation module.

[0263] The sample resource sequence acquisition module is used to obtain the sample historical interaction resource sequence of the sample target object within the sample historical period.

[0264] The sample interest activation module is used to activate the sample historical interaction resource sequence according to the mapping relationship between the sample interest and the resource, and obtain the sample historical interaction resource sequence with interest activation.

[0265] According to an embodiment of the present disclosure, the sample interest activation module includes: a sample interest activation unit.

[0266] The sample interest activation unit is used to activate the interest of each sample historical interaction resource in the sample historical interaction resource sequence about the sample target interest according to the mapping relationship between the sample interest and the resource and the preset activation weight of the sample, so as to obtain the sample historical interaction resource sequence with interest activation.

[0267] According to an embodiment of the present disclosure, the training device of the deep learning model also includes: a sample mapping relationship determination module.

[0268] The sample mapping relationship determination module is used to determine the mapping relationship between sample interests and resources based on sample interest features and sample historical interaction resource features. The sample historical interaction resource features are used to characterize sample resource attribute features of sample historical interaction resources, and the sample interest features are used to characterize features related to sample interest information.

[0269] According to an embodiment of the present disclosure, the training device of the deep learning model also includes: a sample interaction resource feature determination module.

[0270] The sample interaction resource feature determination module is used to obtain the sample object interaction resource feature for each sample object interaction resource among the multiple sample object interaction resources based on the sample resource attribute information of the sample object interaction resource. The sample object interaction resource is used to characterize the sample historical interaction resource with which the sample object has interacted.

[0271] The sample interaction resource feature clustering module is used to cluster multiple sample object interaction resource features to obtain at least one sample object interaction resource feature cluster. The at least one sample object interaction resource feature cluster corresponds to at least one sample interest.

[0272] The sample interest feature determination module is used to obtain the sample interest feature of the sample interest corresponding to the sample object interaction resource feature cluster based on each sample object interaction resource feature cluster.

[0273] According to an embodiment of the present disclosure, the sample object feature determination module includes: a sample interaction resource sequence representation unit and a sample object feature determination unit.

[0274] The sample interaction resource sequence representation unit is used to serialize and represent the sample historical interaction resource sequence of interest activation to obtain the sample interest activation interaction resource feature.

[0275] The sample object feature determination unit is used to obtain the sample object feature of the sample target object related to the sample target interest based on the sample object attribute information of the sample target object and the sample interest activation interaction resource feature.

[0276] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0277] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method as in the embodiment of the present disclosure.

[0278] According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute a method as described in an embodiment of the present disclosure.

[0279] According to an embodiment of the present disclosure, a computer program product includes a computer program, and when the computer program is executed by a processor, the method according to the embodiment of the present disclosure is implemented.

[0280] Fig.12 A schematic block diagram of an example electronic device 1200 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0281] like Fig.12 As shown, the device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 into a random access memory (RAM) 1203. In the RAM 1203, various programs and data required for the operation of the device 1200 can also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0282] A number of components in the device 1200 are connected to an input / output (I / O) interface 1205, including: an input unit 1206, such as a keyboard, a mouse, etc.; an output unit 1207, such as various types of displays, speakers, etc.; a storage unit 1208, such as a disk, an optical disk, etc.; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1209 allows the device 1200 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0283] The computing unit 1201 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1201 performs the various methods and processes described above, such as the recommendation method and the training method of the deep learning model. For example, in some embodiments, the recommendation method and the training method of the deep learning model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1200 via the ROM 1202 and / or the communication unit 1209. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the recommendation method and the training method of the deep learning model described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to execute the recommendation method and the training method of the deep learning model in any other appropriate manner (e.g., by means of firmware).

[0284] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0285] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0286] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0287] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0288] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0289] A computer system may include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises through computer programs running on respective computers and having a client-server relationship to each other. The server may be a cloud server, a server in a distributed system, or a server combined with a blockchain.

