Resource recommendation method, model training method and equipment based on large model

Abstract

The invention provides a resource recommendation method and device based on a large model, a model training method and device and electronic equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, large models, big data, intelligent recommendation and the like. According to the specific implementation scheme, a first resource interested by a target object is determined from historical recommended resources of the target object; determining a second resource to be recommended to the target object this time, and obtaining a third resource recommended to the target object last time; determining a first target representation vector of a resource in which the target object is interested according to the first resource, the second resource and the third resource by using a resource representation generation model; determining a fourth resource matched with the first target representation vector from the candidate resource set; and performing information flow recommendation processing according to the fourth resource. Therefore, the recommended resources can meet the resource demand of the target object, and the accuracy of resource recommendation is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as deep learning, big models, big data, and intelligent recommendation, and in particular to a resource recommendation method, model training method, device, and electronic device based on a big model. Background Art

[0002] With the rapid development of science and technology, various media information service terminals such as the Internet or mobile networks generate massive amounts of information resources every day and present them to users. Therefore, how to select the resources that users are interested in from these information resources and recommend them to users is a very important issue. Summary of the invention

[0003] The present disclosure provides a resource recommendation method based on a large model, a model training method, a device, and an electronic device.

[0004] According to one aspect of the present disclosure, a resource recommendation method based on a large model is provided, the method comprising: determining a first resource of interest to the target object from historically recommended resources of the target object; determining a second resource to be recommended to the target object this time, and obtaining a third resource recommended to the target object last time; using a resource representation generation model to determine a first target representation vector of the resource of interest to the target object based on the first resource, the second resource and the third resource; determining a fourth resource matching the first target representation vector from a set of candidate resources; and performing information flow recommendation processing this time based on the fourth resource.

[0005] According to another aspect of the present disclosure, a method for training a resource representation generation model is provided, comprising: determining a fifth resource of interest to the sample object from historically recommended resources of the sample object; acquiring a sixth resource pushed to the sample object for the jth time, wherein j is an integer greater than or equal to 1; acquiring a seventh resource recommended to the sample object for the j+1th time; determining an eighth resource of interest to the sample object from the seventh resource; determining a second target representation vector of the resource of interest to the sample object based on the fifth resource, the sixth resource and the seventh resource using an initial resource representation generation model; and training the initial resource representation generation model based on the representation vector of the eighth resource and the second target representation vector.

[0006] According to another aspect of the present disclosure, a resource recommendation device based on a large model is provided, the device comprising: a first determination module, used to determine a first resource of interest to the target object from historically recommended resources of the target object; a processing module, used to determine a second resource to be recommended to the target object this time, and obtain a third resource recommended to the target object last time; a second determination module, used to determine a first target representation vector of the resource of interest to the target object based on the first resource, the second resource and the third resource by using a resource representation generation model; a third determination module, used to determine a fourth resource matching the first target representation vector from a set of candidate resources; and a recommendation module, used to perform this information flow recommendation processing based on the fourth resource.

[0007] According to another aspect of the present disclosure, a training device for a resource representation generation model is provided, comprising: a first determination module, used to determine a fifth resource of interest to the sample object from historically recommended resources of the sample object; a first acquisition module, used to obtain a sixth resource pushed to the sample object for the jth time, wherein j is an integer greater than or equal to 1; a second acquisition module, used to obtain a seventh resource recommended to the sample object for the j+1th time; a second determination module, used to determine an eighth resource of interest to the sample object from the seventh resource; a third determination module, used to determine a second target representation vector of the resource of interest to the sample object based on the fifth resource, the sixth resource and the seventh resource using the initial resource representation generation model; and a training module, used to train the initial resource representation generation model based on the representation vector of the eighth resource and the second target representation vector.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the large model-based resource recommendation method proposed in the present disclosure.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the large model-based resource recommendation method proposed above in the present disclosure, or the resource representation generation model training method.

[0010] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the resource recommendation method proposed in the present disclosure, or the training method of the resource representation generation model.

[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0013] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0014] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure;

[0015] Figure 3 It is a schematic diagram of the feature vector of the resource that the resource representation generation model determines the target object's interest;

[0016] Figure 4 is a schematic diagram according to a third embodiment of the present disclosure;

[0017] Figure 5 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0018] Figure 6 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0019] Figure 7 is a schematic diagram according to a sixth embodiment of the present disclosure;

[0020] Figure 8 is a schematic diagram according to a seventh embodiment of the present disclosure;

[0021] Fig. 9 is a schematic block diagram of an electronic device 900 for implementing an embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] Figure 1 It is a schematic diagram according to the first embodiment of the present disclosure. It should be noted that the resource recommendation method of the embodiment of the present disclosure can be applied to a resource recommendation device based on a large model, and the device can be configured in an electronic device so that the electronic device can perform a resource recommendation function.

[0024] The electronic device may be any device with computing capability, for example, the electronic device may be a server, a server set, etc. It should be noted that in the following embodiments, the execution subject is an electronic device as an example for exemplary description.

