Sample generation method, model training method, resource display method and device

Abstract

The invention provides a sample generation method, a model training method and a resource display method and device, and relates to the technical field of artificial intelligence, in particular to the fields of deep learning, intelligent recommendation and the like. The specific implementation scheme is as follows: determining a first resource and at least one second resource meeting an association relationship with the first resource; wherein the first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment; determining an additional duration for the second resource according to a time interval between the first moment and the second moment and a browsing duration of the first object for the second resource; determining a reference duration according to the browsing duration of the first object for the first resource and the respective additional duration of the at least one second resource; a training sample is generated, the training sample comprises the resource attribute of the first resource and the object attribute of the first object, and the label of the training sample comprises the reference duration.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the fields of deep learning and intelligent recommendation. More specifically, the present disclosure provides a sample generation method, a deep learning model training method, a resource display method, a sample generation device, a deep learning model training device, a resource display device, an electronic device, a storage medium, and a computer program product. Background Art

[0002] Recommendation systems usually use algorithms to determine whether resources meet the user's personal preferences and recommend resources that meet the preferences to the user, thereby improving the user experience. Summary of the Invention

[0003] The present disclosure provides a sample generation method, a deep learning model training method, a resource display method, a sample generation device, a deep learning model training device, a resource display device, an electronic device, a storage medium, and a computer program product.

[0004] According to one aspect of the present disclosure, a sample generation method is provided, including: determining a first resource and at least one second resource that has an association relationship with the first resource; wherein the first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment; determining an additional duration for the second resource based on the time interval between the first moment and the second moment, and the browsing duration of the first object for the second resource; determining a reference duration based on the browsing duration of the first object for the first resource and the additional duration of each of the at least one second resource; and generating a training sample, the training sample including resource attributes of the first resource and object attributes of the first object, and the label of the training sample including the reference duration.

[0005] According to another aspect of the present disclosure, a training method for a deep learning model is provided, comprising: obtaining a training sample, the training sample comprising: resource attributes of a first resource and object attributes of a first object, the label of the training sample comprising a reference duration; inputting the resource attributes of the first resource and the object attributes of the first object into the deep learning model to obtain a first estimated duration; and training the deep learning model based on the difference between the first estimated duration and the reference duration; wherein the reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of each of at least one second resource, the first object interacts with the first resource at a first moment, the first object interacts with the second resource at a second moment after the first moment, and the first resource and the second resource satisfy an association relationship; the additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource.

[0006] According to another aspect of the present disclosure, a resource display method is provided, including: inputting resource attributes of a candidate resource and object attributes of a second object into a trained deep learning model to obtain a second estimated duration; and determining a display result of the candidate resource based on the second estimated duration; wherein the deep learning model is trained according to the above-mentioned training method.

[0007] According to another aspect of the present disclosure, a sample generation device is provided, comprising: a resource determination module, an additional duration determination module, a reference duration determination module, and a generation module. The resource determination module is configured to determine a first resource and at least one second resource associated with the first resource; wherein the first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment; the additional duration determination module is configured to determine the additional duration for the second resource based on the time interval between the first moment and the second moment, and the browsing duration of the first object for the second resource; the reference duration determination module is configured to determine the reference duration based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource; and the generation module is configured to generate a training sample, wherein the training sample includes resource attributes of the first resource and object attributes of the first object, and the label of the training sample includes the reference duration.

[0008] According to another aspect of the present disclosure, a training device for a deep learning model is provided, comprising: an acquisition module, an input module, and a training module. The acquisition module is used to acquire training samples, the training samples comprising: resource attributes of a first resource and object attributes of a first object, and the labels of the training samples comprising a reference duration. The input module is used to input the resource attributes of the first resource and the object attributes of the first object into the deep learning model to obtain a first estimated duration. The training module is used to train the deep learning model based on the difference between the first estimated duration and the reference duration. The reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource. The first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment. The first resource and the second resource satisfy an association relationship; the additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource.

[0009] According to another aspect of the present disclosure, a resource display device is provided, comprising: a second estimated duration determination module and a display result determination module. The second estimated duration determination module is configured to input resource attributes of a candidate resource and object attributes of a second object into a trained deep learning model to obtain a second estimated duration; the display result determination module is configured to determine a display result for the candidate resource based on the second estimated duration. The deep learning model is trained using the aforementioned training method.

[0010] 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 to enable the at least one processor to execute the method provided by the present disclosure.

[0011] 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 cause a computer to execute the method provided by the present disclosure.

[0012] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements the method provided in the present disclosure when executed by a processor.

[0013] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0015] Figure 1 is a schematic diagram of an application scenario of the sample generation method, model training method, resource display method and device according to an embodiment of the present disclosure;

[0016] Figure 2 is a schematic flow chart of a sample generation method according to an embodiment of the present disclosure;

[0017] Figure 3 is a schematic diagram of a sample generation method according to an embodiment of the present disclosure;

[0018] Figure 4 is a schematic flow chart of a model training method according to an embodiment of the present disclosure;

[0019] Figure 5 is a schematic diagram of a model training method according to an embodiment of the present disclosure;

[0020] Figure 6 is a schematic flow chart of a resource display method according to an embodiment of the present disclosure;

[0021] Figure 7 is a schematic structural block diagram of a sample generating device according to an embodiment of the present disclosure;

[0022] Figure 8 is a schematic structural block diagram of a model training device according to an embodiment of the present disclosure;

[0023] Figure 9 is a schematic structural block diagram of a resource display device according to an embodiment of the present disclosure; and

[0024] Figure 10 It is a structural block diagram of an electronic device used to implement the sample generation method, model training method, and resource display method of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0025] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize 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.

