Resource recommendation method and device and electronic equipment

Abstract

The invention provides a resource recommendation method and device and electronic equipment, and relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, natural language processing, computer vision, large models and the like. According to the specific implementation scheme, the method comprises the steps of determining a first satisfaction degree of candidate resources according to candidate resource features of the candidate resources, target object features of a target recommendation object and a sorting model; determining a second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, historical resource features of historical satisfaction resources of the target recommendation object and a large language model; according to the first satisfaction degree and the second satisfaction degree, selecting a to-be-recommended resource from the candidate resources for resource recommendation processing; wherein the second satisfaction is determined in combination with the large language model and the historical resource features, and resource recommendation processing is performed in combination with the first satisfaction and the second satisfaction, so that personalized resource recommendation services can be provided, and the resource recommendation efficiency is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to the fields of deep learning, natural language processing, computer vision, large models, etc. In particular, it relates to a resource recommendation method, apparatus and electronic device. Background Art

[0002] The current resource recommendation method mainly combines candidate resources, a recommended object, and a ranking model to select resources to be recommended from the candidate resources for recommendation processing. Among them, the ranking model combines the resource features of the candidate resources and the object features of the recommended object for resource recommendation processing, considering fewer features, resulting in poor resource recommendation efficiency. Summary of the Invention

[0003] The present disclosure provides a resource recommendation method, apparatus and electronic device.

[0004] According to one aspect of the present disclosure, there is provided a resource recommendation method, the method comprising: obtaining candidate resource features of candidate resources, target object features of a target recommended object, and historical resource features of historical satisfactory resources of the target recommended object; determining a first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and a ranking model; determining a second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and a large language model; and selecting resources to be recommended from the candidate resources according to the first satisfaction degree and the second satisfaction degree of the candidate resources for resource recommendation processing of the target recommended object.

[0005] According to another aspect of the present disclosure, there is provided a training method for a large language model for resource recommendation, the method comprising: obtaining a large language model to be trained, and training data of the large language model; the training data includes: sample resource features of sample resources, sample object features of sample recommended objects, sample historical resource features of sample historical satisfactory resources of the sample recommended objects, and sample satisfaction degrees of the sample resources; and using the sample resource features, the sample object features, the sample historical resource features, and the sample satisfaction degrees to perform training processing on the large language model to obtain a trained large language model for resource recommendation processing.

[0006] According to another aspect of the present disclosure, there is provided a resource recommendation device, the device comprising: an acquisition module configured to acquire candidate resource features of candidate resources, target object features of a target recommendation object, and historical resource features of historical satisfied resources of the target recommendation object; a first determination module configured to determine a first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and a ranking model; a second determination module configured to determine a second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and a large language model; and a selection module configured to select a to-be-recommended resource from the candidate resources according to the first satisfaction degree and the second satisfaction degree of the candidate resources for performing a resource recommendation process on the target recommendation object.

[0007] According to another aspect of the present disclosure, there is provided a training device for a large language model for resource recommendation, the device comprising: an acquisition module configured to acquire a large language model to be trained and training data of the large language model; the training data including: sample resource features of sample resources, sample object features of sample recommendation objects, sample historical resource features of sample historical satisfied resources of the sample recommendation objects, and sample satisfaction degrees of the sample resources; and a training module configured to perform a training process on the large language model by using the sample resource features, the sample object features, the sample historical resource features, and the sample satisfaction degrees to obtain a trained large language model for resource recommendation processing.

[0008] According to another aspect of the present disclosure, there is provided an electronic device, 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 when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the resource recommendation method proposed above in the present disclosure; or execute the training method for a large language model for resource recommendation proposed above in the present disclosure.

[0009] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to execute the resource recommendation method proposed above in the present disclosure; or execute the training method for a large language model for resource recommendation proposed above in the present disclosure.

[0010] According to another aspect of the present disclosure, there is provided a computer program product, comprising a computer program which, when executed by a processor, implements the steps of the resource recommendation method proposed above in the present disclosure; or implements the steps of the training method for a large language model for resource recommendation proposed above in the present disclosure.

[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 used to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. Description of the Drawings

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

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

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

[0015] Figure 3 is a schematic diagram according to the third embodiment of the present disclosure;

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

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

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

[0019] Figure 7 is a block diagram of an electronic device for implementing the resource recommendation method of the embodiments of the present disclosure or for training a large language model for resource recommendation. Detailed Embodiments

[0020] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to assist understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can 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.

[0021] The current resource recommendation method mainly combines candidate resources, recommendation objects, and a ranking model to select resources to be recommended from the candidate resources for recommendation processing. Among them, in the training data of the ranking model, the partial order relationship between at least two sample resources is determined only by combining the parameter statistics of the sample resources, such as physical duration, number of images, consumption duration, etc., resulting in poor accuracy of the training data and thus poor resource recommendation efficiency.

[0022] In view of the above problems, the present disclosure proposes a resource recommendation method, apparatus, and electronic device.

[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 embodiments of the present disclosure can be applied to a resource recommendation device, and the device can be configured in an electronic device so that the electronic device can perform the resource recommendation function.

[0024] Among them, the electronic device can be any device with computing power, such as a personal computer (PC for short), a mobile terminal, a server, etc. The mobile terminal can be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a smart speaker, a server, a server cluster, etc., which are hardware devices with various operating systems, touch screens, and / or display screens.

[0025] Among them, the resource recommendation device can also be software in the electronic device, such as resource recommendation software, etc. In the following embodiments, the execution subject is taken as an electronic device for illustration.

[0026] As Figure 1 shown, the resource recommendation method may include the following steps:

[0027] Step 101, obtain the candidate resource features of the candidate resources, the target object features of the target recommendation object, and the historical resource features of the historical satisfactory resources of the target recommendation object.

