Resource recommendation method and device, and electronic device

Abstract

The present disclosure provides a resource recommendation method and device and electronic equipment, relates to the technical field of artificial intelligence, in particular to the technical field of deep learning, cloud computing, natural language processing, intelligent recommendation, large model and the like. The specific implementation scheme is: obtaining a candidate resource set and a historical resource sequence of a target object in a historical time period before a current time point; determining an interest representation vector of the target object at the current time point according to the historical resource sequence; the interest representation vector represents resources of interest to the target object at the current time point; selecting a to-be-recommended resource from the candidate resource set according to the interest representation vector, and then performing resource recommendation processing; wherein the interest representation vector can represent resources of interest to the target object at the current time point, so that the to-be-recommended resource determined is the resource that the target object currently wants to click, thereby improving the matching degree between the to-be-recommended resource and the content of interest to the target object, and improving the resource recommendation accuracy and resource recommendation efficiency.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to the fields of deep learning, cloud computing, natural language processing, intelligent recommendation, and large models, and especially to a resource recommendation method, apparatus, and electronic device. Background Technology

[0002] Current resource recommendation methods learn the characteristics of an object based on its historical resource consumption; based on the object's characteristics, they perform resource recall to obtain resources to be recommended, and then perform resource recommendation.

[0003] In the above scheme, the object's characteristics represent the object's long-term resource preferences, which may not match the resource the object currently wants to click, thus reducing the accuracy and efficiency of resource recommendation. Summary of the Invention

[0004] This disclosure provides a resource recommendation method, apparatus, and electronic device.

[0005] According to one aspect of this disclosure, a resource recommendation method is provided, the method comprising: acquiring a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; determining an interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector representing resources that the target object is interested in at the current time point; selecting resources to be recommended from the candidate resource set based on the interest representation vector; and performing resource recommendation processing on the target object based on the resources to be recommended.

[0006] According to another aspect of this disclosure, a method for training a resource recommendation model is provided. The method includes: acquiring training data; the training data includes structured content of sample historical resources in a sample historical resource sequence, structured content of sample recommended resources, and sample click resources in the sample recommended resources; acquiring an initial resource recommendation model; the resource recommendation model includes an encoding network, a decoding network, and a fully connected network connected in sequence; and training the initial resource recommendation model using the structured content of the sample historical resources, the structured content of the sample recommended resources, and the sample click resources to obtain a trained resource recommendation model.

[0007] According to another aspect of this disclosure, a resource recommendation apparatus is provided, the apparatus comprising: an acquisition module, configured to acquire a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; a determination module, configured to determine an interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector representing resources that the target object is interested in at the current time point; a selection module, configured to select resources to be recommended from the candidate resource set based on the interest representation vector; and a recommendation processing module, configured to perform resource recommendation processing on the target object based on the resources to be recommended.

[0008] According to another aspect of this disclosure, a training apparatus for a resource recommendation model is provided. The apparatus includes: a first acquisition module for acquiring training data; the training data includes structured content of sample historical resources in a sample historical resource sequence, structured content of sample recommendation resources, and sample click resources in the sample recommendation resources; a second acquisition module for acquiring an initial resource recommendation model; the resource recommendation model includes an encoding network, a decoding network, and a fully connected network connected in sequence; and a training processing module for training the initial resource recommendation model using the structured content of the sample historical resources, the structured content of the sample recommendation resources, and the sample click resources to obtain a trained resource recommendation model.

[0009] According to another aspect of this 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, the instructions being executed by the at least one processor to enable the at least one processor to perform the resource recommendation method proposed above in this disclosure; or to perform the training method of the resource recommendation model proposed above in this disclosure.

[0010] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided that stores computer instructions for causing a computer to execute the resource recommendation method proposed in this disclosure above; or, to execute the training method of the resource recommendation model proposed in this disclosure above.

[0011] According to another aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the resource recommendation method proposed above in this disclosure; or, implements the steps of the training method for the resource recommendation model proposed above in this disclosure.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0014] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure;

[0015] Figure 2 This is a schematic diagram according to the second embodiment of the present disclosure;

[0016] Figure 3 This is a schematic diagram according to the third embodiment of the present disclosure;

[0017] Figure 4 This is a schematic diagram according to the fourth embodiment of the present disclosure;

[0018] Figure 5 This is a schematic diagram according to the fifth embodiment of the present disclosure;

[0019] Figure 6 This is a schematic diagram according to the sixth embodiment of the present disclosure;

[0020] Figure 7 This is a block diagram of an electronic device used to implement the resource recommendation method or the training method of the resource recommendation model in the embodiments of this disclosure. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] Current resource recommendation methods learn the characteristics of an object based on its historical resource consumption; based on the object's characteristics, they perform resource recall to obtain resources to be recommended, and then perform resource recommendation.

[0023] In the above scheme, the object's characteristics represent the object's long-term resource preferences, which may not match the resource the object currently wants to click, thus reducing the accuracy and efficiency of resource recommendation.

[0024] To address the aforementioned issues, this disclosure proposes a resource recommendation method, apparatus, and electronic device.

[0025] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure. It should be noted that the resource recommendation method of the present disclosure can be applied to a resource recommendation device, which can be configured in an electronic device so that the electronic device can perform the resource recommendation function.

[0026] Among them, electronic devices can be any device with computing capabilities, such as personal computers (PCs), mobile terminals, servers, etc. Mobile terminals can be, for example, in-vehicle devices, mobile phones, tablets, personal digital assistants, wearable devices, smart speakers, servers, server clusters, and other hardware devices with various operating systems, touch screens and / or displays.

[0027] The resource recommendation device can also be software within an electronic device, such as resource recommendation software. In the following embodiments, an electronic device is used as an example for illustration.

[0028] like Figure 1 As shown, this resource recommendation method may include the following steps:

[0029] Step 101: Obtain the candidate resource set and the historical resource sequence of the target object within the historical time period before the current time point.