[0026] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0027] In the technical solution disclosed herein, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0028] In some technical solutions, recommendation algorithms calculate users' immediate feedback on resources and make recommendations based on this feedback. This feedback includes information such as whether users click on a resource and how long they browse it. However, recommendation algorithms rarely consider users' ongoing preferences, which can lead to biased individual preferences and affect recommendation effectiveness.

[0029] The disclosed embodiments aim to provide a sample generation method, wherein the labels of the training samples determined by the method include a reference duration, which can reflect both the user's immediate preference and the user's ongoing preference. The disclosed embodiments also provide a resource display method, which uses the training samples to train a deep learning model. The second estimated duration determined by the trained deep learning model can reflect the user's immediate preference and ongoing preference for candidate resources, thereby more accurately representing the user's personal preference for the candidate resources, thereby improving the accuracy of resource recommendations and enhancing the user experience.

[0030] The technical solutions provided by the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Figure 1 It is a schematic diagram of an application scenario of the sample generation method, model training method, resource display method and device according to the embodiments of the present disclosure.

[0032] It should be noted that Figure 1 The examples shown are merely examples of system architectures to which the embodiments of the present disclosure may be applied, 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 may not be used in other devices, systems, environments or scenarios.

[0033] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, a server 105, and a database 106. The network 104 is used as a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0034] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Terminal devices 101, 102, and 103 can be various electronic devices with display screens and support web browsing, including but not limited to smartphones, tablet computers, laptop computers, and desktop computers, etc.

[0035] Server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using terminal devices 101, 102, and 103. The background management server may analyze and process received data such as user requests and feedback the processing results to the terminal device.

[0036] The database 106 can be used to store data required or generated during data processing, such as training samples, model parameters, etc.

[0037] For example, during the sample generation process, users can interact with resources through terminal devices 101, 102, and 103, and their historical interaction information (including user information and resource information) can be stored in database 106. Server 105 can generate training samples based on this historical interaction information and store them in database 106.

[0038] For another example, in the model training process, the server 105 can extract training samples from the database 106 and train the deep learning model. The trained deep learning model may not belong to the server 105, but may also be deployed on the terminal devices 101, 102, and 103.

[0039] For another example, in the resource recommendation process, the user's relevant information can be obtained through the terminal devices 101, 102, and 103, and then the second estimated duration can be determined using the trained deep model to make resource recommendations.

[0040] It should be noted that the sample generation method, model training method, and resource display method provided in the embodiments of the present disclosure can generally be executed by the terminal devices 101, 102, 103 or the server 105. Accordingly, the sample generation apparatus, model training apparatus, and resource display method apparatus provided in the embodiments of the present disclosure can generally be set in the terminal devices 101, 102, 103 or the server 105.

[0041] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0042] Figure 2 is a schematic flowchart of a sample generation method according to an embodiment of the present disclosure.

[0043] like Figure 2 As shown, the sample generating method 200 may include operations S210 to S240.

[0044] In operation S210 , a first resource and at least one second resource that satisfies an association relationship with the first resource are determined.

[0045] For example, the first resource is a resource that interacts with the first object, the first object may be a user, and the interaction time is the first moment. The interaction between the first object and the first resource includes: the first object performing at least one action on the first resource, such as clicking, browsing, adding to favorites, liking, commenting, forwarding, or downloading.

[0046] For example, the second resource is a resource that interacts with the first object, and the interaction occurs at the second moment. The interaction between the first object and the second resource includes the first object performing at least one action on the second resource, such as clicking, browsing, adding to favorites, liking, commenting, forwarding, or downloading. The following uses browsing as an example for illustration.

[0047] It should be noted that the second moment can be any moment within the target duration after the first moment. The target duration can be a pre-configured predetermined duration, or a duration determined based on the category of the first resource or other information, or there is no limit on the specific value of the target duration.

[0048] It should be noted that the number of second resources can be one or more, and the second moments at which different second resources interact with the first object may be different. For example, the user browses the first resource at 9:00, browses a second resource at 9:05, and browses another second resource at 9:10. The second moments for the two second resources are 9:05 and 9:10 respectively.

[0049] For example, the association relationship may include: the similarity between the first resource and the second resource is greater than or equal to a similarity threshold; the similarity between the first resource and the second resource can be calculated from at least one category of vectors, such as semantic vectors, content vectors, click vectors, and author vectors; the similarity can use cosine similarity or other similarity; the similarity threshold can be 0.55; this embodiment does not limit the similarity calculation method and similarity threshold. For another example, the association relationship may include: the first resource and the second resource belong to the same category of resources, or the first resource and the second resource have the same author, etc.

[0050] In operation S220 , an additional duration for the second resource is determined according to a time interval between the first moment and the second moment and a browsing duration of the first object for the second resource.