[0025] like Figure 1 As shown, the resource recommendation method based on the large model may include the following steps:

[0026] Step 101: Determine a first resource that the target object is interested in from historically recommended resources for the target object.

[0027] It should be noted that the resources in this embodiment may be various types of resources, such as video, audio, image, article, commodity, etc., and this embodiment does not specifically limit this.

[0028] The target object refers to the object to which resource recommendation is to be performed, for example, the target object may be the user to whom resource recommendation is to be performed.

[0029] In one embodiment of the present disclosure, in order to accurately obtain the resources that the target object is interested in, a possible implementation method for determining the first resource that the target object is interested in from the historically recommended resources of the target object is: obtaining the target object's stay time in the historically recommended resources; and taking the resource whose stay time in the historically recommended resources exceeds a preset time threshold as the first resource. Thus, by analyzing the target object's stay time on the historically recommended resources, the first resource that the target object is interested in can be accurately determined from the historically recommended resources.

[0030] It can be understood that, generally, the longer the target object stays on a historically recommended resource, the higher the target object's interest in the historically recommended resource, and vice versa.

[0031] The preset duration threshold is a duration threshold that is preset according to actual needs, and this embodiment does not specifically limit this.

[0032] In some examples, in order to determine the resources that the target object is interested in recently, the historically recommended resources of the target object within the latest preset time period may be obtained accordingly. In other words, the resources recommended to the target object within the latest preset time period are obtained.

[0033] The most recent preset duration refers to the preset duration before the current time.

[0034] Among them, the length of the most recent preset time can be set and adjusted by the technical staff according to the actual application scenario. For example, the most recent preset time can be the last 1 month, or 2 months, etc. This embodiment does not make specific limitations here.

[0035] Step 102: determine the second resource to be recommended to the target object this time, and obtain the third resource recommended to the target object last time.

[0036] In this embodiment, the second resource to be recommended to the target object this time may be determined based on a recommendation method in related technologies.

[0037] Step 103: using the resource representation generation model, according to the first resource, the second resource and the third resource, determine a first target representation vector of the resource that the target object is interested in.

[0038] In this embodiment, the first resource, the second resource, and the third resource are analyzed by using a resource representation generation model, so as to determine a first target representation vector of the resource that the target object is interested in in this recommendation.

[0039] Among them, the resource representation generation model can be a large model.

[0040] Step 104: Determine a fourth resource that matches the first target representation vector from the candidate resource set.

[0041] In one embodiment of the present disclosure, the matching degree between the first target characterization vector and the characterization vectors of each resource in the candidate resource set may be determined, and a resource whose matching degree exceeds a preset matching degree threshold may be selected from the candidate resource set as the fourth resource.

[0042] The preset matching degree threshold is a matching degree value preset according to actual needs, and this embodiment does not specifically limit the preset matching degree threshold.

[0043] In one embodiment of the present disclosure, the degree of match between the first target representation vector and the representation vectors of each resource in the candidate resource set can be determined, and the resources are sorted in descending order according to the degree of match, and the top N resources in the sorting result are used as the fourth resources, where N is an integer greater than or equal to 1.

[0044] It should be noted that the above-mentioned candidate resource set may include all resources or part of the resources in the preset resource pool.

[0045] As an example, the candidate resource set may include: resources obtained by recalling resources from a preset resource pool. In other words, based on the resource preference of the target object, resources may be recalled from a preset resource pool to obtain the candidate resource set.

[0046] As another example, the candidate resource set may include: resources obtained after performing resource recall processing and rough sorting processing on the resource pool in sequence. That is, based on the resource preference of the target object, resources may be recalled from the preset resource pool, and the recalled resources may be roughly sorted to obtain the candidate resource set.

[0047] As another example, the candidate resource set may include: resources obtained after performing resource recall processing, rough sorting processing, and fine sorting processing on the resource pool in sequence. That is, according to the resource preference of the target object, the preset resource pool is recalled, the recalled resources are roughly sorted, and the resources obtained by the rough sorting are finely sorted to obtain the candidate resource set. Since in this example, the candidate resource set is formed based on the resources obtained after the fine sorting processing, the resources in the preset resource pool that are not related to the resource preference of the target object are filtered out, so that the number of resources included in the candidate resource set is greatly reduced, which can reduce the amount of calculation when determining the fourth resource, improve the efficiency of obtaining the fourth resource, and then improve the efficiency of subsequent resource recommendations.

[0048] Step 105: perform information flow recommendation processing according to the fourth resource.

[0049] In this embodiment, the currently recommended information flow may be formed according to the fourth resource, and the currently recommended information flow may be sent to the target object.