[0030] In this embodiment of the disclosure, the electronic device may perform step 101 as follows: perform resource retrieval and resource filtering based on the attribute information of the target object to obtain a set of candidate resources; obtain the historical resources clicked by the target object within the historical time period before the current time point, and the click time points of the historical resources; sort each historical resource according to the click time points to obtain a historical resource sequence.

[0031] Among them, the historical resources clicked by the target object within the historical time period before the current time point are the target object's behavioral data on resources, which can represent the resources that the target object was interested in during the historical time period. In addition, by combining historical resources, we can accurately predict the resources that the target object may be interested in at the current time point, thereby improving the accuracy of the determined interest representation vector and improving the matching degree between the determined recommended resources and the content that the target object is interested in.

[0032] The target object's attribute information may include at least one of the following: geographic location information, device attribute information, time information, age, gender, occupation type, etc. The device attribute information may include at least one of the following: usage records of various applications on the device, device usage duration, location information when using the device, etc., which can be set according to actual needs and are not specifically limited here.

[0033] The acquisition and use of user object attribute information in this embodiment are known and agreed to by the users, comply with relevant laws and regulations, and do not violate public order and good morals.

[0034] In this embodiment of the disclosure, the process by which the electronic device determines a candidate resource set based on the attribute information of the target object can be as follows: performing resource recall processing based on the attribute information of the target object to obtain a recalled resource set; combining the attribute information and a first click-through rate (CTR) model to determine a first CTR of the recalled resources in the recalled resource set; filtering the recalled resources in the recalled resource set based on the first CTR to obtain a filtered resource set; combining the attribute information and a second CTR model to determine a second CTR of the filtered resources in the filtered resource set; the second CTR model has a larger number of parameters than the first CTR model; and filtering the filtered resources in the filtered resource set based on the second CTR to obtain a candidate resource set.

[0035] Specifically, the electronic device can input the target object's attribute information and the recalled resources from the recall resource set into a first click-through rate (CTR) model to obtain the first CTR output by the first CTR model. The electronic device can also input the target object's attribute information and the filtered resources from the filtered resource set into a second CTR model to obtain the second CTR output by the second CTR model.

[0036] The process by which electronic devices filter recalled resources based on the first click-through rate (CTR) of the recalled resources in the recalled resource set can be, for example, as follows: for each recalled resource, determine whether the first CTR of the recalled resource is less than a first CTR threshold; if the first CTR is less than the first CTR threshold, filter the recalled resource; if the first CTR is greater than or equal to the first CTR threshold, retain the recalled resource.

[0037] The filtering process of electronic devices based on the second click-through rate (CTR) of the filtered resources in the filtered resource set can be, for example, as follows: for each filtered resource, determine whether the second CTR of the filtered resource is less than a second CTR threshold; if the second CTR is less than the second CTR threshold, filter the filtered resource; if the second CTR is greater than or equal to the second CTR threshold, retain the filtered resource. The first CTR threshold can be less than the second CTR threshold.

[0038] The second click-through rate (CTR) model has more parameters than the first CTR model. By combining the first CTR model with the filtering of recalled resources in the recall resource set, the number of resources that the second CTR model needs to process can be reduced, thereby reducing the processing load of the second CTR model and improving processing efficiency.

[0039] Step 102: Based on the historical resource sequence, determine the interest representation vector of the target object at the current time point; the interest representation vector represents the resources that the target object is interested in at the current time point.

[0040] In this embodiment, the interest representation vector can be a representation vector of the resources that the target object is interested in at the current time. Correspondingly, the electronic device can combine an interest prediction model to determine the interest representation vector, and then determine the resources to be recommended to the target object. The interest prediction model can be trained using the target object's historical resource sequence.

[0041] Here, it is assumed that the historical resource sequence includes N historical resources. The training process of the interest prediction model can be as follows: input the first N-1 historical resources into the interest prediction model to obtain the predicted interest representation vector output by the interest prediction model; combine the predicted interest representation vector to determine the Nth predicted resource; combine the Nth predicted resource and the Nth historical resource to determine the value of the loss function of the interest prediction model; train the interest prediction model based on this value to obtain the trained interest prediction model.

[0042] Step 103: Select resources to be recommended from the candidate resource set based on the interest representation vector.

[0043] In one embodiment of this disclosure, an electronic device can determine the similarity between an interest representation vector and the representation vectors of candidate resources in a candidate resource set; and select resources to be recommended from the candidate resource set based on the similarity.

[0044] Step 104: Based on the resources to be recommended, perform resource recommendation processing on the target object.

[0045] The resource recommendation method of this disclosure involves obtaining a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; determining an interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector representing the resources that the target object is interested in at the current time point; selecting resources to be recommended from the candidate resource set based on the interest representation vector; and performing resource recommendation processing on the target object based on the resources to be recommended. The interest representation vector determined by combining the historical resource sequence can represent the resources that the target object is interested in at the current time point, ensuring that the determined resources to be recommended are the resources that the target object currently wants to click on. This improves the matching degree between the resources to be recommended and the content of interest to the target object, thereby improving the accuracy of resource recommendation. Furthermore, accurate resource recommendation can reduce the number of resource recommendations, thus improving resource recommendation efficiency.

[0046] To improve the accuracy of the determined interest representation vector, features of historical resources within the historical resource sequence can be considered. Correspondingly, the electronic device can first determine the representation vectors of historical resources in the historical resource sequence, and then combine these representation vectors to determine the interest representation vector of the target object. For example... Figure 2 As shown, Figure 2 This is a schematic diagram based on the second embodiment of the present disclosure. Figure 2 The illustrated embodiment may include the following steps:

[0047] Step 201: Obtain the candidate resource set and the historical resource sequence of the target object within the historical time period before the current time point.

[0048] Step 202: Determine the representation vector of historical resources in the historical resource sequence.

[0049] In this embodiment of the disclosure, the process of the electronic device performing step 202 may be, for example, determining the structured content of historical resources in the historical resource sequence; the structured content includes at least one metadata of the historical resources; inputting the structured content into the encoding network in the resource recommendation model, and obtaining the representation vector of the historical resources output by the encoding network.