[0051] For example, the time interval can be mapped to a numerical value, and then the numerical value can be multiplied, divided, added, subtracted, or the like by the browsing duration to obtain the additional duration. The specific calculation method needs to be determined according to actual needs and is not limited in this embodiment. For example, if the numerical value is positively correlated with the time interval, the numerical value can be divided or subtracted from the browsing duration. For another example, if the numerical value is negatively correlated with the time interval, the numerical value can be multiplied or added to the browsing duration.

[0052] In operation S230 , a reference duration is determined based on a browsing duration of the first object with respect to the first resource and an additional duration of each of the at least one second resource.

[0053] For example, the browsing time of the first resource and the weighted sum of the additional time of each second resource can be calculated as the reference time. The weight of the first resource and the weight of the second resource can be the same or different, and the weight can be pre-configured.

[0054] In operation S240 , a training sample is generated. The training sample includes resource attributes of the first resource and object attributes of the first object. The label of the training sample includes a reference duration.

[0055] For example, the resource attributes of the first resource may include identifier, category, length, style, author, and pageviews. The object attributes of the first object may include identifier, preference, region, etc. It should be noted that the acquisition and use of the relevant information of the first object are known and agreed upon by the user, comply with relevant laws and regulations, and do not violate public order and good morals.

[0056] According to the sample generation method provided by the embodiment of the present disclosure, the method, for a first resource, first counts other resources browsed by the user within a period of time after browsing the first resource. If the other resources satisfy an association relationship with the first resource, the other resources are regarded as second resources. Then, based on the time interval between the user's interaction with the first resource and the interaction with the second resource, the browsing time of the second resource is converted into additional time, and the reference time is determined by the additional time and the browsing time of the first resource. It should be noted that, first, the browsing time of the first resource can reflect the user's immediate preference. Secondly, the user interacts with the second resource that satisfies an association relationship with the first resource, indicating that the user has an interest in the first resource and that the interest will last for a period of time. Therefore, the additional time can reflect the user's continued preference for the first resource. Since the reference time is determined based on the browsing time of the first resource and the additional time determined based on the second resource, the reference time can reflect both the user's immediate preference and the user's continued preference.

[0057] This embodiment takes a second resource as an example to illustrate the process of determining the additional duration of the second resource.

[0058] In this embodiment, a decay parameter may be determined based on the time interval between the first moment and the second moment, and then the additional time duration may be determined based on the decay parameter and the browsing time duration of the first object for the second resource.

[0059] A reference value may be determined based on a time interval. For example, if the predetermined duration is 5 minutes and the time interval is 18 minutes, the reference value may be the ratio of the time interval to the predetermined duration. In this case, the reference value is 3.6. The reference value may also be an integer determined based on the ratio of the time interval to the predetermined duration. For example, 3.6 may be rounded down to obtain a reference value of 3, or 3.6 may be rounded up to obtain a reference value of 4.

[0060] Next, the attenuation parameter may be determined based on the reference value, for example, the reference value may be used as the attenuation parameter, or the reference value may be mapped to the attenuation parameter through a monotonically increasing function, such that the attenuation parameter is positively correlated with the reference value.

[0061] Next, the additional duration can be calculated using the decay parameter. For example, the decay parameter can be used as an exponent of the initial coefficient, thereby processing the initial coefficient into a target coefficient. The initial coefficient can be greater than 0 and less than 1, such as 0.99. The product of the target coefficient and the browsing duration of the second resource can be used as the additional duration.

[0062] Using the above solution, for a second resource, if the time interval between the first moment and the second moment is larger, the attenuation parameter is larger, the target coefficient is smaller, and the additional duration of the second resource is smaller, so that the additional duration decays to a smaller value as the time interval becomes larger. In this way, the reference duration calculated based on the additional duration is smaller, indicating that the first resource does not quite meet the preference of the first object. If the time interval between the first moment and the second moment is smaller, the calculated reference duration is larger, indicating that the first resource is more in line with the preference of the first object. Therefore, the reference duration can more accurately represent the preference of the first object for the first resource. Therefore, after generating training samples and training the deep learning model, the second estimated duration output by the deep learning model can accurately reflect the second object's preference for the candidate resource, thereby improving the accuracy of resource recommendation.

[0063] The above embodiment takes a second resource as an example to illustrate the process of determining the additional duration of the second resource. When there are multiple second resources, the above scheme can be used to calculate the additional duration of each second resource separately. The sum of the additional durations of multiple second resources can be used as the total additional duration, and the sum of the total additional duration and the browsing duration of the first resource can be used as the reference duration. It should be noted that the more second resources there are, the more similar second resources the user will browse after browsing the first resource, so the first resource is more suitable for recommendation to the user.

[0064] According to another embodiment of the present disclosure, the additional duration satisfies a first condition, and the first condition includes: when the browsing duration of the first object for the second resource is the same, the additional duration is negatively correlated with the decay parameter, and the decay parameter is positively correlated with the reference value determined based on the time interval. That is, when the browsing duration of the first object for the second resource is the same, the longer the time interval between the first moment and the second moment, the longer the time interval between the first moment and the second moment, indicating that the first object browsed the similar second resource after browsing the first resource, and the first object has less interest in the resource. Therefore, the larger the decay parameter, the smaller the additional duration, thereby reducing the reference duration, so that the reference duration more accurately represents the preference of the first object.