[0050] Based on the above description, it can be seen that in this embodiment, after obtaining the second resource to be recommended to the target object, because the target object's viewing behavior of the third resource recommended last time will also affect the target object's viewing of the resource recommended this time, the resource representation generation model in this embodiment combines the first resource of historical interest to the target object and the third resource recommended to the target object last time to analyze the second resource, thereby estimating the representation vector of the resource that the target object is interested in at the time of this recommendation, and determining the fourth resource that matches the representation vector from the set of candidate resources, and performing this information flow recommendation processing based on the fourth resource, thereby recommending the resources of interest to the target object, thereby improving the accuracy of resource recommendation.

[0051] The resource recommendation method based on a large model of the disclosed embodiment determines the first resource that the target object is interested in from the historically recommended resources of the target object; determines the second resource to be recommended to the target object this time, and obtains the third resource recommended to the target object last time; uses the resource representation generation model to determine the first target representation vector of the resource that the target object is interested in based on the first resource, the second resource, and the third resource; determines the fourth resource that matches the first target representation vector from the candidate resource set; and performs information flow recommendation processing this time based on the fourth resource. Thus, through the resource representation generation model, the first resource that the target object is interested in, the second resource to be recommended to the target object this time, and the third resource recommended to the target object last time are analyzed, and the representation vector of the resource that the target object is interested in during this recommendation can be determined, and the fourth resource is determined from the candidate resource set based on the representation vector, and information flow recommendation processing is performed based on the fourth resource, which can make the recommended resources meet the resource needs of the target object and improve the accuracy of resource recommendation.

[0052] In order to further improve the accuracy of the first target representation vector determined by the resource representation generation model, the resource representation generation model can combine the object feature of the target object, the first resource, the first object feature, the second resource and the third resource to determine the first target representation vector of the resource of interest to the target object. Figure 2 This process is described exemplarily.

[0053] Figure 2 is a schematic diagram according to a second embodiment of the present disclosure.

[0054] like Figure 2 As shown, the resource recommendation method based on the large model includes the following steps:

[0055] Step 201 : determining a first resource that the target object is interested in from historically recommended resources for the target object.

[0056] It should be noted that, for the specific description of step 201, reference may be made to the relevant descriptions in other embodiments, which will not be repeated here.

[0057] Step 202: determine the second resource to be recommended to the target object this time, and obtain the third resource recommended to the target object last time.

[0058] In one embodiment of the present disclosure, a possible implementation method for determining the second resource to be recommended to the target object this time may be: according to the resource preference of the target object, resources are recalled from a preset resource pool; the recalled resources are roughly sorted, and the resources obtained by the rough sorting are finely sorted to obtain a candidate resource set, and the candidate resource set is fused and sorted to obtain a fused sorting result; the top K resources in the fused sorting result are used as the second resource, where K is an integer greater than or equal to 1. Thus, in combination with the resource preference of the target object, the second resource to be recommended to the target object this time is accurately determined through resource recall, rough sorting, fine sorting, and fused sorting, so that the determined second resource matches the resource preference, and the accuracy of subsequent resource recommendation can be further improved.

[0059] Step 203: Acquire a first object feature of the target object.

[0060] The first object feature is used to describe and characterize various attributes and behavior data of the target object.

[0061] Step 204 , using the resource representation generation model, according to the first resource, the first object feature, the second resource and the third resource, a first target representation vector of the resource of interest to the target object is determined.

[0062] In one embodiment of the present disclosure, in the case where the resource representation generation model includes an encoding network and a decoding network, a possible implementation method of using the resource representation generation model to determine the feature vector of the resource of interest to the target object according to the first resource, the first object feature, the second resource and the third resource is as follows: inputting the representation vector of the first resource and the representation vector of the first object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the first resource and the representation vector of the first object feature through the encoding network, and obtain the feature vector of the first resource and the feature vector of the first object feature; inputting the feature vector of the first resource, the feature vector of the first object feature, the representation vector of the second resource and the representation vector of the third resource into the decoding network, so as to determine the first target representation vector of the resource of interest to the target object. Thus, performing attention cross-learning on the representation vector of the first resource and the representation vector of the first object feature through the encoding network can help the model focus on the most critical part of the object feature and the first resource, and inputting the content output by the encoding network into the decoding network, so that the decoding network can more accurately determine the first target representation vector of the resource of interest to the target object in this recommendation according to the representation vector of the second resource, the representation vector of the third resource, and the output content of the encoding network.

[0063] The characterization vector of the first resource refers to a vectorized representation of the features of the first resource.

[0064] The characteristics of the first resource may include characteristics of the first resource in at least one characteristic dimension.

[0065] Among them, when there are multiple feature dimensions, the features of the first resource in each feature dimension can be vectorized and the representation vectors of the features in each feature dimension can be concatenated to obtain the representation vector of the first resource.

[0066] Among them, at least one characteristic dimension may include at least one of the following dimensions: semantic dimension, resource attribute dimension, stay time dimension, interaction dimension, etc.

[0067] The representation vector of the object feature refers to the vectorized representation of the object feature.

[0068] The characterization vector of the second resource refers to a vectorized representation of the characteristics of the second resource.

[0069] The characteristics of the second resource may include characteristics of the second resource in at least one characteristic dimension.