[0050] The metadata in structured content can include at least one of the following: the resource title, keywords in the resource, the resource's domain, and the target audience for the resource. The resource's domain can be any domain at various levels. For example, the resource's domain can include at least one of the following: the primary domain to which the resource belongs, the secondary domain to which the resource belongs under the primary domain, the tertiary domain to which the resource belongs under the secondary domain, etc. There are no specific limitations here; these can be set according to actual needs.

[0051] As an alternative, electronic devices can perform word segmentation on the metadata in the structured content to obtain individual words; determine the representation vector of each word; concatenate the representation vectors of each word to obtain a first concatenation result; and use the first concatenation result as the structured content to be input into the encoding network.

[0052] As an alternative, the electronic device can assign numbers to each word obtained from the segmentation process; and store the correspondence between the representation vectors and codes of each word. The numbers of each word are then concatenated to obtain a second concatenation result; this second concatenation result is used as structured content to be input into the encoding network. Correspondingly, the encoding network can query the representation vectors corresponding to each number in the second concatenation result; thereby obtaining the representation vectors of each word; and based on the representation vectors of each word, determining the representation vector of the resource.

[0053] The structured content setting allows the encoding network to determine the representation vector of newly released resources without needing to perform learning processing. This enables the resource recommendation method disclosed herein to be applied to recommending newly released resources, further improving resource recommendation efficiency.

[0054] Step 203: Determine the interest representation vector of the target object at the current time point based on the representation vectors of each historical resource in the historical resource sequence.

[0055] In this embodiment of the disclosure, the electronic device may perform step 203 as follows: determine a representation vector sequence based on the representation vectors of each historical resource in the historical resource sequence; input the representation vector sequence into the decoding network in the resource recommendation model to obtain the interest representation vectors output by the decoding network.

[0056] In this method, the interest representation vector is determined by combining the representation vector sequence and the decoding network. This allows the decoding network to learn the correlation between the various representation vectors in the representation vector sequence and to learn the changing trends of the resources that the target object is interested in. This improves the accuracy of the interest representation vector of the target object at the current time point and further improves the efficiency of resource recommendation.

[0057] In this embodiment, since the resources of interest to the target object are unlikely to change in a short period of time, in order to weaken the influence of the order of historical resources in the historical resource sequence on the interest representation vector, the historical resources in the historical resource sequence can be divided according to time points to obtain multiple historical resource combinations; the interest representation vector is determined by combining the combinations to which the historical resources in the historical resource sequence belong. Correspondingly, before step 203, the electronic device can also perform the following process: dividing each historical resource in the historical resource sequence according to the collection time point to obtain multiple historical resource combinations; determining the combination number of the historical resource combination and the number vector corresponding to the combination number; for each historical resource in the historical resource sequence, concatenating the number vector corresponding to the combination number of the historical resource combination to which the historical resource belongs with the representation vector of the historical resource to obtain the processed representation vector; updating the representation vector of the historical resource based on the processed representation vector.

[0058] The electronic device can divide the historical resource sequence into time periods based on the collection time points of each historical resource to obtain multiple historical time periods; each historical time period corresponds to a historical resource combination; the collection time points of the historical resources in the historical resource combination are located within the historical time period corresponding to the historical resource combination; a combination number is set for each historical resource combination; the combination number is vectorized to obtain the number vector.

[0059] Historical resources belonging to the same historical resource group are assigned the same group number, which can weaken the temporal order of historical resources to a certain extent and reduce the impact of the temporal order of historical resources on the interest representation vector.

[0060] Step 204: Select resources to be recommended from the candidate resource set based on the interest representation vector.

[0061] Step 205: Based on the resources to be recommended, perform resource recommendation processing on the target object.

[0062] It should be noted that for details regarding steps 201, 204, and 205, please refer to [the relevant documentation / reference]. Figure 1 Steps 101, 103 to 104 in the illustrated embodiment will not be described in detail here.

[0063] The resource recommendation method of this disclosure involves obtaining a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; determining the representation vector of each historical resource in the historical resource sequence; determining the interest representation vector of the target object at the current time point based on the representation vector of each historical resource in the historical resource sequence; selecting resources to be recommended from the candidate resource set based on the interest representation vector; and performing resource recommendation processing on the target object based on the resources to be recommended. The method combines first determining the representation vector of historical resources in the historical resource sequence, and then using the representation vector of historical resources to determine the interest representation vector of the target object. This approach considers more features of historical resources, further ensuring that the resources to be recommended are the resources the target object currently wants to click on. This improves the matching degree between the resources to be recommended and the content of interest to the target object, thereby improving the accuracy of resource recommendation. Furthermore, accurate resource recommendation reduces the number of resource recommendations, thus improving resource recommendation efficiency.

[0064] To further improve the matching degree between the identified resources to be recommended and the target audience, and to increase the click-through rate of the target audience on the resources to be recommended, the predicted click-through rate of the candidate resources can be determined based on the interest representation vector and the representation vectors of the candidate resources in the candidate resource set. Then, the resources to be recommended are selected based on the predicted click-through rate. For example... Figure 3 As shown, Figure 3 This is a schematic diagram based on the third embodiment of the present disclosure. Figure 3 The illustrated embodiment may include the following steps:

[0065] Step 301: Obtain the candidate resource set and the historical resource sequence of the target object within the historical time period before the current time point.

[0066] Step 302: Based on the historical resource sequence, determine the interest representation vector of the target object at the current time point; the interest representation vector represents the resources that the target object is interested in at the current time point.

[0067] Step 303: Determine the representation vector of the candidate resources in the candidate resource set.

[0068] In this embodiment of the disclosure, the process of the electronic device performing step 303 may be, for example, determining the structured content of the candidate resource; the structured content includes at least one metadata of the candidate resource; inputting the structured content of the candidate resource into the encoding network in the resource recommendation model, and obtaining the representation vector of the candidate resource output by the encoding network.

[0069] Step 304: Determine the predicted click-through rate of the candidate resources based on the interest representation vector and the representation vector of the candidate resources.

[0070] In this embodiment of the disclosure, the electronic device may perform step 304 by inputting the interest representation vector and the representation vector of the candidate resource into the fully connected network in the resource recommendation model, and obtaining the predicted click-through rate output by the fully connected network.