[0065] According to another embodiment of the present disclosure, the additional duration satisfies a second condition, which includes: under the same decay parameter, the additional duration is positively correlated with the first subject's browsing duration for the second resource. That is, under the same decay parameter, the longer the first subject browses the second resource, the higher the first subject's preference for the second resource. Consequently, the additional duration for the second resource will increase, thereby increasing the reference duration, so that the reference duration more accurately represents the first subject's preference.

[0066] According to another embodiment of the present disclosure, the additional duration satisfies the first condition and the second condition.

[0067] According to another embodiment of the present disclosure, the training sample includes the resource attributes of the first resource, the object attributes of the first object and the first historical resource sequence. The first historical resource sequence includes at least one historical resource that interacted with the first object before the first moment. A predetermined number of resources with the most recent interaction time before the first moment can be screened and added to the first historical resource sequence. The predetermined number can be 20. The first historical resource sequence is used to be fused with the object attributes of the first object to obtain the first object features of the first object. In this way, the first object features can be jointly determined by the object attributes of the first object and the first historical resource sequence with which the first object has historically interacted, thereby more accurately characterizing the first object. In some embodiments, the features of each historical resource in the first historical resource sequence used in the fusion process may include only the identification of the resource, and may also include category, length, style, author, number of views, etc.

[0068] Figure 3 is a schematic diagram of a sample generation method according to an embodiment of the present disclosure.

[0069] In this embodiment, it is possible to first determine the first resource 302 and at least one second resource 303 that the first object 301 has browsed, and the first resource 302 and the second resource 303 satisfy an association relationship, for example, the similarity between the two is greater than a similarity threshold. In addition, the first object 301 interacts with the first resource 302 at a first moment T1, and the first object 301 interacts with the second resource 303 at any second moment T2 within a predetermined time period after the first moment T1. Next, the attenuation parameter 306 can be determined based on the time interval between the first moment T1 and the second moment T2, and then the additional duration 307 can be determined based on the attenuation parameter 306 and the browsing duration 305 of the first object 301 for the second resource 303.

[0070] Next, a reference duration 308 can be determined based on the browsing duration 304 of the first object 301 for the first resource 302 and the additional duration 307 of each of the at least one second resource 303. For example, the sum of the browsing duration values ​​of each resource is used as the reference duration 308. This can generate a training sample 311. The training sample 311 includes the resource attribute 309 of the first resource 302 and the object attribute 310 of the first object 301. The label of the training sample 311 includes the reference duration 308.

[0071] Figure 4 It is a schematic flowchart of the model training method according to an embodiment of the present disclosure.

[0072] like Figure 4 As shown, the deep learning model training method 400 may include operations S410 to S430.

[0073] In operation S410 , a training sample is acquired. The training sample includes resource attributes of a first resource and object attributes of a first object. A label of the training sample includes a reference duration.

[0074] For example, a training sample can be obtained using any of the above-mentioned sample generation methods. For example, the reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource. The first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment. The first resource and the second resource satisfy an association relationship. The additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource. Other methods may also be used to obtain training samples, which are not limited in this embodiment.

[0075] In operation S420 , resource attributes of the first resource and object attributes of the first object are input into a deep learning model to obtain a first estimated duration.

[0076] For example, the resource attributes of the first resource may include identification, category, length, style, author, number of views, etc. The object attributes of the first object may include identification, preference, region, etc. The resource attributes of the first resource and the object attributes of the first object are input into the deep learning model to be trained, and the deep learning model to be trained outputs a first estimated duration. The first estimated duration can represent both the duration of the first object browsing the first resource and the additional duration of the first object browsing the related second resource after browsing the first resource, and the additional duration decays with the time interval between browsing the first resource and browsing the second resource. Therefore, the first estimated duration determined by the deep learning model can represent the first object's preference for the first resource from the two dimensions of immediate preference and continuous preference.

[0077] In operation S430 , a deep learning model is trained based on the difference between the first estimated duration and the reference duration.

[0078] For example, a predetermined loss function can be used to calculate the first estimated duration and the reference duration to obtain a loss value, and then the parameters of the deep learning model to be trained can be adjusted based on the loss value until the deep learning model converges. In the actual training process, the deep learning model can include multiple network layers, and the deep learning model can be trained as a whole or in part.

[0079] The disclosed embodiments utilize training samples to train a deep learning model. This allows the first estimated duration output by the deep learning model to be close to the reference duration in the label, improving the accuracy of the first estimated duration. Consequently, when using the deep learning model for reasoning, the second subject's preference for the candidate resource can be more accurately determined.

[0080] According to another embodiment of the present disclosure, a deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model. Next, the process of determining the first estimated duration is described. In this embodiment, resource attributes of a first resource are input into the first sub-model to obtain a first resource feature. Object attributes of a first object are processed based on the encoding sub-model to obtain a first object feature. The first resource feature and the first object feature are then input into the second sub-model to obtain the first estimated duration.

[0081] This embodiment utilizes the first sub-model and the encoding sub-model to process the resource attributes of the first resource and the object attributes of the first object respectively, so that the first object characteristics and the first resource characteristics can be determined more accurately.

[0082] Figure 5 It is a schematic diagram of the model training method according to an embodiment of the present disclosure.