[0070] Among them, when there are multiple feature dimensions, the features of the second resource in each feature dimension can be vectorized and the representation vectors of the features in each feature dimension can be concatenated to obtain the representation vector of the second resource.

[0071] Among them, at least one characteristic dimension may include at least one of the following dimensions: semantic dimension, resource attribute dimension, stay time dimension, interaction dimension, etc.

[0072] The characterization vector of the third resource refers to a vectorized representation of the characteristics of the third resource.

[0073] An example process in which the resource representation generation model combines the first resource, the first object feature, the second resource, and the third resource to determine a feature vector of a resource of interest to the target object is as follows: Figure 3 As shown. It should be noted that Figure 3In the example, the number of first resources is 30, and the number of second resources and third resources are both 6. In this example, the first representation sequence can be obtained by vectorizing the features of each resource in the first resource sequence, wherein the first resource sequence is obtained by sorting the 30 first resources. Specifically, the representation vector of the first object feature and the first representation sequence are input into the encoding network of the resource representation generation model, so that the first representation vector and the representation vector in the first representation sequence are cross-learned through the encoding network to obtain the feature vector and feature sequence of the object feature, wherein the feature sequence includes the feature vectors of each representation vector in the first representation sequence. Then, the second representation sequence, the third representation sequence, the feature vector and the feature sequence of the object feature are input into the decoding network of the resource representation generation model to obtain the first target representation vector sequence of the resources that the target object is interested in in this recommendation, wherein the second representation sequence is obtained by vectorizing the features of each resource in the second resource sequence. Among them, the third representation sequence is obtained by vectorizing the features of each resource in the third resource sequence. Among them, the second resource sequence is obtained by sorting 6 second resources, and the third resource sequence is obtained by sorting 6 third sequences. Among them, it should be noted that, Figure 3 X in user The representation vector representing the object features, H user The feature vector representing the characteristics of the object, X nidj represents the i-th representation vector in the first representation sequence, H nidj represents the jth eigenvector in the feature sequence, where the value of j ranges from 1 to 30; C nidk represents the kth representation vector in the second representation sequence, S nidk represents the kth representation vector in the third representation sequence, O nidk represents the kth target representation vector in the first target representation vector sequence, where the value of k ranges from 1 to 6.

[0074] Step 205: Determine a fourth resource that matches the first target representation vector from the candidate resource set.

[0075] Step 206: perform information flow recommendation processing according to the fourth resource.

[0076] It should be noted that, for the specific description of step 205 to step 206, reference may be made to the relevant description in other embodiments, which will not be repeated here.

[0077] In this embodiment, through the resource representation generation model, the object characteristics of the target object, the first resource of interest to the target object, the second resource to be recommended to the target object this time, and the third resource recommended to the target object last time are analyzed, and the representation vector of the resource that the target object is interested in this time can be determined, and the fourth resource is determined from the candidate resource set based on the representation vector, and information flow recommendation processing is performed based on the fourth resource, which can make the recommended resources more in line with the resource needs of the target object and improve the accuracy of resource recommendation.

[0078] Figure 4 is a schematic diagram according to a third embodiment of the present disclosure.

[0079] like Figure 4 As shown, the method may include:

[0080] Step 401: determine the first resource that the target object is interested in from the historically recommended resources of the target object.

[0081] Step 402: recall resources from a preset resource pool according to the resource preference of the target object.

[0082] Step 403: perform rough sorting on the recalled resources.

[0083] Step 404, performing fine sorting on the resources obtained by the rough sorting process to obtain a candidate resource set.

[0084] Step 405: perform fusion sorting processing on the candidate resource set to obtain a fusion sorting result.

[0085] Step 406: Use the top K resources in the fusion sorting result as the second resources, where K is an integer greater than or equal to 1.

[0086] Step 407: Obtain the third resource recommended to the target object last time.

[0087] Step 408: Acquire a first object feature of the target object.

[0088] Step 409: Input the representation vector of the first resource and the representation vector of the first object feature into the encoding network of the resource representation generation model, so as to perform attention cross-learning on the representation vector of the first resource and the representation vector of the first object feature through the encoding network to obtain the feature vector of the first resource and the feature vector of the first object feature.

[0089] Step 410, input the feature vector of the first resource, the feature vector of the first object feature, the representation vector of the second resource and the representation vector of the third resource into a decoding network of a resource representation generation model to determine a first target representation vector of a resource of interest to the target object.

[0090] Step 411: Determine a fourth resource matching the first target representation vector from the candidate resource set.

[0091] Step 412: perform information flow recommendation processing according to the fourth resource.

[0092] In order to implement the above embodiments, the present disclosure also provides a method for training a resource representation generation model.

[0093] Figure 5 It is a schematic diagram according to the fourth embodiment of the present disclosure. It should be noted that the training method of the resource representation generation model of the embodiment of the present disclosure can be applied to a training device for the resource representation generation model, and the device can be configured in an electronic device so that the electronic device can perform a model training function.