[0028] In the embodiments of the present disclosure, the candidate resources can be the resources in the candidate resource list; the candidate resource list is a resource list obtained by the target recommendation object through resource query processing based on the target query statement; or, it is a resource list triggered after the target recommendation object selects a resource entry.

[0029] Among them, the resources in the resource list can be resources related to the target query statement or resource entry of the target recommendation object, and are the resources that the target recommendation object wants to browse. Performing resource recommendation processing based on the resources in the resource list enables the above resource recommendation method to be applicable to resource query scenarios or resource entry trigger scenarios, thereby expanding the applicable scenarios of the resource recommendation method.

[0030] In the embodiments of the present disclosure, the candidate resources can be resources of at least one of the following types: video resources, dynamic resources, and graphic and text resources. The candidate resource features include the text-related content of the candidate resources; the text-related content includes at least one of the following: resource title, resource introduction, and content obtained after text recognition of videos and / or images in the resources.

[0031] Among them, dynamic resources such as slides, animations, etc. Among them, in addition to the text-related content of the candidate resources, the candidate resource features may also include the attribute features of the candidate resources. Taking video resources as an example, the attribute features may include at least one of the following: physical duration, consumption duration, etc. Taking dynamic resources as an example, the attribute features may include at least one of the following: total number of images, number of consumed images, etc. Taking graphic and text resources as an example, the attribute features may include at least one of the following: number of characters in the resource, number of consumed characters, etc.

[0032] Among them, the setting of multiple types of candidate resources enables the resource recommendation method to perform resource recommendation processing based on multiple types of candidate resources, thereby further expanding the applicable scenarios of the resource recommendation method.

[0033] Among them, the setting of the text-related content enables the sorting model and / or the large language model to perform satisfaction determination processing in combination with the text-related content, and can consider the features of the content dimension in the candidate resources, thereby further improving the accuracy of the determined satisfaction.

[0034] In the embodiments of the present disclosure, the target object features may include the basic attributes, behavioral attributes, preference attributes, etc. of the target object. Among them, the basic attributes such as age, gender, etc. The behavioral attributes such as activity level, etc. The preference attributes such as the preferred resource type, etc.

[0035] Among them, the historical satisfied resources of the target recommended object can be resources that the target recommended object has browsed and meet the conditions. Among them, the conditions are, for example, that the satisfaction of the target recommended object with respect to the resource is greater than or equal to the satisfaction threshold; the viewing completion rate of the target recommended object with respect to the resource is greater than or equal to the completion rate threshold.

[0036] Step 102, determine the first satisfaction of the candidate resource according to the candidate resource features, the target object features, and the sorting model.

[0037] In the embodiments of the present disclosure, the electronic device may input the candidate resource features and the target object features into the sorting model, obtain the satisfaction value output by the sorting model; and determine this satisfaction value as the first satisfaction of the candidate resource.

[0038] Among them, the sorting model can be, for example, a Pairwise Comparison Model, etc. The training data of the sorting model can include: the sample resource features of the sample resources, the sample object features of the sample recommended objects, and the sample satisfaction of the sample resources. Among them, the sample satisfaction of the sample resources is determined by combining the basic attributes and consumption attributes of the sample resources. Among them, for video resources, the basic attribute can be, for example, the physical duration; the consumption attribute can be, for example, the consumption duration. For dynamic resources, the basic attribute can be, for example, the total number of images; the consumption attribute can be, for example, the number of consumed images. For graphic-text resources, the basic attribute can be, for example, the number of words in the resource; the consumption attribute can be, for example, the number of consumed words.

[0039] Among them, the process of determining the sample satisfaction of the sample resources can be, for example, to determine the bucket where the sample resources are located according to the basic attributes of the sample resources; based on the fitting function corresponding to the bucket and the consumption attribute, determine the sample satisfaction of the sample resources. Among them, the number of buckets can be multiple, obtained after range partitioning based on the basic attributes. The process of determining the fitting function corresponding to the bucket can be, for example, to sort the multiple resources in the bucket according to the consumption attribute to obtain a sorting result; determine the consumption attributes at multiple specified quantiles in the sorting result; perform function fitting processing according to the multiple specified quantiles and the consumption attributes at the multiple specified quantiles to obtain a fitting function; the fitting function is a mapping relationship between the consumption attribute and the quantile. Among them, the quantile can be used to represent the satisfaction.

[0040] Step 103, determine the second satisfaction of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model.

[0041] In the embodiments of the present disclosure, the electronic device can input the candidate resource features, the target object features, and the historical resource features into the large language model to obtain the satisfaction value output by the large language model; determine the satisfaction value as the second satisfaction of the candidate resources.

[0042] Among them, the large language model (Large Language Model, LLM) has powerful text representation capabilities and can perform satisfaction determination processing based on the extracted text features, thereby improving the accuracy of the second satisfaction obtained by de - multifunction.

[0043] In the embodiments of the present disclosure, the large language model is trained by combining the sample resource features of at least two sample resources, the sample object features of the sample objects, the sample historical resource features of the sample historical satisfied resources of the sample objects, and the sample satisfaction partial order relationship between at least two sample resources; among them, the sample satisfaction partial order relationship between at least two sample resources is determined by combining the satisfaction of the sample object for at least two sample resources.

[0044] Among them, the sample satisfaction partial order relationship between at least two sample resources can indicate the size relationship of satisfaction degrees between at least two sample resources. Suppose that among at least two sample resources, there are sample resource A and sample resource B. The sample satisfaction partial order relationship can indicate that the satisfaction degree of sample resource A is greater than that of sample resource B, or indicate that the satisfaction degree of sample resource B is greater than that of sample resource A.

[0045] Among them, during the training process of the large language model, considering the satisfaction degrees of the sample object for at least two sample resources and the historical satisfied resources of the sample object can improve the accuracy of the trained large language model, and further improve the accuracy of the determined second satisfaction degree.