[0290] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0291] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A recommendation method, comprising: Determine a historical interaction resource sequence activated by the interest of the target object, wherein the historical interaction resource sequence activated by the interest includes a plurality of historical interaction resources that the target object has interacted with in a historical period, and the plurality of historical interaction resources are all resources of the target interest; Determining an object feature of the target object with respect to the target interest based on a historical interaction resource sequence activated by the interest and object attribute information of the target object; Determine a plurality of candidate resource features about the target interest, wherein the plurality of candidate resource features correspond one-to-one to a plurality of candidate resources, the candidate resource features are used to characterize resource attribute features of the candidate resources corresponding to the candidate resource features, and the plurality of candidate resources are all resources of the target interest; and Determining a resource to be recommended from the multiple candidate resources based on the multiple candidate resource characteristics and the object characteristics; The interest-activated historical interaction resource sequence is obtained by performing interest activation on the historical interaction resource sequence according to a mapping relationship between interests and resources.

2. The method according to claim 1, wherein: The determining of the historical interaction resource sequence activated by the interest of the target object comprises: Based on the identification information of the target object and the interest information of the target interest, a historical interaction resource sequence activated by the interest of the target object is determined from an offline database, wherein the offline database is generated using a plurality of historical interaction resource sequences activated by interests.

3. The method according to claim 1, further comprising: Acquire a historical interaction resource sequence of the target object within the historical period; as well as According to the mapping relationship between interests and resources, interest activation is performed on the historical interaction resource sequence to obtain the interest-activated historical interaction resource sequence.

4. The method according to claim 3, wherein: The step of performing interest activation on the historical interaction resource sequence according to the mapping relationship between the interest and the resource to obtain the interest-activated historical interaction resource sequence includes: According to the mapping relationship between interests and resources and in accordance with preset activation weights, interest activation is performed on each historical interaction resource in the historical interaction resource sequence with respect to the target interest to obtain the interest-activated historical interaction resource sequence.

5. The method according to claim 1, further comprising: Based on the interest feature and the historical interaction resource feature, a mapping relationship between the interest and the resource is determined, wherein the historical interaction resource feature is used to characterize the resource attribute feature of the historical interaction resource, and the interest feature is used to characterize the feature related to the interest information.

6. The method according to claim 5, further comprising: For each object interaction resource among the plurality of object interaction resources, obtaining an object interaction resource feature based on resource attribute information of the object interaction resource, wherein the object interaction resource is used to characterize a historical interaction resource with which the object has interacted; Clustering the plurality of object interaction resource features to obtain at least one object interaction resource feature cluster, wherein the at least one object interaction resource feature cluster corresponds one-to-one to at least one interest; and Based on each of the object interaction resource feature clusters, an interest feature of an interest corresponding to the object interaction resource feature cluster is obtained.

7. The method according to any one of claims 1 to 6, wherein: The determining the object feature of the target object with respect to the target interest based on the historical interaction resource sequence activated by the interest and the object attribute information of the target object includes: Serializing and representing the historical interaction resource sequence activated by the interest to obtain an interest-activated interaction resource feature; and The object feature of the target object related to the target interest is obtained based on the object attribute information of the target object and the interest activation interaction resource feature.

8. The method according to any one of claims 1 to 6, wherein: The determining of a plurality of candidate resource features related to the target interest comprises: Determining a target resource index corresponding to the target interest from a plurality of resource indexes; and Based on the target resource index, the plurality of candidate resource features matching the interest feature of the target interest are determined.

9. The method according to claim 8, further comprising: For each of the plurality of interests, determining a resource feature set based on respective resource attribute information of the plurality of resources; as well as Based on the multiple resource feature sets, the multiple resource indexes are generated, wherein the multiple resource indexes correspond to the multiple interests one by one.

10. The method according to any one of claims 1 to 6, wherein: The determining a resource to be recommended from the multiple candidate resources based on the multiple candidate resource features and the object features includes: Determining the similarity between the object feature and each of the candidate resource features to obtain multiple similarities; and Based on the multiple similarities, the resource to be recommended is determined from the multiple candidate resources.

11. The method according to claim 10, wherein: The resources to be recommended include multiple resources; The method further comprises: Sorting the plurality of resources to be recommended according to similarity to obtain a sorting result; and According to the sorting results, multiple resources to be recommended are recommended.