[0083] In this embodiment, the deep learning model includes a first sub-model M_1, a second sub-model M_2 and an encoding sub-model, and the encoding sub-model includes a target network M_31 and an encoder M_32.

[0084] In practical applications, encoder M_32 may be a Transformer encoder. The first sub-model M_1, the second sub-model M_2, and the target network M_31 may each include at least one fully connected layer. This embodiment does not limit the network structures of the first sub-model M_1, the second sub-model M_2, the encoder M_32, and the target network M_31.

[0085] The training samples used to train the deep learning model include: resource attribute F_1 of the first resource, object attribute F_3 of the first object, and a first historical resource sequence that interacted with the first object before the first moment. The label of the training sample is the reference duration Label.

[0086] For example, the object attribute F_3 of the first object is input into the target network M_31 to obtain the first initial feature F_4. The object attribute F_3 of the first object includes an identifier and other attributes, such as region and age. It should be noted that the feature dimension of the object attribute F_3 of the first object is different from the dimension of the features of each historical resource in the first historical resource sequence. The target network M_31 is mainly used to change the feature dimension of the object attribute F_3 of the first object, for example, by reducing the dimension so that the dimension of the first initial feature F_4 is the same as the dimension of the features of each historical resource in the first historical resource sequence.

[0087] The first initial feature F_4 and the features of each historical resource in the first historical resource sequence are then input into encoder M_32. The input features of encoder M_32 correspond one-to-one with the output features. The first object feature F_5 can be determined based on the output features of encoder M_32 corresponding to the first initial feature F_4. For example, the object attribute F_3 of the first object is reduced in dimension by the target network M_31 to become the first initial feature F_4. This first initial feature F_4 is then concatenated with the features of each historical resource. The encoder M_32 learns the user representation, and the first output vector is used as the first object feature F_5. This first vector is the hidden feature 0 in the figure.

[0088] The above describes the process of determining the first object feature F_5. It should be noted that encoder M_32 and target network M_31 are primarily used to fuse the first object's object attribute F_3 and the first historical resource sequence into the first object feature F_5. In other embodiments, other fusion methods may also be used, for example, by concatenating the features of the first object's object attribute F_3 with the features of each historical resource in the first historical resource sequence. This embodiment does not limit the fusion method.

[0089] In addition, the resource attribute F_1 of the first resource is input into the first sub-model M_1. The first sub-model M_1 may include two fully connected networks. The resource representation is learned through the fully connected networks to obtain the first resource feature F_2.

[0090] The first resource feature F_2 and the first object feature F_5 can then be fused together through concatenation or other methods to obtain a fused feature F_6. This fused feature F_6 is then input into the second sub-model M_2, which then outputs a first estimated duration Output. The loss function is then used to calculate the difference between the first estimated duration Output and the reference duration Label in the label to obtain the loss value Loss.

[0091] Next, the deep learning model can be trained based on the loss value Loss, for example, by adjusting the network parameters of the first sub-model M_1, the second sub-model M_2, the target network M_31, and the encoder M_32.

[0092] It should be noted that in the actual resource recommendation process, some applications use a single-column recommendation method, displaying only one column of target resources on the front-end page. Other applications use a dual-column recommendation method, displaying two columns of target resources on the front-end page. Different loss functions can be configured for different recommendation scenarios.

[0093] Figure 6 is a schematic flow chart of a resource display method according to an embodiment of the present disclosure.

[0094] like Figure 6As shown, the resource display method 600 may include operations S610 to S620.

[0095] In operation S610 , resource attributes of the candidate resource and object attributes of the second object are input into a trained deep learning model to obtain a second estimated duration.

[0096] For example, the object attributes of the second object may include identifiers, preferences, and regions. It should be noted that the acquisition and use of the relevant information of the second object are known and consistent to the user, comply with relevant laws and regulations, and do not violate public order and good morals. Relevance between the second object and various resources in the database can be preliminarily determined using relevant algorithms, and the relevant resources can be selected as candidate resources.

[0097] For example, the deep learning model is a model for calculating duration. The second estimated duration calculated by the model can represent both the duration of the second object browsing the candidate resource and the additional duration of the second object browsing other related resources after browsing the last resource. Moreover, the additional duration decays with the time interval between browsing the candidate resource and other resources. Therefore, the second estimated duration determined by the deep learning model can represent the second object's preference for the candidate resource from the two dimensions of immediate preference and continuous preference. The deep learning model can be trained using the above-mentioned deep learning model training method or other methods, which is not limited in this embodiment.

[0098] In operation S620 , a presentation result of the candidate resource is determined according to the second estimated duration.

[0099] For example, at least one target resource to be displayed can be determined from multiple candidate resources based on the second estimated duration. The larger the second estimated duration, the higher the preference of the second object for the candidate resource. The initial evaluation value of each candidate resource can be calculated by a relevant evaluation algorithm, and the initial evaluation value can be adjusted based on the second estimated duration to obtain the target evaluation value of each candidate resource. If the candidate target evaluation value meets the threshold condition or the ranking of the target evaluation value in the evaluation value sequence meets the predetermined order condition, the candidate resource can be determined as the target resource and recommended to the user. These target resources are the resources that need to be displayed to the second object. Otherwise, the candidate resource will not be recommended to the user, that is, the candidate resource will not be displayed.