[0046] Step 104: Select a resource to be recommended from the candidate resources according to the first satisfaction degree and the second satisfaction degree of the candidate resources for resource recommendation processing of the target recommendation object.

[0047] The resource recommendation method of the embodiments of the present disclosure obtains the candidate resource features of the candidate resources, the target object features of the target recommendation object, and the historical resource features of the historical satisfied resources of the target recommendation object; determines the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the ranking model; determines the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model; selects a resource to be recommended from the candidate resources according to the first satisfaction degree and the second satisfaction degree of the candidate resources for resource recommendation processing of the target recommendation object; among them, the second satisfaction degree is determined by combining the large language model and the historical resource features of the historical satisfied resources of the target recommendation object. Combining the first satisfaction degree and the second satisfaction degree for resource recommendation processing can provide a personalized resource recommendation service and improve the resource recommendation efficiency.

[0048] Among them, in order to further improve the selection accuracy of the resource to be recommended, for any resource pair composed of two candidate resources among the candidate resources, the electronic device can first combine the first satisfaction degree and the second satisfaction degree of the two candidate resources in the resource pair to determine the target satisfaction partial order relationship between the two candidate resources, and then combine the target satisfaction degree for the selection processing of the resource to be recommended. As Figure 2 shown, Figure 2 is a schematic diagram according to the second embodiment of the present disclosure, Figure 2 The embodiment shown may include the following steps:

[0049] Step 201: Obtain the candidate resource features of the candidate resources, the target object features of the target recommendation object, and the historical resource features of the historical satisfied resources of the target recommendation object.

[0050] Step 202: Determine the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the ranking model.

[0051] Step 203: Determine the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model.

[0052] Step 204: For any resource pair composed of two candidate resources among the candidate resources, determine the target satisfaction partial order relationship between the two candidate resources according to the first satisfaction degree and the second satisfaction degree of the two candidate resources in the resource pair.

[0053] In the embodiment of the present disclosure, the process of the electronic device executing Step 204 may be, for example, determining the first weight corresponding to the ranking model and the second weight corresponding to the large language model; performing weighted processing on the first satisfaction degree and the second satisfaction degree of the candidate resources in the resource pair according to the first weight and the second weight to obtain the target satisfaction degree of the candidate resources in the resource pair; determining the target satisfaction partial order relationship between the two candidate resources according to the target satisfaction degrees of the two candidate resources in the resource pair.

[0054] Among them, the first weight corresponding to the ranking model and the second weight corresponding to the large language model may be preset, or determined in combination with the accuracy of the ranking model and the large language model.

[0055] Among them, determining the target satisfaction partial order relationship by combining the first satisfaction degree and the second satisfaction degree of the candidate resources, as well as the first weight corresponding to the ranking model and the second weight corresponding to the large language model, can comprehensively consider the emphasis degrees of the ranking model and the large language model in the resource recommendation process, and further improve the accuracy of the to-be-recommended resources determined, and further improve the resource recommendation efficiency.

[0056] In the embodiment of the present disclosure, the process of the electronic device determining the resource pair may be, for example, for each candidate resource, the electronic device may perform pairing processing on the candidate resource with each of the other candidate resources respectively to obtain a plurality of resource pairs; then, perform duplicate removal processing on the resource pairs.

[0057] Step 205: Select the to-be-recommended resources from the candidate resources according to the target satisfaction partial order relationship between the two candidate resources in each resource pair.

[0058] In the embodiment of the present disclosure, the target satisfaction partial order relationship between the two candidate resources in the resource pair may indicate the satisfaction degree size relationship between the two candidate resources in the resource pair. The process of the electronic device executing Step 205 may be, for example, determining the first candidate resource according to the target satisfaction partial order relationship between the two candidate resources in each resource pair; the satisfaction degree of the first candidate resource is greater than the satisfaction degrees of a preset number of non-first candidate resources; select the to-be-recommended resources from the first candidate resources.

[0059] Among them, the satisfaction degree of the first candidate resource is greater than that of a preset number of non-first candidate resources, which indicates that the first candidate resource is the resource that the target recommended object is more inclined to. Selecting the resource to be recommended from the first candidate resources for recommendation processing can further improve the accuracy of the determined resource to be recommended and further improve the resource recommendation efficiency.

[0060] Among them, as an alternative, the process of the electronic device executing step 205 can be, for example, according to the satisfaction degree size relationship between two candidate resources in each resource pair, performing a satisfaction degree size sorting process on each candidate resource to obtain a sorting result; in the case where the sorting result is in descending order of satisfaction degree, determining multiple candidate resources at the front as the resources to be recommended; in the case where the sorting result is in ascending order of satisfaction degree, determining multiple candidate resources at the back as the resources to be recommended.

[0061] Among them, it should be noted that for the detailed content of steps 201 to 203, reference can be made to Figure 1 Steps 101 to 103 in the illustrated embodiment, and details will not be described here again.

[0062] The resource recommendation method of the present disclosure embodiment includes obtaining the candidate resource features of candidate resources, the target object features of the target recommended object, and the historical resource features of the historical satisfied resources of the target recommended object; determining the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the sorting model; determining the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model; for any two candidate resources in the candidate resources that form a resource pair, determining the target satisfaction partial order relationship between the two candidate resources according to the first satisfaction degree and the second satisfaction degree of the two candidate resources in the resource pair; selecting the resources to be recommended from the candidate resources according to the target satisfaction partial order relationship between the two candidate resources in each resource pair; among them, when determining the target satisfaction partial order relationship, the satisfaction degrees output by the sorting model and the large language model can be comprehensively considered, so as to improve the accuracy of the determined target satisfaction partial order relationship, and further improve the resource recommendation efficiency.