[0094] like Figure 5 As shown, the training method of the resource representation generation model may include:

[0095] Step 501: Determine a fifth resource that the sample object is interested in from historically recommended resources for the sample object.

[0096] The sample object refers to the object used to train the initial resource representation generation model.

[0097] In one embodiment of the present disclosure, in order to accurately obtain the resources that the sample object is interested in, a possible implementation method for determining the first resource that the sample object is interested in from the historically recommended resources of the sample object is: obtaining the length of time that the sample object stays in the historically recommended resources; and taking the resource whose length of stay in the historically recommended resources exceeds a preset length threshold as the first resource. Thus, by analyzing the length of time that the sample object stays in the historically recommended resources, the first resource that the sample object is interested in can be accurately determined from the historically recommended resources.

[0098] It can be understood that, generally, the longer the sample object stays on a historically recommended resource, the higher the sample object's interest in the historically recommended resource, and vice versa.

[0099] The preset duration threshold is a duration threshold that is preset according to actual needs, and this embodiment does not specifically limit this.

[0100] In some examples, in order to further determine the resources that the sample object is interested in recently, correspondingly, the historically recommended resources of the sample object within the latest preset time period may be obtained, that is, the resources recommended to the sample object within the latest preset time period are obtained.

[0101] The most recent preset duration refers to the preset duration before the current time.

[0102] The length of the most recent preset duration may be set and adjusted by a technician according to actual application scenarios, and is not specifically limited in this embodiment.

[0103] Step 502: Acquire the sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1.

[0104] The number of the sixth resources may be one or more. For example, the number of the sixth resources may be 6. This embodiment does not specifically limit the number of the sixth resources.

[0105] Step 503: Obtain the seventh resource recommended to the sample object for the j+1th time.

[0106] The number of the seventh resources may be one or more, or the number of the seventh resources may be 6. This embodiment does not specifically limit the number of the seventh resources.

[0107] Step 504: determine an eighth resource that the sample object is interested in from the seventh resource.

[0108] Step 505 : using the initial resource representation generation model, determine a second target representation vector of the resource that the sample object is interested in based on the fifth resource, the sixth resource, and the seventh resource.

[0109] Step 506: Train the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector.

[0110] In one embodiment of the present disclosure, a possible implementation method for training the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector is: determine the loss function value according to the loss function of the initial resource representation generation model, the representation vector of the eighth resource and the second target representation vector; and train the initial resource representation generation model according to the loss function value. Thus, by training the initial resource representation generation model in combination with the loss function value, it is possible to intuitively understand how large the prediction error of the model is, and to purposefully adjust the model parameters of the model according to the loss function value, thereby helping the loss function value to converge efficiently, reducing the training time of the model, and improving the training efficiency of the model.

[0111] The loss function in this embodiment is used to calculate the cross entropy loss between the representation vector of the eighth resource and the second target representation vector.

[0112] The training method of the resource representation generation model provided by the embodiment of the present disclosure determines the fifth resource of interest to the sample object from the historically recommended resources of the sample object; obtains the sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1; obtains the seventh resource recommended to the sample object for the j+1th time; determines the eighth resource of interest to the sample object from the seventh resource; uses the initial resource representation generation model to determine the second target representation vector of the resource of interest to the sample object based on the fifth resource, the sixth resource and the seventh resource; and trains the initial resource representation generation model based on the representation vector of the eighth resource and the second target representation vector. Thus, in combination with the fifth resource of interest to the sample object, the sixth resource to be recommended to the target object for the jth time, and the seventh resource and the eighth resource recommended to the target object for the j+1th time, the initial resource representation generation model is trained, so that the trained resource representation generation model can accurately output the representation vector of the resource of interest to the target object, which facilitates the subsequent generation of the representation vector of the resource of interest to the target object based on the trained resource representation generation model.

[0113] Figure 6 is a schematic diagram according to a fifth embodiment of the present disclosure.

[0114] like Figure 6 As shown, the training method of the resource representation generation model may include:

[0115] Step 601: Determine a fifth resource that the sample object is interested in from historically recommended resources for the sample object.

[0116] Step 602: Acquire the sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1.

[0117] Step 603: Obtain the seventh resource recommended to the sample object for the j+1th time.

[0118] Step 604: determine an eighth resource that the sample object is interested in from the seventh resource.

[0119] Step 605: Acquire the second object feature of the sample object.

[0120] Step 606: using the initial resource representation generation model, determine a second target representation vector of the resource of interest to the sample object according to the fifth resource, the second object feature, the sixth resource, and the seventh resource.