[0100] In this embodiment, the second estimated duration output by the deep learning model can accurately characterize the second object's preference for candidate resources from the two dimensions of immediate preference and continuous preference. Therefore, resource recommendations based on the second estimated duration can provide the second object with target resources with a high degree of continuous preference, thereby improving the recommendation effect and increasing the second object's satisfaction with the recommended target resources.

[0101] According to another embodiment of the present disclosure, a deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model. During inference, resource attributes of a candidate resource may be input into the first sub-model to obtain candidate resource features. Furthermore, object attributes of a second object may be processed based on the encoding sub-model to obtain second object features. The candidate resource features and the second object features are then input into the second sub-model to obtain a second estimated duration.

[0102] This embodiment uses the first sub-model and the encoding sub-model to process the resource attributes of the candidate resource and the object attributes of the second object respectively, which can more accurately determine the candidate resource characteristics and the second object characteristics.

[0103] According to another embodiment of the present disclosure, the deep learning model includes a first sub-model, a second sub-model and an encoding sub-model, and the encoding sub-model includes an encoder and a target network.

[0104] For example, the object attributes of the second object are input into the target network to obtain the second initial feature. The object attributes of the first object include an identifier and other attributes, such as region, age, etc. The target network changes the feature dimension of the object attributes of the second object so that the dimension of the second initial feature is the same as the dimension of the feature of each historical resource in the second historical resource sequence. The above-mentioned second historical resource sequence includes historical resources that have interacted with the second object in the past period of time. The features of each historical resource in the second historical resource sequence may include only the identifier of the resource, and may also include category, length, style, author, number of views, etc.

[0105] Then, the second initial feature and the features of each historical resource in the first historical resource sequence are input into the encoder, and the second object feature is determined according to the output feature output by the encoder corresponding to the second initial feature.

[0106] Alternatively, the resource attributes of the candidate resource may be input into the first sub-model to obtain a first resource feature. The candidate resource feature and the second object feature may then be concatenated to obtain a fused feature, which is then input into the second sub-model to obtain a second estimated duration output by the second sub-model.

[0107] Figure 7 is a schematic structural block diagram of a sample generating device according to an embodiment of the present disclosure.

[0108] like Figure 7 As shown, the sample generating apparatus 700 may include a resource determining module 710 , an additional duration determining module 720 , a reference duration determining module 730 and a generating module 740 .

[0109] The resource determination module 710 is used to determine a first resource and at least one second resource that is associated with the first resource; wherein the first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment.

[0110] The additional duration determining module 720 is configured to determine an additional duration for the second resource according to a time interval between the first moment and the second moment and a browsing duration of the first object for the second resource.

[0111] The reference duration determining module 730 is configured to determine a reference duration according to a browsing duration of the first object for the first resource and an additional duration of each of the at least one second resource.

[0112] The generating module 740 is used to generate a training sample. The training sample includes resource attributes of the first resource and object attributes of the first object. The label of the training sample includes a reference duration.

[0113] According to another embodiment of the present disclosure, an additional duration determination module includes: a decay parameter determination submodule and an additional duration determination submodule. The decay parameter determination submodule is configured to determine a decay parameter based on a time interval between a first moment and a second moment. The additional duration determination submodule is configured to determine an additional duration based on the decay parameter and a duration during which the first object browses the second resource.

[0114] According to another embodiment of the present disclosure, the additional duration satisfies at least one of the following conditions: when the browsing duration of the first object for the second resource is the same, the additional duration is negatively correlated with the decay parameter, and the decay parameter is positively correlated with a reference value determined based on the time interval; and when the decay parameters are the same, the additional duration is positively correlated with the browsing duration of the first object for the second resource.

[0115] According to another embodiment of the present disclosure, the training sample further includes a first historical resource sequence that interacts with the first object before the first moment, and the first historical resource sequence is used to be fused with the object attribute of the first object to obtain a first object feature of the first object.

[0116] According to another embodiment of the present disclosure, the association relationship includes: the similarity between the second resource and the first resource is greater than or equal to a similarity threshold.

[0117] According to another embodiment of the present disclosure, the interaction between the first object and the resource includes: the first object performs at least one of the following actions on the resource: click, browse, collect, like, comment, forward, download; wherein the resource is any one of the first resource and the second resource.

[0118] Figure 8 It is a schematic structural block diagram of a model training device according to an embodiment of the present disclosure.

[0119] like Figure 8 As shown, the training device 800 of the deep learning model may include an acquisition module 810, an input module 820 and a training module 830.

[0120] The acquisition module 810 is used to acquire a training sample. The training sample includes: resource attributes of the first resource and object attributes of the first object. The label of the training sample includes a reference duration.

[0121] The input module 820 is used to input the resource attributes of the first resource and the object attributes of the first object into the deep learning model to obtain a first estimated duration.

[0122] The training module 830 is used to train the deep learning model according to the difference between the first estimated duration and the reference duration.

[0123] The reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource. The first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment. The first resource and the second resource satisfy an association relationship; the additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource.