[0063] Among them, in order to further improve the accuracy of the determined target satisfaction partial order relationship and further improve the resource recommendation efficiency, considering that the sorting model mainly learns the partial order relationship and the accuracy of the output satisfaction partial order relationship is relatively high, the target satisfaction partial order relationship can be determined by combining the first satisfaction partial order relationship based on the first satisfaction degree and the second satisfaction partial order relationship based on the second satisfaction degree. As Figure 3 shown, Figure 3 is a schematic diagram according to the third embodiment of the present disclosure, Figure 3 The illustrated embodiment may include the following steps:

[0064] Step 301: Obtain the candidate resource features of the candidate resources, the target object features of the target recommended object, and the historical resource features of the historical satisfied resources of the target recommended object.

[0065] Step 302: Determine the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the sorting model.

[0066] Step 303: Determine the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model.

[0067] Step 304: Determine the first satisfaction partial order relationship between two candidate resources according to the first satisfaction degrees of the two candidate resources in the resource pair.

[0068] Among them, the first satisfaction partial order relationship can indicate a size relationship of satisfaction degrees between two candidate resources. Suppose the two candidate resources are candidate resource A and candidate resource B respectively. The first satisfaction partial order relationship can indicate that the satisfaction degree of candidate resource A is greater than that of candidate resource B, or indicate that the satisfaction degree of candidate resource B is greater than that of candidate resource A.

[0069] Among them, the satisfaction degree of candidate resource A is greater than that of candidate resource B, that is, among candidate resource A and candidate resource B, the target recommended object is more inclined to candidate resource A. The satisfaction degree of candidate resource B is greater than that of candidate resource A, that is, among candidate resource A and candidate resource B, the target recommended object is more inclined to candidate resource B.

[0070] Step 305: Determine the second satisfaction partial order relationship between two candidate resources according to the second satisfaction degrees of the two candidate resources in the resource pair.

[0071] Among them, the second satisfaction partial order relationship can indicate another size relationship of satisfaction degrees between two candidate resources.

[0072] Step 306: Determine the target satisfaction partial order relationship between two candidate resources according to the first satisfaction partial order relationship and the second satisfaction partial order relationship.

[0073] In an embodiment of the present disclosure, the process of the electronic device executing step 306 may be, for example, in the case where the first satisfaction partial order relation is inconsistent with the second satisfaction partial order relation, determining the satisfaction deviation value between two candidate resources in the resource pair according to the second satisfaction partial order relation; adjusting the first satisfaction degrees of the two candidate resources according to the satisfaction deviation values of the two candidate resources to obtain the adjusted first satisfaction degrees; determining the target satisfaction partial order relation between the two candidate resources according to the adjusted first satisfaction degrees of the two candidate resources; in the case where the first satisfaction partial order relation is consistent with the second satisfaction partial order relation, determining the first satisfaction partial order relation or the second satisfaction partial order relation as the target satisfaction partial order relation.

[0074] Among them, in the case where the first satisfaction partial order relation is inconsistent with the second satisfaction partial order relation, the magnitude relationship between the satisfaction deviation values of the two candidate resources in the resource pair may be the same as the magnitude relationship between the two second satisfaction degrees in the second satisfaction partial order relation. That is, assuming that the two candidate resources are candidate resource A and candidate resource B respectively, in the case where the second satisfaction degree of candidate resource A is greater than the second satisfaction degree of candidate resource B, the satisfaction deviation value of candidate resource A may be greater than the satisfaction deviation value of candidate resource B; in the case where the second satisfaction degree of candidate resource A is less than the second satisfaction degree of candidate resource B, the satisfaction deviation value of candidate resource B may be greater than the satisfaction deviation value of candidate resource A.

[0075] Among them, the satisfaction deviation value may be preset, or mapped in combination with the second satisfaction degree of the candidate resource.

[0076] Among them, according to whether the first satisfaction partial order relation is consistent with the second satisfaction partial order relation, determining whether to determine the target satisfaction partial order relation by combining the satisfaction deviation value and the first satisfaction partial order relation, or directly using the first satisfaction partial order relation or the second satisfaction partial order relation as the target satisfaction partial order relation, can consider the influence of the second satisfaction partial order relation on the first satisfaction partial order relation, thereby further improving the accuracy of the determined target satisfaction partial order relation.

[0077] Step 307, selecting the resource to be recommended from the candidate resources according to the target satisfaction partial order relation between the two candidate resources in each resource pair.

[0078] Among them, it should be noted that for the detailed content of steps 301 to 303, reference may be made to Figure 1 Steps 101 to 103 in the illustrated embodiment, and details will not be described herein again.

[0079] The resource recommendation method according to the embodiments of the present disclosure includes obtaining the candidate resource features of candidate resources, the target object features of the target recommendation object, and the historical resource features of the historical satisfactory resources of the target recommendation object; determining the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the ranking model; determining the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model; determining the first satisfaction partial order relationship between two candidate resources in a resource pair according to the first satisfaction degrees of the two candidate resources in the resource pair; determining the second satisfaction partial order relationship between two candidate resources in a resource pair according to the second satisfaction degrees of the two candidate resources in the resource pair; determining the target satisfaction partial order relationship between two candidate resources according to the first satisfaction partial order relationship and the second satisfaction partial order relationship; and selecting the resources to be recommended from the candidate resources according to the target satisfaction partial order relationships between two candidate resources in each resource pair. Among them, considering that the ranking model mainly learns the partial order relationship and the accuracy of the output satisfaction partial order relationship is relatively high, determining the target satisfaction partial order relationship based on the first satisfaction partial order relationship based on the first satisfaction degree and the second satisfaction partial order relationship based on the second satisfaction degree can further improve the accuracy of the determined target satisfaction partial order relationship, and further improve the resource recommendation efficiency.

[0080] Figure 4 FIG. is a schematic diagram according to the fourth embodiment of the present disclosure. It should be noted that the training method of the large language model for resource recommendation according to the embodiments of the present disclosure can be applied to the training device of the large language model for resource recommendation, and this device can be configured in an electronic device so that the electronic device can execute the training function of the large language model for resource recommendation.