[0071] Among them, the resource recommendation model can be used to determine the predicted click-through rate of candidate resources by combining the fully connected network in the resource recommendation model after training, which can improve the accuracy of the predicted click-through rate.

[0072] Step 305: Select resources to be recommended from the candidate resource set based on the predicted click-through rate.

[0073] In one example of this disclosure, the electronic device performing step 305 may be, for example, sorting the candidate resources in the candidate resource set in descending order according to the predicted click-through rate to obtain a sorting result; and determining at least one candidate resource ranked first in the sorting result as a resource to be recommended.

[0074] As an alternative, the electronic device can determine whether the predicted click-through rate of each candidate resource in the candidate resource set is greater than or equal to a predicted click-through rate threshold; if the predicted click-through rate of the candidate resource is greater than or equal to the predicted click-through rate threshold, the candidate resource is identified as a resource to be recommended.

[0075] Among these methods, identifying candidate resources with higher predicted click-through rates as recommended resources can increase the probability of these recommended resources being clicked and reduce the number of times the target audience refreshes the recommended resources, thereby further improving the efficiency of resource recommendation.

[0076] In another example, the electronic device performing step 305 may be as follows: determining the value of a candidate resource in the candidate resource set on at least one recommendation metric; determining the score of the candidate resource based on the value of the candidate resource on at least one recommendation metric and the predicted click-through rate; sorting the candidate resources in the candidate resource set in descending order according to the score to obtain a sorting result; and determining at least one candidate resource ranked first in the sorting result as the resource to be recommended.

[0077] The recommendation metrics can include at least one of the following: browsing time metrics, interaction frequency metrics, and pageview frequency metrics, which can be set according to actual needs. The interaction frequency metrics can include at least one of the following: number of favorites, number of comments, number of likes, number of shares, number of subscriptions, and number of downloads.

[0078] The process of determining the score of candidate resources by electronic devices can be as follows: the electronic device determines at least one recommendation indicator and the weight of the predicted click-through rate; and performs a weighted summation of the values ​​of the at least one recommendation indicator and the predicted click-through rate according to the weight to obtain the score of the candidate resource.

[0079] In particular, by combining the values ​​of at least one recommendation metric with the predicted click-through rate to determine the score of candidate resources, the accuracy of the determined score can be improved, thereby further improving the matching degree between the recommended resources and the content of interest to the target audience, and further improving the accuracy of resource recommendation.

[0080] Step 306: Based on the resources to be recommended, perform resource recommendation processing on the target object.

[0081] It should be noted that for details regarding steps 301 to 302 and steps 306, please refer to [the relevant documentation / reference]. Figure 1 Steps 101 to 102 and step 104 in the illustrated embodiment will not be described in detail here.

[0082] The resource recommendation method of this disclosure involves obtaining a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; determining the interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector representing the resources that the target object is interested in at the current time point; determining the representation vectors of candidate resources in the candidate resource set; determining the predicted click-through rate of candidate resources based on the interest representation vector and the representation vectors of candidate resources; selecting resources to be recommended from the candidate resource set based on the predicted click-through rate; and performing resource recommendation processing on the target object based on the resources to be recommended. Specifically, determining the predicted click-through rate of candidate resources based on the interest representation vector and the representation vectors of candidate resources in the candidate resource set, and then selecting resources to be recommended based on the predicted click-through rate, can further improve the matching degree between the determined resources to be recommended and the content of interest to the target object, thereby further improving the accuracy of resource recommendation. Furthermore, accurate resource recommendation can reduce the number of resource recommendations, thus improving resource recommendation efficiency.

[0083] Figure 4 The diagram is based on the fourth embodiment of this disclosure. It should be noted that the training method of the resource recommendation model in this embodiment can be applied to a training device for the resource recommendation model. This device can be configured in an electronic device so that the electronic device can perform the training function of the resource recommendation model.

[0084] Among them, electronic devices can be any device with computing capabilities, such as personal computers (PCs), mobile terminals, servers, etc. Mobile terminals can be, for example, in-vehicle devices, mobile phones, tablets, personal digital assistants, wearable devices, smart speakers, servers, server clusters, and other hardware devices with various operating systems, touch screens and / or displays.

[0085] The training device for the resource recommendation model can also be software within an electronic device, such as training software for the resource recommendation model. The following embodiments use an electronic device as an example for illustration.

[0086] like Figure 4 As shown, the training method for this resource recommendation model may include the following steps:

[0087] Step 401: Obtain training data; the training data includes the structured content of the historical resources of the samples in the historical resource sequence, the structured content of the recommended resources of the samples, and the clicked resources of the recommended resources of the samples.

[0088] In this embodiment of the disclosure, the sample history resource can be the recommendation processing tag for sample objects, or the historical resources clicked by the sample objects. The sample click resource can be the resource clicked by the sample object in the sample recommendation resource.

[0089] The structured content of the sample historical resources may include at least one metadata element from the sample historical resources. This metadata may include at least one of the following: the resource title, keywords from the resource, the domain to which the resource belongs, and the recipient of the resource.

[0090] Step 402: Obtain the initial resource recommendation model; the resource recommendation model includes an encoding network, a decoding network, and a fully connected network connected in sequence.

[0091] In this embodiment, an encoding network is used to determine the representation vector of the sample historical resources by combining the structured content of the sample historical resources, and to determine the representation vector of the sample recommended resources by combining the structured content of the sample recommended resources. A decoding network is used to determine the interest representation vector of the sample object to which the sample historical resources belong by combining the representation vectors of the sample historical resources in the sample historical resource sequence. A fully connected network is used to determine the predicted click-through rate of the sample recommended resources by combining the interest representation vectors and the representation vectors of the sample recommended resources.

[0092] The encoding network can be equipped with a multi-head attention mechanism to learn the relationships between various metadata in the structured content of the resource, thereby extracting deep features from the structured content and obtaining the representation vector of the resource.

[0093] The decoding network can also be equipped with a multi-head attention mechanism to learn the correlation between the representation vectors of historical resources of each sample, learn the changing trend of the resources of interest of the target object, and thus determine the interest representation vector of the target object.