[0124] According to another embodiment of the present disclosure, a deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model; and an input module includes a first processing sub-module, a second processing sub-module, and a third processing sub-module. The first processing sub-module is configured to input resource attributes of a first resource into the first sub-model to obtain a first resource feature. The second processing sub-module is configured to process object attributes of a first object based on the encoding sub-model to obtain a first object feature. The third processing sub-module is configured to input the first resource feature and the first object feature into the second sub-model to obtain a first estimated duration.

[0125] According to another embodiment of the present disclosure, the training sample also includes a first historical resource sequence that interacted with the first object before the first moment, and the encoding sub-model includes an encoder and a target network; the second processing sub-module includes: a first processing unit, a second processing unit, and a third processing unit. The first processing unit is used to input the object attributes of the first object into the target network to obtain a first initial feature; the dimension of the first initial feature is the same as the dimension of the feature of each historical resource in the first historical resource sequence. The second processing unit is used to input the first initial feature and the feature of each historical resource in the first historical resource sequence into the encoder. The third processing unit is used to determine the first object feature based on the output feature output by the encoder corresponding to the first initial feature.

[0126] Figure 9It is a schematic structural block diagram of a resource display device according to an embodiment of the present disclosure.

[0127] like Figure 9 As shown, the resource display device 900 may include a second estimated duration determination module 910 and a display result determination module 920 .

[0128] The second estimated duration determining module 910 is configured to input the resource attributes of the candidate resource and the object attributes of the second object into the trained deep learning model to obtain a second estimated duration; and

[0129] The presentation result determination module 920 is configured to determine the presentation result of the candidate resource according to the second estimated duration. The deep learning model is trained using the above training method.

[0130] According to another embodiment of the present disclosure, a deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model; and a second estimated duration determination module includes a fourth processing sub-module, a fifth processing sub-module, and a sixth processing sub-module. The fourth processing sub-module is configured to input resource attributes of a candidate resource into the first sub-model to obtain candidate resource features. The fifth processing sub-module is configured to process object attributes of a second object based on the encoding sub-model to obtain second object features. The sixth processing sub-module is configured to input candidate resource features and second object features into the second sub-model to obtain a second estimated duration.

[0131] According to another embodiment of the present disclosure, the encoding sub-model includes an encoder and a target network; the fifth processing sub-module includes: a fourth processing unit, a fifth processing unit, and a sixth processing unit. The fourth processing unit is used to input the object attributes of the second object into the target network to obtain a second initial feature; the dimension of the second initial feature is the same as the dimension of the feature of each historical resource in the second historical resource sequence, and the second historical resource sequence includes historical resources that interact with the second object. The fifth processing unit is used to input the second initial feature and the features of each historical resource in the first historical resource sequence into the encoder. The sixth processing unit is used to determine the second object feature based on the output feature output by the encoder corresponding to the second initial feature.

[0132] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions that can be executed 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 any one of the above methods.

[0133] According to an embodiment of the present disclosure, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute any one of the above methods.

[0134] According to an embodiment of the present disclosure, the present disclosure further provides a computer program product, including a computer program, which implements any of the above methods when executed by a processor.

[0135] Figure 10 1 is a block diagram of an electronic device for implementing the sample generation method, model training method, and resource display method of an embodiment of the present disclosure. 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 may also represent various forms of mobile devices, such as personal digital assistants, 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 claimed herein.

[0136] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. RAM 1003 may also store various programs and data required for the operation of device 1000. Computing unit 1001, ROM 1002, and RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to bus 1004.

[0137] Various components in device 1000 are connected to I / O interface 1005, including an input unit 1006, such as a keyboard, mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a magnetic disk, optical disk, etc.; and a communication unit 1009, such as a network card, modem, wireless communication transceiver, etc. The communication unit 1009 allows device 1000 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0138] Computing unit 1001 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 1001 performs the various methods and processes described above, such as any of the aforementioned methods. For example, in some embodiments, any of the aforementioned methods may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed onto device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by computing unit 1001, one or more steps of any of the aforementioned methods described above may be performed. Alternatively, in other embodiments, computing unit 1001 may be configured to perform any of the aforementioned methods via any other suitable means (e.g., via firmware).

[0139] Various embodiments of the systems and techniques described above 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), system-on-a-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes 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.

[0140] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0141] 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, apparatus, or device. 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, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, 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), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] 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).

[0143] The systems and techniques described herein can 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 can 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.

[0144] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

[0145] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0146] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A sample generation method, comprising: Determining a first resource and at least one second resource that is associated with the first resource; wherein a first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment; Determining an additional duration for the second resource according to a time interval between the first moment and the second moment and a browsing duration of the first object for the second resource; determining a reference duration according to a browsing duration of the first object for the first resource and the additional duration of each of the at least one second resource; and A training sample is generated, where the training sample includes resource attributes of the first resource and object attributes of the first object, and a label of the training sample includes the reference duration.

2. The method according to claim 1, wherein The determining, based on the time interval between the first moment and the second moment and the browsing time of the first object for the second resource, of the additional duration for the second resource includes: determining a decay parameter according to a time interval between the first moment and the second moment; and The additional duration is determined according to the decay parameter and a browsing duration of the first object with respect to the second resource.

3. The method according to claim 2, wherein: The additional duration satisfies at least one of the following conditions: In a case where the browsing durations of the first object for the second resource are the same, the additional duration is negatively correlated with the decay parameter, and the decay parameter is positively correlated with a reference value determined based on the time interval; as well as When the decay parameters are the same, the additional duration is positively correlated with the browsing duration of the first object for the second resource.