[0121] In one embodiment of the present disclosure, when the initial resource representation generation model includes an encoding network and a decoding network, a possible implementation method of determining the second target representation vector of the resource of interest to the sample object based on the fifth resource, the second object feature, the sixth resource and the seventh resource using the initial resource representation generation model is as follows: inputting the representation vector of the fifth resource and the representation vector of the second object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the fifth resource and the representation vector of the second object feature through the encoding network, and obtain the feature vector of the fifth resource and the feature vector of the second object feature; inputting the feature vector of the fifth resource, the feature vector of the second object feature, the representation vector of the sixth resource and the representation vector of the seventh resource into the decoding network, so as to determine the second target representation vector of the resource of interest to the sample object. Thus, performing attention cross-learning on the representation vector of the fifth resource and the representation vector of the second object feature through the encoding network can help the model focus on the most critical part of the object feature and the fifth resource, and inputting the content output by the encoding network into the decoding network, so that the decoding network can more accurately determine the first target representation vector of the resource of interest to the target object in this recommendation based on the representation vector of the second resource, the representation vector of the third resource and the output content of the encoding network.

[0122] In this embodiment, cross-learning of the representation vector of the fifth resource and the representation vector of the second object feature is performed through the encoding network, which can help the model focus on the most critical parts of the second object feature and the fifth resource, ignore unimportant information, and reduce the impact of unimportant information on the model. This helps to improve the training speed of the model while improving the accuracy and robustness of the model.

[0123] Step 607: train the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector.

[0124] In this embodiment, the initial resource representation generation model is trained by combining the second object feature of the sample object, the fifth resource of interest to the sample object, the sixth resource to be recommended to the target object for the jth time, and the seventh and eighth resources recommended to the target object for the j+1th time, so that the trained resource representation generation model can more accurately output the representation vector of the resource of interest to the target object.

[0125] In order to implement the above embodiments, the present disclosure also provides a resource recommendation device based on a large model.

[0126] Figure 7 is a schematic diagram according to a sixth embodiment of the present disclosure.

[0127] like Figure 7As shown, the resource recommendation device 70 based on the large model may include: a first determination module 701, a processing module 702, a second determination module 703, a third determination module 704 and a recommendation module 705, wherein:

[0128] The first determining module 701 is used to determine a first resource that the target object is interested in from historically recommended resources for the target object.

[0129] The processing module 702 is used to determine the second resource to be recommended to the target object this time, and obtain the third resource recommended to the target object last time.

[0130] The second determination module 703 is used to determine a first target representation vector of a resource that the target object is interested in based on the first resource, the second resource and the third resource by using the resource representation generation model.

[0131] The third determination module 704 is used to determine a fourth resource matching the first target representation vector from the candidate resource set.

[0132] The recommendation module 705 is used to perform the current information flow recommendation process according to the fourth resource.

[0133] As a possible implementation of the embodiment of the present disclosure, the device may further include:

[0134] An acquisition module, used for acquiring a first object feature of a target object;

[0135] The second determination module 703 is specifically used to: determine a first target representation vector of a resource of interest to the target object according to the first resource, the first object feature, the second resource and the third resource by using the resource representation generation model.

[0136] As a possible implementation method of an embodiment of the present disclosure, the resource representation generation model includes an encoding network and a decoding network, and the second determination module 703 is specifically used to: input the representation vector of the first resource and the representation vector of the first object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the first resource and the representation vector of the first object feature through the encoding network to obtain the feature vector of the first resource and the feature vector of the first object feature; input the feature vector of the first resource, the feature vector of the first object feature, the representation vector of the second resource and the representation vector of the third resource into the decoding network to determine the first target representation vector of the resource of interest to the target object.

[0137] As a possible implementation method of the embodiment of the present disclosure, the method for obtaining the candidate resource set is: according to the resource preference of the target object, resources are recalled from a preset resource pool; the recalled resources are roughly sorted, and the resources obtained by the rough sorting are finely sorted to obtain a candidate resource set.

[0138] As a possible implementation method of the embodiment of the present disclosure, determining the second resource to be recommended to the target object this time includes: performing fusion sorting processing on the candidate resource set to obtain a fusion sorting result; and taking the top K resources in the fusion sorting result as the second resource, where K is an integer greater than or equal to 1.

[0139] As a possible implementation of the embodiment of the present disclosure, the first determination module 701 is specifically used to: obtain the target object's stay time in the historically recommended resources; and use the resource whose stay time in the historically recommended resources exceeds a preset time threshold as the first resource.

[0140] It should be noted that the aforementioned explanation of the resource recommendation method based on a large model is also applicable to the resource recommendation device based on a large model in this embodiment, and will not be repeated here.

[0141] The resource recommendation device based on a large model of the disclosed embodiment determines the first resource that the target object is interested in from the historically recommended resources of the target object; determines the second resource to be recommended to the target object this time, and obtains the third resource recommended to the target object last time; uses the resource representation generation model to determine the first target representation vector of the resource that the target object is interested in based on the first resource, the second resource, and the third resource; determines the fourth resource that matches the first target representation vector from the candidate resource set; and performs information flow recommendation processing this time based on the fourth resource. Thus, through the resource representation generation model, the first resource that the target object is interested in, the second resource to be recommended to the target object this time, and the third resource recommended to the target object last time are analyzed, and the representation vector of the resource that the target object is interested in during this recommendation can be determined, and the fourth resource is determined from the candidate resource set based on the representation vector, and information flow recommendation processing is performed based on the fourth resource, which can make the recommended resources meet the resource needs of the target object and improve the accuracy of resource recommendation.