[0081] Among them, the electronic device can be any device with computing power, such as a personal computer (PC for short), a mobile terminal, a server, etc. The mobile terminal can be, for example, a vehicle-mounted device, a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a smart speaker, a server, a server cluster, etc., which are hardware devices with various operating systems, touch screens, and / or display screens.

[0082] Among them, the training device of the large language model for resource recommendation can also be software in the electronic device, such as the training software of the large language model for resource recommendation, etc. In the following embodiments, the execution subject is taken as an example of an electronic device for description.

[0083] As Figure 4 shown, the training method of the large language model for resource recommendation may include the following steps:

[0084] Step 401: Obtain the large language model to be trained and the training data of the large language model; the training data includes: the sample resource features of the sample resources, the sample object features of the sample recommended objects, the sample historical resource features of the sample historical satisfied resources of the sample recommended objects, and the sample satisfaction of the sample resources.

[0085] In the embodiments of the present disclosure, the sample resources may be resources of at least one of the following types: video resources, dynamic resources, and graphic and text resources. The sample resource features include the text-related content of the sample resources; the text-related content includes at least one of the following: resource title, resource introduction, and the content obtained after text recognition of videos and / or images in the resources.

[0086] Among them, in addition to the text-related content of the sample resources, the sample resource features may further include the attribute features of the sample resources. Taking video resources as an example, the attribute features may include at least one of the following: physical duration, consumption duration, etc. Taking dynamic resources as an example, the attribute features may include at least one of the following: total number of images, number of consumed images, etc. Taking graphic and text resources as an example, the attribute features may include at least one of the following: number of words in the resources, number of consumed words, etc.

[0087] Among them, the setting of multiple types of sample resources enables the large language model to be applicable to resource recommendation processing for multiple types of resources, thereby further expanding the applicable scenarios of the large language model.

[0088] Step 402: Use the sample resource features, sample object features, sample historical resource features, and sample satisfaction to perform training processing on the large language model to obtain a trained large language model for resource recommendation processing.

[0089] In the embodiments of the present disclosure, in one example, the process of the electronic device executing Step 402 may be, for example, inputting the sample resource features, sample object features, and sample historical resource features into the large language model to obtain the predicted satisfaction output by the large language model; according to the sample satisfaction, predicted satisfaction, and the first loss function of the large language model, perform parameter adjustment processing on the large language model to obtain a trained large language model.

[0090] Specifically, the electronic device may determine the value of the first loss function according to the sample satisfaction, predicted satisfaction, and the first loss function of the large language model; combine the value of the first loss function to perform parameter adjustment processing on the large language model to obtain a trained large language model.

[0091] The first loss function can be used to calculate the difference between the sample satisfaction and the predicted satisfaction.

[0092] Among them, by combining the sample satisfaction, the predicted satisfaction, and the first loss function of the large language model to perform parameter adjustment processing on the large language model, the difference between the predicted satisfaction output by the large language model and the sample satisfaction can be gradually reduced, thereby improving the accuracy of the trained large language model.

[0093] In an embodiment of the present disclosure, in another example, the training data may further include: the sample satisfaction partial order relationship between at least two sample resources; the sample satisfaction partial order relationship between at least two sample resources is determined in combination with the sample satisfaction of at least two sample resources; the sample satisfaction of the sample resource is the satisfaction of the sample object with respect to the sample resource.

[0094] Among them, during the training process of the large language model, considering the satisfaction of the sample object with respect to at least two sample resources and the historical satisfied resources of the sample object can improve the accuracy of the trained large language model.

[0095] In this example, the process of the electronic device executing step 402 may be, for example, inputting the sample resource features, sample object features, and sample historical resource features of at least two sample resources into the large language model to obtain the predicted satisfaction partial order relationship output by the large language model; according to the sample satisfaction partial order relationship, the predicted satisfaction partial order relationship, and the second loss function of the large language model, performing parameter adjustment processing on the large language model to obtain the trained large language model.

[0096] Among them, the second loss function can be used to calculate the difference between the sample satisfaction partial order relationship and the predicted satisfaction partial order relationship.

[0097] Among them, by combining the sample satisfaction partial order relationship, the predicted satisfaction partial order relationship, and the second loss function of the large language model to perform parameter adjustment processing on the large language model, the difference between the predicted satisfaction partial order relationship output by the large language model and the sample satisfaction partial order relationship can be gradually reduced, thereby improving the accuracy of the trained large language model.

[0098] The training method of the large language model for resource recommendation according to the embodiments of the present disclosure includes obtaining the large language model to be trained and the training data of the large language model; the training data includes: the sample resource features of the sample resources, the sample object features of the sample recommendation objects, the sample historical resource features of the sample historical satisfied resources of the sample recommendation objects, and the sample satisfaction degrees of the sample resources; using the sample resource features, the sample object features, the sample historical resource features, and the sample satisfaction degrees to perform training processing on the large language model to obtain the trained large language model for resource recommendation processing; wherein, the large language model is trained in combination with the sample historical resource features of the sample historical satisfied resources of the sample recommendation objects, improving the accuracy of the trained large language model and enabling the large language model to personalized determine the satisfaction degree of the object for the resource.

[0099] To implement the above embodiments, the present disclosure also provides a resource recommendation device. As Figure 5 shown, Figure 5 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 first determination module 502, a second determination module 503, and a selection module 504.

[0100] Among them, the acquisition module 501 is used to obtain the candidate resource features of the candidate resources, the target object features of the target recommendation object, and the historical resource features of the historical satisfied resources of the target recommendation object; the first determination module 502 is used to determine the first satisfaction degree of the candidate resources according to the candidate resource features, the target object features, and the sorting model; the second determination module 503 is used to determine the second satisfaction degree of the candidate resources according to the candidate resource features, the target object features, the historical resource features, and the large language model; the selection module 504 is used to select the resources to be recommended from the candidate resources according to the first satisfaction degree and the second satisfaction degree of the candidate resources for resource recommendation processing of the target recommendation object.