4. The method according to claim 1, wherein The training sample further includes a first historical resource sequence that interacted with the first object before the first moment, and the first historical resource sequence is used to be fused with the object attribute of the first object to obtain a first object feature of the first object.

5. The method according to claim 1, wherein The association relationship includes: The similarity between the second resource and the first resource is greater than or equal to a similarity threshold.

6. The method according to claim 1, wherein The interaction between the first object and the resource includes: The first object generates at least one of the following actions on the resource: click, browse, favorite, like, comment, forward, download; The resource is any one of the first resource and the second resource.

7. A method for training a deep learning model, comprising: Acquire a training sample, the training sample including: resource attributes of the first resource and object attributes of the first object, the label of the training sample including a reference duration; Inputting the resource attributes of the first resource and the object attributes of the first object into a deep learning model to obtain a first estimated duration; and Training the deep learning model based on a difference between the first estimated duration and the reference duration; In which, the reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource. The first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment. The first resource and the second resource satisfy an association relationship; the additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource.

8. The method according to claim 7, wherein: The deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model; and inputting the resource attributes of the first resource and the object attributes of the first object into the deep learning model to obtain the first estimated duration includes: Inputting the resource attribute of the first resource into the first sub-model to obtain a first resource feature; Processing the object attributes of the first object based on the encoding sub-model to obtain a first object feature; and The first resource feature and the first object feature are input into the second sub-model to obtain the first estimated duration.

9. The method according to claim 8, wherein The training sample further includes a first historical resource sequence that interacted with the first object before the first moment, the encoding sub-model includes an encoder and a target network; and the step of processing the object attributes of the first object based on the encoding sub-model to obtain the first object feature includes: Inputting the object attribute of the first object into the target network to obtain a first initial feature; the dimension of the first initial feature is the same as the dimension of the feature of each historical resource in the first historical resource sequence; Inputting the first initial feature and the feature of each historical resource in the first historical resource sequence into the encoder; and The first object feature is determined according to an output feature output by the encoder corresponding to the first initial feature.

10. A resource display method, comprising: Inputting the resource attributes of the candidate resource and the object attributes of the second object into the trained deep learning model to obtain a second estimated duration; as well as determining a display result of the candidate resource according to the second estimated duration; The deep learning model is trained according to the method described in any one of claims 7 to 9.

11. The method according to claim 10, wherein: The deep learning model includes a first sub-model, a second sub-model, and an encoding sub-model; and inputting the resource attributes of the candidate resource and the object attributes of the second object into the trained deep learning model to obtain the second estimated duration includes: Inputting the resource attributes of the candidate resource into the first sub-model to obtain the candidate resource features; Processing the object attributes of the second object based on the encoding sub-model to obtain a second object feature; and The candidate resource feature and the second object feature are input into the second sub-model to obtain the second estimated duration.

12. The method according to claim 11, wherein The encoding sub-model includes an encoder and a target network; and processing the object attribute of the second object based on the encoding sub-model to obtain the second object feature includes: Inputting the object attributes of the second object into the target network to obtain a second initial feature; the dimension of the second initial feature is the same as the dimension of the feature of each historical resource in a second historical resource sequence, and the second historical resource sequence includes historical resources that interact with the second object; inputting the second initial feature and the feature of each historical resource in the first historical resource sequence into the encoder; and The second object feature is determined according to the output feature output by the encoder corresponding to the second initial feature.

13. A sample generating device, comprising: a resource determination module, configured to determine a first resource and at least one second resource associated with the first resource; wherein a first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment; an additional duration determining module, configured to determine an additional duration for the second resource based on a time interval between the first moment and the second moment and a browsing duration of the first object for the second resource; a reference duration determining module, configured to determine a reference duration based on a browsing duration of the first object for the first resource and the additional duration of each of the at least one second resource; and The generating module is configured to generate a training sample, wherein the training sample includes resource attributes of the first resource and object attributes of the first object, and the label of the training sample includes the reference duration.

14. A training device for a deep learning model, comprising: An acquisition module, configured to acquire a training sample, wherein the training sample includes: resource attributes of the first resource and object attributes of the first object, and a label of the training sample includes a reference duration; an input module, configured to input the resource attributes of the first resource and the object attributes of the first object into a deep learning model to obtain a first estimated duration; and a training module, configured to train the deep learning model according to a difference between the first estimated duration and the reference duration; In which, the reference duration is determined based on the browsing duration of the first object for the first resource and the additional duration of at least one second resource. The first object interacts with the first resource at a first moment, and the first object interacts with the second resource at a second moment after the first moment. The first resource and the second resource satisfy an association relationship; the additional duration of the second resource is determined based on the time interval between the first moment and the second moment and the browsing duration of the first object for the second resource.

15. A resource display device comprising: A second estimated duration determination module, configured to input resource attributes of the candidate resource and object attributes of the second object into a trained deep learning model to obtain a second estimated duration; as well as a presentation result determination module, configured to determine a presentation result of the candidate resource according to the second estimated duration; Wherein, the deep learning model is trained according to the device according to claim 14.

16. 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 that can be executed 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 12.

17. 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 12.

18. 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 12.

Images