[0094] Step 403: The initial resource recommendation model is trained using the structured content of the sample historical resources, the structured content of the sample recommended resources, and the sample click resources to obtain the trained resource recommendation model.

[0095] In this embodiment of the disclosure, the electronic device performing step 403 may, for example, involve inputting the structured content of the sample historical resources and the structured content of the sample recommended resources into the encoding network of the resource recommendation model to obtain the representation vectors of the sample historical resources and the sample recommended resources; inputting the representation vectors of each sample historical resource in the sample historical resource sequence into the decoding network of the resource recommendation model to obtain the interest representation vector of the object to which the sample historical resource belongs; inputting the interest representation vector and the representation vector of the sample recommended resources into the fully connected network of the resource recommendation model to obtain the predicted click probability of the sample recommended resources; combining the sample click resources in the sample recommended resources, the predicted click probability of the sample recommended resources, and the loss function of the resource recommendation model to determine the value of the loss function; and combining the value of the loss function to perform parameter adjustment processing on the encoding network, the decoding network, and the fully connected network to obtain the trained resource recommendation model.

[0096] The process of determining the loss function value by combining sample click resources, predicted click probabilities of sample recommended resources, and the loss function of the resource recommendation model can be as follows: For sample recommended resources, the actual click probability of the sample recommended resource can be determined by whether it is a sample click resource; the value of the loss function can be determined by combining the predicted click probability, the actual click probability, and the loss function. The loss function is used to calculate the difference between the predicted click probability and the actual click probability. The smaller the difference, the smaller the value of the loss function; the larger the difference, the larger the value of the loss function.

[0097] Specifically, if the sample recommended resource is a sample click resource, the actual click probability of the sample recommended resource can be determined to be 1; if the sample recommended resource is not a sample click resource, the actual click probability of the sample recommended resource can be determined to be 0.

[0098] Specifically, by combining the structured content of each historical resource in the historical resource sequence with the structured content of the recommended resources, the predicted click probability of the recommended resources is determined. By combining the predicted click probability and the actual click probability of the recommended resources, the parameters of the resource recommendation model are adjusted so that the predicted click probability of the recommended resources determined by the trained resource recommendation model tends to the actual click probability of the recommended resources, thereby improving the accuracy of the trained resource recommendation model.

[0099] The method for training a resource recommendation model according to this embodiment involves acquiring training data, including structured content of historical resources, structured content of recommended resources, and clicked resources in the recommended resources; acquiring an initial resource recommendation model, which includes an encoding network, a decoding network, and a fully connected network connected in sequence; and training the initial resource recommendation model using the structured content of historical resources, the structured content of recommended resources, and clicked resources to obtain a trained resource recommendation model. The training process using the structured content of historical resources, the structured content of recommended resources, and clicked resources enables the resource recommendation model to learn the relationships between historical resources and the interest representation vectors of objects, thereby improving the accuracy of the trained resource recommendation model.

[0100] To implement the above embodiments, this disclosure also provides a resource recommendation device. For example... Figure 5 As shown, Figure 5 This is a schematic diagram according to the fifth embodiment of the present disclosure. The resource recommendation device 50 may include: an acquisition module 501, a determination module 502, a selection module 503, and a recommendation processing module 504.

[0101] The acquisition module 501 is used to acquire a set of candidate resources and a historical resource sequence of the target object within a historical time period prior to the current time point; the determination module 502 is used to determine the interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector represents the resources that the target object is interested in at the current time point; the selection module 503 is used to select resources to be recommended from the set of candidate resources based on the interest representation vector; and the recommendation processing module 504 is used to perform resource recommendation processing on the target object based on the resources to be recommended.

[0102] As one possible implementation of this disclosure, the determining module 502 includes a first determining unit and a second determining unit; the first determining unit is used to determine the representation vector of historical resources in the historical resource sequence; the second determining unit is used to determine the interest representation vector of the target object at the current time point based on the representation vector of each historical resource in the historical resource sequence.

[0103] As one possible implementation of this disclosure, the first determining unit is specifically used to: determine the structured content of historical resources in the historical resource sequence; the structured content includes at least one metadata of the historical resources; input the structured content into the encoding network in the resource recommendation model, and obtain the representation vector of the historical resources output by the encoding network.

[0104] As one possible implementation of this disclosure, the metadata in the structured content includes at least one of the following: the title of the resource, keywords in the resource, the domain to which the resource belongs, and the object to which the resource is published.

[0105] As one possible implementation of this disclosure, the second determining unit is specifically used to: determine a representation vector sequence based on the representation vectors of each historical resource in the historical resource sequence; input the representation vector sequence into the decoding network in the resource recommendation model to obtain the interest representation vector output by the decoding network.

[0106] As one possible implementation of this disclosure, the determining module 502 further includes: a partitioning processing unit, a third determining unit, a splicing processing unit, and an updating processing unit; the partitioning processing unit is used to partition each historical resource in the historical resource sequence according to the collection time point to obtain multiple historical resource combinations; the third determining unit is used to determine the combination number of the historical resource combination and the number vector corresponding to the combination number; the splicing processing unit is used to splice the number vector corresponding to the combination number of the historical resource combination to which the historical resource belongs and the representation vector of the historical resource for each historical resource in the historical resource sequence to obtain a processed representation vector; the updating processing unit is used to update the representation vector of the historical resource according to the processed representation vector.

[0107] As one possible implementation of this disclosure, the acquisition module 501 is specifically used to: perform resource retrieval processing and resource filtering processing based on the attribute information of the target object to obtain the candidate resource set; acquire the historical resources clicked by the target object within a historical time period before the current time point, and the click time points of the historical resources; and sort each of the historical resources according to the click time points to obtain the historical resource sequence.