[0101] As a possible implementation manner of the embodiments of the present disclosure, the large language model is trained in combination with the sample resource features of at least two sample resources, the sample object features of the sample objects, the sample historical resource features of the historical satisfied resources of the sample objects, and the sample satisfaction partial order relationship between at least two of the sample resources; wherein, the sample satisfaction partial order relationship between at least two of the sample resources is determined in combination with the satisfaction degrees of the sample object for at least two of the sample resources.

[0102] As a possible implementation manner of the embodiment of the present disclosure, the selection module 504 includes a determination unit and a selection unit; the determination unit is configured to determine a target satisfaction partial order relationship between any two candidate resources in the candidate resources according to the first satisfaction and the second satisfaction of the two candidate resources in the resource pair; the selection unit is configured to select the resource to be recommended from the candidate resources according to the target satisfaction partial order relationship between the two candidate resources in each resource pair.

[0103] As a possible implementation manner of the embodiment of the present disclosure, the determination unit is specifically configured to determine a first weight corresponding to the ranking model and a second weight corresponding to the large language model; perform weighted processing on the first satisfaction and the second satisfaction of the candidate resources in the resource pair according to the first weight and the second weight to obtain the target satisfaction of the candidate resources in the resource pair; determine the target satisfaction partial order relationship between the two candidate resources according to the target satisfaction of the two candidate resources in the resource pair.

[0104] As a possible implementation manner of the embodiment of the present disclosure, the determination unit is specifically configured to determine a first satisfaction partial order relationship between the two candidate resources according to the first satisfaction of the two candidate resources in the resource pair; determine a second satisfaction partial order relationship between the two candidate resources according to the second satisfaction of the two candidate resources in the resource pair; determine the target satisfaction partial order relationship between the two candidate resources according to the first satisfaction partial order relationship and the second satisfaction partial order relationship.

[0105] As a possible implementation manner of the embodiment of the present disclosure, the determination unit is specifically further configured to, when the first satisfaction partial order relationship is inconsistent with the second satisfaction partial order relationship, determine a satisfaction deviation value between the two candidate resources in the resource pair according to the second satisfaction partial order relationship; perform adjustment processing on the first satisfaction of the two candidate resources according to the satisfaction deviation value of the two candidate resources to obtain an adjusted first satisfaction; determine the target satisfaction partial order relationship between the two candidate resources according to the adjusted first satisfaction of the two candidate resources.

[0106] As a possible implementation manner of the embodiment of the present disclosure, the determination unit is specifically further configured to, when the first satisfaction partial order relationship is consistent with the second satisfaction partial order relationship, determine the first satisfaction partial order relationship or the second satisfaction partial order relationship as the target satisfaction partial order relationship.

[0107] As a possible implementation manner of the embodiments of the present disclosure, the target satisfaction partial order relationship between two candidate resources in the resource pair indicates the satisfaction degree relationship between the two candidate resources in the resource pair; specifically, the selection unit is configured to determine a first candidate resource according to the target satisfaction partial order relationship between two candidate resources in each resource pair; the satisfaction degree of the first candidate resource is greater than the satisfaction degrees of a preset number of non-first candidate resources; and select the resource to be recommended from the first candidate resources.

[0108] As a possible implementation manner of the embodiments of the present disclosure, the candidate resource is a resource in a candidate resource list; the candidate resource list is a resource list obtained by the target recommendation object performing resource query processing based on a target query statement; or is a resource list triggered after the target recommendation object selects a resource entry.

[0109] As a possible implementation manner of the embodiments of the present disclosure, the candidate resource is a resource of at least one of the following types: video resource, dynamic resource, graphic and text resource.

[0110] As a possible implementation manner of the embodiments of the present disclosure, the candidate resource features include the text-related content of the candidate resource; the text-related content includes at least one of the following: resource title, resource introduction, and content obtained after text recognition of videos and / or images in the resource.

[0111] The resource recommendation device of the embodiments of the present disclosure obtains candidate resource features of candidate resources, target object features of a target recommendation object, and historical resource features of historical satisfied resources of the target recommendation object; determines a first satisfaction degree of the candidate resource according to the candidate resource features, the target object features, and a sorting model; determines a second satisfaction degree of the candidate resource according to the candidate resource features, the target object features, the historical resource features, and a large language model; and selects a resource to be recommended from the candidate resources for performing resource recommendation processing on the target recommendation object; wherein, the second satisfaction degree is determined by combining the large language model and the historical resource features of the historical satisfied resources of the target recommendation object. Performing resource recommendation processing by combining the first satisfaction degree and the second satisfaction degree can provide a personalized resource recommendation service and improve the resource recommendation efficiency.

[0112] To implement the above embodiments, the present disclosure also provides a training device for a large language model for resource recommendation. As Figure 6 shown, Figure 6 is a schematic diagram according to the sixth embodiment of the present disclosure. The training device 60 for the large language model for resource recommendation may include: an acquisition module 601 and a training module 602.

[0113] Among them, the acquisition module 601 is used to acquire the large language model to be trained and the training data of the large language model; the training data includes: the sample resource features of the sample resources, the sample object features of the sample recommended objects, the sample historical resource features of the sample historical satisfied resources of the sample recommended objects, and the sample satisfaction of the sample resources; the training module 602 is used to perform training processing on the large language model by using the sample resource features, the sample object features, the sample historical resource features, and the sample satisfaction to obtain the trained large language model for resource recommendation processing.

[0114] As a possible implementation manner of the embodiment of the present disclosure, the training module 602 is specifically configured to input the sample resource features, the sample object features, and the sample historical resource features into the large language model to obtain the predicted satisfaction output by the large language model; and perform parameter adjustment processing on the large language model according to the sample satisfaction, the predicted satisfaction, and the first loss function of the large language model to obtain the trained large language model.