12. The method according to any one of claims 1 to 6, further comprising: A historical interaction resource sequence of the target object from a starting time to a current time is obtained, wherein the starting time is used to represent a historical registration time of the target object.

13. A method for training a deep learning model, comprising: Determine a sample history interaction resource sequence activated by the interest of the sample target object, wherein the sample history interaction resource sequence activated by the interest includes a plurality of sample history interaction resources that the sample target object has interacted with within a sample history period, and the plurality of sample history interaction resources are all resources of the sample target interest; Determine a target sample historical interaction resource from a sequence of sample historical interaction resources activated by the interest of the sample target object; The interest-activated sample history interaction resource sequence, the sample object attribute information of the sample target object, and the target sample history interaction resource are input into the deep learning model, so as to use the deep learning model to perform the following operations: Determining a sample object feature of the sample target object with respect to the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object; Obtaining sample resource characteristics based on the sample resource attribute information of the target sample historical interaction resource; and Based on the sample object characteristics and the sample resource characteristics, adjusting the parameters of the deep learning model to complete the training; The interest-activated sample historical interaction resource sequence is obtained by performing interest activation on the sample historical interaction resource sequence according to a mapping relationship between sample interests and resources.

14. The method according to claim 13, wherein: The determining of the sample historical interaction resource sequence activated by the interest of the sample target object includes: Based on the sample identification information of the sample target object and the sample interest information of the sample target interest, a sample historical interaction resource sequence activated by the interest of the sample target object is determined from a sample offline database, wherein the sample offline database is generated using multiple sample historical interaction resource sequences activated by interests.

15. The method according to claim 13, further comprising: Acquire a sample history interaction resource sequence of the sample target object within the sample history period; as well as According to the mapping relationship between sample interests and resources, interest activation is performed on the sample historical interaction resource sequence to obtain the interest-activated sample historical interaction resource sequence.

16. The method according to claim 15, wherein: The step of performing interest activation on the sample history interaction resource sequence according to the mapping relationship between the sample interest and the resource to obtain the interest activated sample history interaction resource sequence includes: According to the mapping relationship between the sample interest and the resource and according to the preset activation weight of the sample, each sample historical interaction resource in the sample historical interaction resource sequence is activated with respect to the sample target interest to obtain the interest-activated sample historical interaction resource sequence.

17. The method according to claim 13, further comprising: Based on the sample interest feature and the sample history interaction resource feature, the mapping relationship between the sample interest and the resource is determined, wherein the sample history interaction resource feature is used to characterize the sample resource attribute feature of the sample history interaction resource, and the sample interest feature is used to characterize the feature related to the sample interest information.

18. The method according to claim 17, further comprising: For each sample object interaction resource among the multiple sample object interaction resources, based on the sample resource attribute information of the sample object interaction resource, a sample object interaction resource feature is obtained, wherein the sample object interaction resource is used to characterize a sample historical interaction resource with which the sample object has interacted; Clustering the plurality of sample object interaction resource features to obtain at least one sample object interaction resource feature cluster, wherein the at least one sample object interaction resource feature cluster corresponds one-to-one to at least one sample interest; Based on each of the sample object interaction resource feature clusters, a sample interest feature of the sample interest corresponding to the sample object interaction resource feature cluster is obtained.

19. The method according to any one of claims 13 to 18, wherein: The determining, based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object, the sample object feature of the sample target object for the sample target interest comprises: Serializing and characterizing the sample historical interaction resource sequence activated by the interest to obtain a sample interest activation interaction resource feature; and Based on the sample object attribute information of the sample target object and the sample interest activation interaction resource feature, the sample object feature of the sample target object related to the sample target interest is obtained.