[0108] As one possible implementation of this disclosure, the acquisition module 501 is further configured to: perform resource recall processing based on the attribute information of the target object to obtain a recalled resource set; combine the attribute information and a first click-through rate (CTR) model to determine a first CTR of the recalled resources in the recalled resource set; perform filtering processing on the recalled resources in the recalled resource set based on the first CTR to obtain a filtered resource set; combine the attribute information and a second CTR model to determine a second CTR of the filtered resources in the filtered resource set; the number of parameters in the second CTR model is greater than the number of parameters in the first CTR model; and perform filtering processing on the filtered resources in the filtered resource set based on the second CTR to obtain the candidate resource set.

[0109] As one possible implementation of this disclosure, the selection module 503 includes a fourth determining unit, a fifth determining unit, and a selection unit; the fourth determining unit is used to determine the representation vector of the candidate resource in the candidate resource set; the fifth determining unit is used to determine the predicted click-through rate of the candidate resource based on the interest representation vector and the representation vector of the candidate resource; the selection unit is used to select the resource to be recommended from the candidate resource set based on the predicted click-through rate.

[0110] As one possible implementation of this disclosure, the fifth determining unit is specifically used to input the interest representation vector and the representation vector of the candidate resource into a fully connected network in the resource recommendation model, and obtain the predicted click-through rate output by the fully connected network.

[0111] As one possible implementation of this disclosure, the selection unit is specifically used to: sort the candidate resources in the candidate resource set in descending order according to the predicted click-through rate to obtain a sorting result; and determine at least one candidate resource ranked first in the sorting result as the resource to be recommended.

[0112] As one possible implementation of this disclosure, the selection unit is specifically configured to: determine the value of the candidate resource in the candidate resource set on at least one recommendation metric; determine the score of the candidate resource based on the value of the candidate resource on at least one recommendation metric and the predicted click-through rate; sort the candidate resources in the candidate resource set in descending order based on the score to obtain a sorting result; and determine at least one candidate resource ranked first in the sorting result as the resource to be recommended.

[0113] The resource recommendation device of this embodiment acquires a candidate resource set and a historical resource sequence of a target object within a historical time period prior to the current time point; determines an interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector represents the resources that the target object is interested in at the current time point; selects resources to be recommended from the candidate resource set based on the interest representation vector; and performs resource recommendation processing on the target object based on the resources to be recommended. The interest representation vector determined by combining the historical resource sequence can represent the resources that the target object is interested in at the current time point, making the determined resources to be recommended the resources that the target object currently wants to click, thereby improving the matching degree between the resources to be recommended and the content of interest to the target object, and improving the accuracy of resource recommendation; moreover, accurate resource recommendation can reduce the number of resource recommendations, thereby improving resource recommendation efficiency.

[0114] To implement the above embodiments, this disclosure also provides a training apparatus for a resource recommendation model. For example... Figure 6 As shown, Figure 6 This is a schematic diagram according to the sixth embodiment of the present disclosure. The training device 60 for the resource recommendation model may include: a first acquisition module 601, a second acquisition module 602, and a training processing module 603.

[0115] The system includes a first acquisition module 601 for acquiring training data, which includes structured content of historical resources, structured content of recommended resources, and click resources in the recommended resources. A second acquisition module 602 is used to acquire an initial resource recommendation model, which includes a sequentially connected encoding network, a decoding network, and a fully connected network. A training processing module 603 is used to train the initial resource recommendation model using the structured content of the historical resources, the structured content of the recommended resources, and the click resources to obtain a trained resource recommendation model.

[0116] As one possible implementation of this disclosure, the training processing module 603 is specifically configured to: input the structured content of the sample historical resources and the structured content of the sample recommended resources into the encoding network of the resource recommendation model to obtain the representation vectors of the sample historical resources and the sample recommended resources; input the representation vectors of each sample historical resource in the sample historical resource sequence into the decoding network of the resource recommendation model to obtain the interest representation vector of the object to which the sample historical resource belongs; input the interest representation vector and the representation vector of the sample recommended resource into the fully connected network of the resource recommendation model to obtain the predicted click probability of the sample recommended resource; combine the sample click resources in the sample recommended resources, the predicted click probability of the sample recommended resource, and the loss function of the resource recommendation model to determine the value of the loss function; and combine the value of the loss function to perform parameter adjustment processing on the encoding network, the decoding network, and the fully connected network to obtain the trained resource recommendation model.

[0117] The training apparatus for the resource recommendation model in this embodiment acquires training data, including structured content of historical resources, structured content of recommended resources, and clicked resources in the recommended resources. It then acquires an initial resource recommendation model, which includes a sequentially connected encoding network, decoding network, and fully connected network. The initial resource recommendation model is trained using the structured content of historical resources, the structured content of recommended resources, and the clicked resources to obtain a trained resource recommendation model. This training process enables the resource recommendation model to learn the relationships between historical resources and the interest representation vectors of objects, thereby improving the accuracy of the trained resource recommendation model.

[0118] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information are all carried out with the consent of the users, and all comply with the provisions of relevant laws and regulations, and do not violate public order and good morals.

[0119] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0120] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments 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 may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0121] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0122] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0123] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose 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. The computing unit 701 performs the various methods and processes described above, such as resource recommendation methods or resource recommendation model training methods. For example, in some embodiments, the resource recommendation method or resource recommendation model training method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the resource recommendation method or resource recommendation model training method described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured in any other suitable manner (e.g., by means of firmware) to perform a resource recommendation method or a training method for a resource recommendation model.

[0124] Various embodiments 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-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0125] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0126] In the context of this disclosure, a machine-readable medium can 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 can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0127] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, 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 sound input, voice input, or tactile input).