[0115] As a possible implementation manner of the embodiment of the present disclosure, the training data further includes: the sample satisfaction partial order relationship between at least two sample resources; the sample satisfaction partial order relationship between at least two sample resources is determined by combining the sample satisfaction of at least two sample resources; the sample satisfaction of the sample resource is the satisfaction of the sample object with respect to the sample resource.

[0116] As a possible implementation manner of the embodiment of the present disclosure, the training module 602 is specifically configured to input the sample resource features, the sample object features, and the sample historical resource features of at least two sample resources into the large language model to obtain the predicted satisfaction partial order relationship output by the large language model; and perform parameter adjustment processing on the large language model according to the sample satisfaction partial order relationship, the predicted satisfaction partial order relationship, and the second loss function of the large language model to obtain the trained large language model.

[0117] As a possible implementation manner of the embodiment of the present disclosure, the sample resource features include the text-related content of the sample resource; the text-related content includes at least one of the following: the resource title, the resource introduction, and the content obtained by performing text recognition on the video and / or image in the resource.

[0118] The training device for a large language model for resource recommendation according to an embodiment of the present disclosure obtains a large language model to be trained and training data for the large language model; the training data includes: sample resource features of sample resources, sample object features of sample recommendation objects, sample historical resource features of sample historical satisfied resources of sample recommendation objects, and sample satisfaction degrees of sample resources; the large language model is trained using the sample resource features, sample object features, sample historical resource features, and sample satisfaction degrees to obtain a trained large language model for resource recommendation processing; wherein, the large language model is trained in combination with the sample historical resource features of the sample historical satisfied resources of the sample recommendation object, improving the accuracy of the trained large language model and enabling the large language model to personalized determine the satisfaction degree of an object for a resource.

[0119] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information and other processes are all carried out on the premise of obtaining the consent of the user, and all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0120] 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.

[0121] Figure 7 A schematic block diagram of an exemplary electronic device 700 that can be used to implement the embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device, 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.

[0122] As Figure 7 shown, the device 700 includes a computing unit 701, which can execute various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0123] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as a keyboard, mouse, etc.; output unit 707, such as various types of displays, speakers, etc.; storage unit 708, such as a disk, optical disc, etc.; and communication unit 709, such as a network card, modem, wireless communication transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0124] Computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing unit 701 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, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 701 executes the various methods and processes described above, such as a resource recommendation method or a training method for a large language model for resource recommendation. For example, in some embodiments, a resource recommendation method or a training method for a large language model for resource recommendation can be implemented as a computer software program, which is 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 onto device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by computing unit 701, one or more steps of the resource recommendation method or the training method for a large language model for resource recommendation described above can be executed. Alternatively, in other embodiments, computing unit 701 can be configured to execute the resource recommendation method or the training method for a large language model for resource recommendation in any other suitable manner (e.g., by means of firmware).

[0125] 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), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: 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, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0126] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program codes can be executed entirely on the machine, partially on the machine, executed partially on the machine as an independent software package and partially on a remote machine, or executed entirely on a remote machine or server.

[0127] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection 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 include, 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 a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0128] In order 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 a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the computer. Other kinds 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).

[0129] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0130] A computer system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client - server relationship is created by computer programs running on the respective computers and having a client - server relationship with each other. The server can be a cloud server, a server of a distributed system, or a server incorporating blockchain.

[0131] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0132] The above - described specific embodiments do not constitute a limitation on the protection scope 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 shall be included within the protection scope of this disclosure.

Claims

1. A resource recommendation method, the method comprising: Acquire candidate resource features of the candidate resource, target object features of the target recommendation object, and historical resource features of historically satisfactory resources of the target recommendation object; Determining a first satisfaction level of the candidate resource according to the candidate resource characteristics, the target object characteristics and the ranking model; Determining a second satisfaction level of the candidate resource according to the candidate resource feature, the target object feature, the historical resource feature, and the large language model; According to the first satisfaction level and the second satisfaction level of the candidate resources, a resource to be recommended is selected from the candidate resources for performing resource recommendation processing on the target recommendation object.

2. The method according to claim 1, wherein: The large language model is trained by combining sample resource features of at least two sample resources, sample object features of sample objects, sample historical resource features of historically satisfactory resources of sample objects, and sample satisfaction partial order relations between at least two of the sample resources; The sample satisfaction partial order relationship between at least two of the sample resources is determined by combining the satisfaction of the sample object with respect to at least two of the sample resources.

3. The method according to claim 1, wherein: The selecting a resource to be recommended from the candidate resources according to the first satisfaction level and the second satisfaction level of the candidate resources comprises: For a resource pair consisting of any two candidate resources among the candidate resources, determining a target satisfaction partial order relationship between the two candidate resources according to first satisfaction levels and second satisfaction levels of the two candidate resources in the resource pair; The resource to be recommended is selected from the candidate resources according to the target satisfaction partial order relationship between two candidate resources in each of the resource pairs.

4. The method according to claim 3, wherein: The determining, according to the first satisfaction level and the second satisfaction level of the two candidate resources in the resource pair, a target satisfaction partial order relationship between the two candidate resources comprises: Determining a first weight corresponding to the ranking model and a second weight corresponding to the large language model; Performing weighted processing on the first satisfaction degree and the second satisfaction degree of the candidate resources in the resource pair according to the first weight and the second weight to obtain a target satisfaction degree of the candidate resources in the resource pair; According to the target satisfaction levels of two candidate resources in the resource pair, a target satisfaction partial order relationship between the two candidate resources is determined.