[0142] In order to implement the above embodiments, the present disclosure also provides a training device for a resource representation generation model.

[0143] Figure 8 is a schematic diagram according to a seventh embodiment of the present disclosure.

[0144] like Figure 8 As shown, the training device 80 for the resource representation generation model may include: a first determination module 801, a first acquisition module 802, a second acquisition module 803, a second determination module 804, a third determination module 805 and a training module 806, wherein:

[0145] The first determining module 801 is configured to determine a fifth resource that the sample object is interested in from historically recommended resources for the sample object.

[0146] The first acquisition module 802 is configured to acquire a sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1.

[0147] The second acquisition module 803 is used to acquire the seventh resource recommended to the sample object for the j+1th time.

[0148] The second determining module 804 is configured to determine an eighth resource that the sample object is interested in from the seventh resource.

[0149] The third determination module 805 is used to determine a second target representation vector of the resource that the sample object is interested in according to the fifth resource, the sixth resource and the seventh resource by using the initial resource representation generation model.

[0150] The training module 806 is used to train the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector.

[0151] As a possible implementation method of the embodiment of the present disclosure, the training module 806 is specifically used to: determine the loss function value according to the loss function of the initial resource representation generation model, the representation vector of the eighth resource and the second target representation vector; and train the initial resource representation generation model according to the loss function value.

[0152] As a possible implementation of the embodiment of the present disclosure, the device may further include:

[0153] A third acquisition module, used to acquire a second object feature of the sample object;

[0154] The training module 806 is specifically used to: use the initial resource representation generation model to determine the second target representation vector of the resource that the sample object is interested in according to the fifth resource, the second object feature, the sixth resource and the seventh resource.

[0155] As a possible implementation method of an embodiment of the present disclosure, the initial resource representation generation model includes an encoding network and a decoding network, and a training module 806 is specifically used to: input the representation vector of the fifth resource and the representation vector of the second object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the fifth resource and the representation vector of the second object feature through the encoding network to obtain a feature vector of the fifth resource and a feature vector of the second object feature; input the feature vector of the fifth resource, the feature vector of the second object feature, the representation vector of the sixth resource and the representation vector of the seventh resource into the decoding network to determine the second target representation vector of the resource of interest to the sample object.

[0156] It should be noted that the aforementioned explanation of the training method of the resource representation generation model is also applicable to the training device of the resource representation generation model of this embodiment, and will not be repeated here.

[0157] The training device of the resource representation generation model provided by the embodiment of the present disclosure determines the fifth resource that the sample object is interested in from the historically recommended resources of the sample object; obtains the sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1; obtains the seventh resource recommended to the sample object for the j+1th time; determines the eighth resource that the sample object is interested in from the seventh resource; uses the initial resource representation generation model to determine the second target representation vector of the resource that the sample object is interested in based on the fifth resource, the sixth resource and the seventh resource; and trains the initial resource representation generation model based on the representation vector of the eighth resource and the second target representation vector. Thus, in combination with the fifth resource that the sample object is interested in, the sixth resource to be recommended to the target object for the jth time, and the seventh resource and the eighth resource recommended to the target object for the j+1th time, the initial resource representation generation model is trained, so that the trained resource representation generation model can accurately output the representation vector of the resource that the target object is interested in.

[0158] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out with the user's consent, comply with the relevant laws and regulations, and do not violate public order and good morals.

[0159] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0160] Fig. 9 900 is a schematic block diagram of an electronic device 900 for implementing 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 processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0161] like Fig. 9As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0162] A number of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0163] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as a resource recommendation method based on a large model, or a training method for a resource representation generation model. For example, in some embodiments, a resource recommendation method based on a large model, or a training method for a resource representation generation model may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the resource recommendation method based on the large model or the training method for the resource representation generation model described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute a resource recommendation method based on a large model, or a training method for a resource representation generation model, in any other appropriate manner (for example, by means of firmware).

[0164] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0165] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0166] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0167] 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).

[0168] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0169] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0170] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0171] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A resource recommendation method based on a large model, comprising: Determining a first resource that the target object is interested in from historically recommended resources for the target object; Determine the second resource to be recommended to the target object this time, and obtain the third resource recommended to the target object last time; Determine, by using a resource representation generation model, a first target representation vector of a resource of interest to the target object based on the first resource, the second resource, and the third resource; Determining, from the candidate resource set, a fourth resource that matches the first target representation vector; The current information flow recommendation process is performed according to the fourth resource.

2. The method according to claim 1, wherein: The method further comprises: Acquire a first object feature of the target object; The using of the resource representation generation model to determine a first target representation vector of the resource of interest to the target object according to the first resource, the second resource and the third resource includes: The resource representation generation model is used to determine a first target representation vector of a resource of interest to the target object according to the first resource, the first object feature, the second resource, and the third resource.