[0128] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0129] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0130] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0131] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A resource recommendation method, the method comprising: Obtain a set of candidate resources and a sequence of historical resources of the target object within a historical time period prior to the current time point. The set of candidate resources is determined based on the attribute information of the target object. Based on the historical resource sequence, determine the interest representation vector of the target object at the current time point; the interest representation vector represents the resources that the target object is interested in at the current time point. Determine the representation vector of the candidate resource in the candidate resource set, wherein the structured content of the candidate resource is determined, the structured content includes at least one metadata of the candidate resource, the structured content of the candidate resource is input into the encoding network in the resource recommendation model, and the representation vector of the candidate resource output by the encoding network is obtained. The interest representation vector and the candidate resource representation vector are input into the fully connected network in the resource recommendation model to obtain the predicted click-through rate output by the fully connected network. Based on the predicted click-through rate, resources to be recommended are selected from the candidate resource set; Based on the resources to be recommended, resource recommendation processing is performed on the target object; Determining the interest representation vector of the target object at the current time point based on the historical resource sequence includes: Determine the representation vectors of historical resources in the historical resource sequence; wherein, the representation vectors include: The historical resources in the historical resource sequence are divided according to the collection time point to obtain multiple combinations of historical resources; Determine the combination number of the historical resource combination, and the number vector corresponding to the combination number; For each historical resource in the historical resource sequence, the number vector corresponding to the combination number of the historical resource combination to which the historical resource belongs is concatenated with the representation vector of the historical resource to obtain the processed representation vector. The representation vector of the historical resource is updated based on the processed representation vector. Based on the representation vectors of each historical resource in the historical resource sequence, determine the interest representation vector of the target object at the current time point; wherein, it includes: determining a representation vector sequence based on the representation vectors of each historical resource in the historical resource sequence; The sequence of representation vectors is input into the decoding network in the resource recommendation model to obtain the interest representation vector output by the decoding network; wherein, the decoding network is used to learn the correlation between the representation vectors of historical resources of each sample, as well as the changing trend of the resources of interest of the target object.

2. The method according to claim 1, wherein, Determining the representation vector of historical resources in the historical resource sequence includes: Determine the structured content of historical resources in the historical resource sequence; the structured content includes at least one metadata element from the historical resource. The structured content is input into the encoding network of the resource recommendation model to obtain the representation vector of the historical resource output by the encoding network.

3. The method according to claim 2, wherein, The metadata in the structured content includes at least one of the following: the title of the resource, the keywords in the resource, the domain to which the resource belongs, and the target audience for the resource.

4. The method according to claim 1, wherein, The acquisition of the candidate resource set and the historical resource sequence of the target object within the historical time period prior to the current time point includes: Based on the attribute information of the target object, resource retrieval and resource filtering processes are performed to obtain the candidate resource set. Obtain the historical resources clicked by the target object within a historical time period prior to the current time point, as well as the click time points of the historical resources; The historical resources are sorted according to the click time to obtain the historical resource sequence.

5. The method according to claim 4, wherein, The step of performing resource retrieval and resource filtering based on the attribute information of the target object to obtain the candidate resource set includes: Based on the attribute information of the target object, resource retrieval processing is performed to obtain a set of recalled resources; By combining the attribute information and the first click-through rate model, the first click-through rate of the recalled resources in the recalled resource set is determined; Based on the first click-through rate, the recalled resources in the recalled resource set are filtered to obtain a filtered resource set; Combining the attribute information and the second click-through rate model, the second click-through rate of the filtered resources in the filtered resource set is determined; the number of parameters in the second click-through rate model is greater than the number of parameters in the first click-through rate model. Based on the second click rate, the filtered resources in the filtered resource set are filtered to obtain the candidate resource set.

6. The method according to claim 1, wherein, The step of selecting the resource to be recommended from the candidate resource set based on the predicted click-through rate includes: Based on the predicted click-through rate, the candidate resources in the candidate resource set are sorted in descending order to obtain the sorting result; At least one candidate resource ranked first in the sorting results is identified as the resource to be recommended.

7. The method according to claim 1, wherein, The step of selecting the resource to be recommended from the candidate resource set based on the predicted click-through rate includes: Determine the value of the candidate resource in the candidate resource set on at least one recommendation metric; The score of the candidate resource is determined based on the value of the candidate resource on at least one recommendation metric and the predicted click-through rate. Based on the scores, the candidate resources in the candidate resource set are sorted in descending order to obtain the sorting result; At least one candidate resource ranked first in the sorting results is identified as the resource to be recommended.

8. A method for training a resource recommendation model, the method comprising: Obtain training data; The training data includes structured content of historical resources in the sample historical resource sequence, structured content of recommended resources, and sample click resources in the recommended resources. The structured content of the historical resources includes at least one metadata of the historical resources. The metadata in the structured content includes at least one of the following: the title of the resource, the keywords in the resource, the domain of the resource, and the target audience of the resource. Obtain an initial resource recommendation model; the resource recommendation model includes an encoding network, a decoding network, and a fully connected network connected in sequence; The initial resource recommendation model is trained using the structured content of the sample historical resources, the structured content of the sample recommendation resources, and the sample click resources to obtain a trained resource recommendation model. The process of training the initial resource recommendation model using the structured content of the sample historical resources, the structured content of the sample recommendation resources, and the structured content of the sample click resources to obtain the trained resource recommendation model includes: The structured content of the sample historical resources and the structured content of the sample recommended resources are respectively input into the encoding network in the resource recommendation model to obtain the representation vector of the sample historical resources and the representation vector of the sample recommended resources. The representation vectors of each historical resource in the sample historical resource sequence are input into the decoding network in the resource recommendation model to obtain the interest representation vector of the object to which the sample historical resource belongs. The decoding network is used to learn the correlation between the representation vectors of each historical resource and the changing trend of the target object's interest resources. The interest representation vector and the representation vector of the sample recommended resource are input into the fully connected network in the resource recommendation model to obtain the predicted click probability of the sample recommended resource. The value of the loss function is determined by combining the sample click resources in the sample recommended resources, the predicted click probability of the sample recommended resources, and the loss function of the resource recommendation model; By combining the loss function values, the parameters of the encoding network, the decoding network, and the fully connected network are adjusted to obtain the trained resource recommendation model.