20. A recommendation device, comprising: A resource sequence determination module, used to determine a historical interaction resource sequence activated by the interest of the target object, wherein the historical interaction resource sequence activated by the interest includes a plurality of historical interaction resources that the target object has interacted with in a historical period, and the plurality of historical interaction resources are all resources of the target interest; An object feature determination module, configured to determine an object feature of the target object with respect to the target interest based on a historical interaction resource sequence activated by the interest and object attribute information of the target object; A candidate resource feature determination module, used to determine a plurality of candidate resource features about the target interest, wherein the plurality of candidate resource features correspond one-to-one to a plurality of candidate resources, the candidate resource features are used to characterize resource attribute features of the candidate resources corresponding to the candidate resource features, and the plurality of candidate resources are all resources of the target interest; A to-be-recommended resource determination module, configured to determine a to-be-recommended resource from the plurality of candidate resources based on the plurality of candidate resource features and the object features; The interest-activated historical interaction resource sequence is obtained by performing interest activation on the historical interaction resource sequence according to a mapping relationship between interests and resources.

21. The device according to claim 20, wherein: The resource sequence determination module includes: A resource sequence determination unit is used to determine the historical interaction resource sequence activated by the interest of the target object from an offline database based on the identification information of the target object and the interest information of the target interest, wherein the offline database is generated using multiple historical interaction resource sequences activated by interests.

22. The apparatus according to claim 20, further comprising: A resource sequence acquisition module, used to acquire a historical interaction resource sequence of the target object within the historical period; The interest activation module is used to activate the interest of the historical interaction resource sequence according to the mapping relationship between the interest and the resource, so as to obtain the interest-activated historical interaction resource sequence.

23. The device according to claim 22, wherein: The interest activation module includes: The interest activation unit is used to activate the interest of each historical interaction resource in the historical interaction resource sequence about the target interest according to the mapping relationship between interest and resource and the preset activation weight, so as to obtain the interest-activated historical interaction resource sequence.

24. The apparatus according to claim 20, further comprising: A mapping relationship determination module is used to determine the mapping relationship between the interest and the resource based on the interest feature and the historical interaction resource feature, wherein the historical interaction resource feature is used to characterize the resource attribute feature of the historical interaction resource, and the interest feature is used to characterize the feature related to the interest information.

25. The apparatus according to claim 24, further comprising: An interaction resource feature determination module, configured to obtain an object interaction resource feature for each object interaction resource among a plurality of object interaction resources based on resource attribute information of the object interaction resource, wherein the object interaction resource is used to characterize a historical interaction resource with which the object has interacted; An interactive resource feature clustering module, configured to cluster the plurality of object interactive resource features to obtain at least one object interactive resource feature cluster, wherein the at least one object interactive resource feature cluster corresponds one-to-one to at least one interest; The interest feature determination module is used to obtain the interest feature of the interest corresponding to the object interaction resource feature cluster based on each of the object interaction resource feature clusters.

26. The device according to any one of claims 20 to 25, wherein: The object feature determination module includes: An interactive resource sequence characterization unit, used for serializing and characterizing the historical interactive resource sequence activated by the interest to obtain an interactive resource feature activated by the interest; The object feature determination unit is used to obtain the object feature of the target object related to the target interest based on the object attribute information of the target object and the interest activation interaction resource feature.

27. The device according to any one of claims 20 to 25, wherein: The candidate resource feature determination module includes: a target resource index determining unit, configured to determine a target resource index corresponding to the target interest from a plurality of resource indexes; The candidate resource feature determination unit is used to determine the multiple candidate resource features that match the interest feature of the target interest based on the target resource index.

28. The apparatus according to claim 27, further comprising: A resource feature set determination module, configured to determine a resource feature set for each of the plurality of interests based on respective resource attribute information of the plurality of resources; The resource index generating module is used to generate the multiple resource indexes based on the multiple resource feature sets, wherein the multiple resource indexes correspond to the multiple interests one by one.

29. The device according to any one of claims 20 to 25, wherein: The module for determining resources to be recommended includes: A similarity determination unit, used to determine the similarity between the object feature and each of the candidate resource features to obtain multiple similarities; The to-be-recommended resource determining unit is configured to determine the to-be-recommended resource from the multiple candidate resources based on the multiple similarities.

30. The device according to claim 29, wherein: The resources to be recommended include multiple resources; The device also includes: A sorting module, used to sort the plurality of resources to be recommended according to similarity to obtain a sorting result; The recommendation module is used to recommend a plurality of resources to be recommended according to the sorting results.