3. The method according to claim 2, wherein: The resource representation generation model includes an encoding network and a decoding network, and using the resource representation generation model to determine a feature vector of a resource of interest to the target object according to the first resource, the first object feature, the second resource, and the third resource, includes: Inputting the representation vector of the first resource and the representation vector of the first object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the first resource and the representation vector of the first object feature through the encoding network, thereby obtaining a feature vector of the first resource and a feature vector of the first object feature; The feature vector of the first resource, the feature vector of the first object feature, the representation vector of the second resource and the representation vector of the third resource are input into the decoding network to determine a first target representation vector of the resource of interest to the target object.

4. The method according to claim 1, wherein: The method for obtaining the candidate resource set is as follows: Recalling resources from a preset resource pool according to the resource preference of the target object; The recalled resources are roughly sorted, and the resources obtained by the rough sorting are finely sorted to obtain the candidate resource set.

5. The method according to claim 4, wherein: The determining the second resource to be recommended to the target object this time includes: Performing fusion sorting processing on the candidate resource set to obtain a fusion sorting result; The top K resources in the fusion sorting result are used as the second resources, where K is an integer greater than or equal to 1.

6. The method according to any one of claims 1 to 5, wherein: The step of determining the first resource that the target object is interested in from the historically recommended resources of the target object includes: Obtaining the target object's stay time in historically recommended resources; The resource whose stay time exceeds a preset time threshold among the historically recommended resources is used as the first resource.

7. A method for training a resource representation generation model, comprising: Determining a fifth resource that the sample object is interested in from historically recommended resources of the sample object; Obtaining a sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1; Obtaining the seventh resource recommended to the sample object for the j+1th time; Determining an eighth resource of interest to the sample object from the seventh resource; Determine, by using the initial resource representation generation model, a second target representation vector of the resource of interest to the sample object according to the fifth resource, the sixth resource, and the seventh resource; The initial resource representation generation model is trained according to the representation vector of the eighth resource and the second target representation vector.

8. The method according to claim 7, wherein: The training of the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector includes: Determine a loss function value according to the loss function of the initial resource representation generation model, the representation vector of the eighth resource, and the second target representation vector; The initial resource representation generation model is trained according to the loss function value.

9. The method according to claim 7, wherein: The method further comprises: Acquire a second object feature of the sample object; The using of the initial resource representation generation model to determine the second target representation vector of the resource of interest to the target object according to the fifth resource, the sixth resource and the seventh resource includes: The initial resource representation generation model is used to determine a second target representation vector of the resource of interest to the sample object according to the fifth resource, the second object feature, the sixth resource, and the seventh resource.

10. The method according to claim 9, wherein: The initial resource representation generation model includes an encoding network and a decoding network, and the determining, using the initial resource representation generation model and according to the fifth resource, the second object feature, the sixth resource, and the seventh resource, a second target representation vector of the resource of interest to the sample object includes: Inputting the representation vector of the fifth resource and the representation vector of the second object feature into the encoding network, so as to perform attention cross-learning on the representation vector of the fifth resource and the representation vector of the second object feature through the encoding network, thereby obtaining a feature vector of the fifth resource and a feature vector of the second object feature; The feature vector of the fifth resource, the feature vector of the second object feature, the representation vector of the sixth resource and the representation vector of the seventh resource are input into the decoding network to determine a second target representation vector of the resource of interest to the sample object.

11. A resource recommendation device based on a large model, comprising: A first determining module, configured to determine a first resource that the target object is interested in from historically recommended resources for the target object; A processing module, used to determine the second resource to be recommended to the target object this time, and obtain the third resource recommended to the target object last time; A second determination module is used to determine a first target representation vector of a resource of interest to the target object based on the first resource, the second resource and the third resource by using a resource representation generation model; A third determination module, configured to determine a fourth resource matching the first target representation vector from the candidate resource set; The recommendation module is used to perform the current information flow recommendation process according to the fourth resource.

12. A training device for a resource representation generation model, comprising: A first determining module, configured to determine a fifth resource that the sample object is interested in from historically recommended resources of the sample object; A first acquisition module, configured to acquire a sixth resource pushed to the sample object for the jth time, where j is an integer greater than or equal to 1; A second acquisition module, configured to acquire a seventh resource recommended to the sample object for the j+1th time; A second determining module, configured to determine an eighth resource of interest to the sample object from the seventh resource; A third determination module is used to determine a second target representation vector of the resource of interest to the sample object based on the fifth resource, the sixth resource and the seventh resource by using the initial resource representation generation model; A training module is used to train the initial resource representation generation model according to the representation vector of the eighth resource and the second target representation vector.

13. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 6, or the method of any one of claims 7 to 10.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 6, or the method according to any one of claims 7 to 10.

15. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 6, or the method according to any one of claims 7 to 10.

Images