9. A resource recommendation device, the device comprising: The acquisition module is used to acquire a set of candidate resources and the historical resource sequence of the target object within a historical time period before the current time point; The determination module is used to determine the interest representation vector of the target object at the current time point based on the historical resource sequence; the interest representation vector represents the resources that the target object is interested in at the current time point. The selection module is used to select resources to be recommended from the candidate resource set based on the interest representation vector. The recommendation processing module is used to perform resource recommendation processing on the target object based on the resource to be recommended; The selection module includes a fourth determining unit, a fifth determining unit, and a selection unit. The fourth determining unit is used to determine the representation vector of the candidate resource in the candidate resource set, wherein the structured content of the candidate resource is determined, the structured content includes at least one metadata of the candidate resource, the structured content of the candidate resource is input into the encoding network in the resource recommendation model, and the representation vector of the candidate resource output by the encoding network is obtained. The fifth determining unit is used to input the interest representation vector and the representation vector of the candidate resource into the fully connected network in the resource recommendation model to obtain the predicted click-through rate output by the fully connected network. The selection unit is used to select the resource to be recommended from the candidate resource set based on the predicted click-through rate; The determining module includes a first determining unit and a second determining unit; The first determining unit is used to determine the representation vector of the historical resources in the historical resource sequence; The second determining unit is used to determine the interest representation vector of the target object at the current time point based on the representation vector of each historical resource in the historical resource sequence. The second determining unit is specifically used for, Based on the representation vectors of each historical resource in the historical resource sequence, determine the representation vector sequence; The sequence of representation vectors is input into the decoding network in the resource recommendation model to obtain the interest representation vector output by the decoding network; wherein, the decoding network is used to learn the correlation between the representation vectors of historical resources of each sample, as well as the changing trend of the resources of interest of the target object; The determining module further includes: a division processing unit, a third determining unit, a splicing processing unit, and an update processing unit; The partitioning processing unit is used to partition each historical resource in the historical resource sequence according to the collection time point to obtain multiple historical resource combinations. The third determining unit is used to determine the combination number of the historical resource combination and the number vector corresponding to the combination number; The splicing processing unit is used to splice the number vector corresponding to the combination number of the historical resource combination to which the historical resource belongs and the representation vector of the historical resource for each historical resource in the historical resource sequence, so as to obtain the processed representation vector. The update processing unit is used to update the representation vector of the historical resource according to the processed representation vector.

10. The apparatus according to claim 9, wherein, The first determining unit is specifically used for, Determine the structured content of historical resources in the historical resource sequence; the structured content includes at least one metadata element from the historical resource. The structured content is input into the encoding network of the resource recommendation model to obtain the representation vector of the historical resource output by the encoding network.

11. The apparatus according to claim 10, wherein, The metadata in the structured content includes at least one of the following: the title of the resource, the keywords in the resource, the domain to which the resource belongs, and the target audience for the resource.

12. The apparatus according to claim 9, wherein, The acquisition module is specifically used for, Based on the attribute information of the target object, resource retrieval and resource filtering processes are performed to obtain the candidate resource set. Obtain the historical resources clicked by the target object within a historical time period prior to the current time point, as well as the click time points of the historical resources; The historical resources are sorted according to the click time to obtain the historical resource sequence.

13. The apparatus according to claim 12, wherein, The acquisition module is further specifically used for, Based on the attribute information of the target object, resource retrieval processing is performed to obtain a set of recalled resources; By combining the attribute information and the first click-through rate model, the first click-through rate of the recalled resources in the recalled resource set is determined; Based on the first click-through rate, the recalled resources in the recalled resource set are filtered to obtain a filtered resource set; Based on the attribute information and the second click-through rate model, the second click-through rate of the filtered resources in the filtered resource set is determined; The second click-through rate model has more parameters than the first click-through rate model; Based on the second click rate, the filtered resources in the filtered resource set are filtered to obtain the candidate resource set.

14. The apparatus according to claim 9, wherein, The selection unit is specifically used for, Based on the predicted click-through rate, the candidate resources in the candidate resource set are sorted in descending order to obtain the sorting result; At least one candidate resource ranked first in the sorting results is identified as the resource to be recommended.

15. The apparatus according to claim 9, wherein, The selection unit is specifically used for, Determine the value of the candidate resource in the candidate resource set on at least one recommendation metric; The score of the candidate resource is determined based on the value of the candidate resource on at least one recommendation metric and the predicted click-through rate. Based on the scores, the candidate resources in the candidate resource set are sorted in descending order to obtain the sorting result; At least one candidate resource ranked first in the sorting results is identified as the resource to be recommended.

16. A training apparatus for a resource recommendation model, the apparatus comprising: The first acquisition module is used to acquire training data; The training data includes structured content of historical resources in the sample historical resource sequence, structured content of recommended resources, and sample click resources in the recommended resources. The structured content of the historical resources includes at least one metadata of the historical resources. The metadata in the structured content includes at least one of the following: the title of the resource, the keywords in the resource, the domain of the resource, and the target audience of the resource. The second acquisition module is used to acquire an initial resource recommendation model; the resource recommendation model includes an encoding network, a decoding network, and a fully connected network connected in sequence. The training processing module is used to train the initial resource recommendation model using the structured content of the sample historical resources, the structured content of the sample recommendation resources, and the sample click resources, to obtain the trained resource recommendation model; specifically, the training processing module is used for... The structured content of the sample historical resources and the structured content of the sample recommended resources are respectively input into the encoding network in the resource recommendation model to obtain the representation vector of the sample historical resources and the representation vector of the sample recommended resources. The representation vectors of each historical resource in the sample historical resource sequence are input into the decoding network in the resource recommendation model to obtain the interest representation vector of the object to which the sample historical resource belongs. The decoding network is used to learn the correlation between the representation vectors of each historical resource and the changing trend of the target object's interest resources. The interest representation vector and the representation vector of the sample recommended resource are input into the fully connected network in the resource recommendation model to obtain the predicted click probability of the sample recommended resource. The value of the loss function is determined by combining the sample click resources in the sample recommended resources, the predicted click probability of the sample recommended resources, and the loss function of the resource recommendation model; By combining the loss function values, the parameters of the encoding network, the decoding network, and the fully connected network are adjusted to obtain the trained resource recommendation model.

17. 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, which, when executed by the at least one processor, enable the at least one processor to perform the method of any one of claims 1 to 7; Alternatively, the method described in claim 8 may be performed.

18. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7; or to perform the method according to claim 8.

19. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7; or implements the method according to claim 8.

Images