31. The apparatus according to any one of claims 20 to 25, further comprising: The resource sequence acquisition module is used to acquire the historical interaction resource sequence of the target object from the starting time to the current time, wherein the starting time is used to represent the historical registration time of the target object.

32. A training device for a deep learning model, comprising: A sample resource sequence determination module, used to determine a sample history interaction resource sequence activated by the interest of a sample target object, wherein the sample history interaction resource sequence activated by the interest includes a plurality of sample history interaction resources that the sample target object has interacted with within a sample history period, and the plurality of sample history interaction resources are all resources of the sample target interest; A sample interaction resource determination module, configured to determine a target sample history interaction resource from a sequence of sample history interaction resources activated by the interest of the sample target object; Wherein, the training device is also used for: The interest-activated sample history interaction resource sequence, the sample object attribute information of the sample target object, and the target sample history interaction resource are input into the deep learning model, so as to use the deep learning model to perform the following operations: A sample object feature determination module, configured to determine the sample object feature of the sample target object with respect to the sample target interest based on the sample historical interaction resource sequence activated by the interest and the sample object attribute information of the sample target object; A sample resource feature determination module, configured to obtain sample resource features based on sample resource attribute information of the target sample historical interaction resource; A parameter adjustment module, used to adjust the parameters of the deep learning model based on the sample object characteristics and the sample resource characteristics to complete the training; The interest-activated sample historical interaction resource sequence is obtained by performing interest activation on the sample historical interaction resource sequence according to a mapping relationship between sample interests and resources.

33. The device according to claim 32, wherein: The sample resource sequence determination module includes: A sample resource sequence determination unit is used to determine a sample historical interaction resource sequence activated by the interest of the sample target object from a sample offline database based on the sample identification information of the sample target object and the sample interest information of the sample target interest, wherein the sample offline database is generated using multiple sample historical interaction resource sequences activated by interests.

34. The apparatus of claim 32, further comprising: A sample resource sequence acquisition module, used to acquire a sample history interaction resource sequence of the sample target object within the sample history period; The sample interest activation module is used to activate the sample historical interaction resource sequence according to the mapping relationship between the sample interest and the resource, so as to obtain the sample historical interaction resource sequence with the interest activated.

35. The device according to claim 34, wherein Sample interest activation modules include: A sample interest activation unit is used to activate the interest of each sample historical interaction resource in the sample historical interaction resource sequence about the sample target interest according to the mapping relationship between the sample interest and the resource and the sample preset activation weight, so as to obtain the sample historical interaction resource sequence with the interest activated.

36. The apparatus of claim 32, further comprising: A sample mapping relationship determination module is used to determine the mapping relationship between the sample interest and the resource based on the sample interest feature and the sample historical interaction resource feature, wherein the sample historical interaction resource feature is used to characterize the sample resource attribute feature of the sample historical interaction resource, and the sample interest feature is used to characterize the feature related to the sample interest information.

37. The apparatus of claim 36, further comprising: a sample interaction resource feature determination module, configured to obtain a sample object interaction resource feature for each sample object interaction resource among a plurality of sample object interaction resources based on sample resource attribute information of the sample object interaction resource, wherein the sample object interaction resource is used to characterize a sample historical interaction resource with which the sample object has interacted; A sample interaction resource feature clustering module, configured to cluster the plurality of sample object interaction resource features to obtain at least one sample object interaction resource feature cluster, wherein the at least one sample object interaction resource feature cluster corresponds one-to-one to at least one sample interest; The sample interest feature determination module is used to obtain, based on each of the sample object interaction resource feature clusters, the sample interest feature of the sample interest corresponding to the sample object interaction resource feature cluster.

38. The device according to any one of claims 32 to 37, wherein: The sample object feature determination module includes: A sample interaction resource sequence characterization unit, used to serialize and characterize the sample historical interaction resource sequence activated by the interest, and obtain a sample interest activation interaction resource feature; The sample object feature determination unit is used to obtain the sample object feature of the sample target object related to the sample target interest based on the sample object attribute information of the sample target object and the sample interest activation interaction resource feature.

39. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 19.

40. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 19.

41. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 19.

Images