5. The method according to claim 3, wherein: The determining, according to the first satisfaction level and the second satisfaction level of the two candidate resources in the resource pair, a target satisfaction partial order relationship between the two candidate resources comprises: Determine a first satisfactory partial order relationship between the two candidate resources according to first satisfaction levels of the two candidate resources in the resource pair; Determine a second satisfactory partial order relationship between the two candidate resources according to the second satisfaction levels of the two candidate resources in the resource pair; A target satisfactory partial order relationship between the two candidate resources is determined according to the first satisfactory partial order relationship and the second satisfactory partial order relationship.

6. The method according to claim 5, wherein: The determining, according to the first satisfactory partial order relationship and the second satisfactory partial order relationship, a target satisfactory partial order relationship between the two candidate resources includes: In the case where the first satisfactory partial order relationship is inconsistent with the second satisfactory partial order relationship, determining satisfaction deviation values ​​of two candidate resources in the resource pair according to the second satisfactory partial order relationship; According to the satisfaction degree deviation values ​​of the two candidate resources, adjusting the first satisfaction degrees of the two candidate resources to obtain adjusted first satisfaction degrees; A target satisfaction partial order relationship between the two candidate resources is determined according to the adjusted first satisfaction levels of the two candidate resources.

7. The method according to claim 6, wherein: The determining a target satisfactory partial order relationship between the two candidate resources according to the first satisfactory partial order relationship and the second satisfactory partial order relationship also includes: When the first satisfactory partial order relationship is consistent with the second satisfactory partial order relationship, the first satisfactory partial order relationship or the second satisfactory partial order relationship is determined as the target satisfactory partial order relationship.

8. The method according to claim 3, wherein: A target satisfaction partial order relationship between two candidate resources in the resource pair, indicating a satisfaction degree relationship between the two candidate resources in the resource pair; The selecting the resource to be recommended from the candidate resources according to the target satisfaction partial order relationship between the two candidate resources in each resource pair includes: Determine a first candidate resource according to a target satisfaction partial order relationship between two candidate resources in each resource pair; the satisfaction level of the first candidate resource is greater than the satisfaction levels of a preset number of non-first candidate resources; The resource to be recommended is selected from the first candidate resources.

9. The method according to claim 1, wherein: The candidate resource is a resource in a candidate resource list; The candidate resource list is a resource list obtained by performing resource query processing on the target recommendation object based on the target query statement; or, it is a resource list triggered after the target recommendation object selects a resource entry.

10. The method according to claim 1, wherein: The candidate resource is a resource of at least one of the following types: a video resource, a dynamic resource, and a graphic resource.

11. The method according to claim 1 or 10, wherein: The candidate resource features include text-related content of the candidate resource; The text-related content includes at least one of the following: a resource title, a resource introduction, and content obtained by performing text recognition on a video and / or image in the resource.

12. A method for training a large language model for resource recommendation, the method comprising: Obtaining a large language model to be trained and training data of the large language model; The training data includes: sample resource characteristics of sample resources, sample object characteristics of sample recommended objects, sample historical resource characteristics of sample historically satisfied resources of the sample recommended objects, and sample satisfaction of the sample resources; The large language model is trained using the sample resource features, the sample object features, the sample historical resource features and the sample satisfaction to obtain a trained large language model for resource recommendation processing.

13. The method according to claim 12, wherein: The large language model is trained using the sample resource feature, the sample object feature, the sample historical resource feature, and the sample satisfaction level to obtain a trained large language model, including: Inputting the sample resource feature, the sample object feature and the sample historical resource feature into the large language model to obtain the predicted satisfaction level output by the large language model; According to the sample satisfaction, the predicted satisfaction and the first loss function of the large language model, parameter adjustment processing is performed on the large language model to obtain a trained large language model.

14. The method according to claim 12, wherein: The training data also includes: a sample satisfaction partial order relationship between at least two sample resources; The sample satisfaction partial order relationship between at least two of the sample resources is determined by combining the sample satisfaction of at least two of the sample resources; the sample satisfaction of the sample resources is the satisfaction of the sample object with respect to the sample resources.

15. The method according to claim 14, wherein: The large language model is trained using the sample resource feature, the sample object feature, the sample historical resource feature, and the sample satisfaction level to obtain a trained large language model, including: Inputting sample resource features of at least two of the sample resources, the sample object features, and the sample historical resource features into the large language model to obtain a predicted satisfactory partial order relationship output by the large language model; According to the sample satisfaction partial order relationship, the predicted satisfaction partial order relationship and the second loss function of the large language model, parameter adjustment processing is performed on the large language model to obtain a trained large language model.

16. The method according to claim 12, wherein: The sample resource characteristics include text-related content of the sample resource; The text-related content includes at least one of the following: a resource title, a resource introduction, and content obtained by performing text recognition on a video and / or image in the resource.

17. A resource recommendation device, comprising: An acquisition module, used to acquire candidate resource features of candidate resources, target object features of target recommendation objects, and historical resource features of historically satisfactory resources of the target recommendation objects; A first determination module, configured to determine a first satisfaction level of the candidate resource according to the candidate resource characteristics, the target object characteristics and a ranking model; A second determination module, configured to determine a second satisfaction level of the candidate resource according to the candidate resource feature, the target object feature, the historical resource feature and the large language model; A selection module is used to select a resource to be recommended from the candidate resources according to a first satisfaction level and a second satisfaction level of the candidate resources, and to perform resource recommendation processing on the target recommendation object.

18. The device according to claim 17, wherein: The large language model is trained by combining sample resource features of at least two sample resources, sample object features of sample objects, sample historical resource features of historically satisfactory resources of sample objects, and sample satisfaction partial order relations between at least two of the sample resources; The sample satisfaction partial order relationship between at least two of the sample resources is determined by combining the satisfaction of the sample object with respect to at least two of the sample resources.

19. The device according to claim 17, wherein: The selection module includes a determination unit and a selection unit; The determining unit is used to determine, for a resource pair consisting of any two candidate resources among the candidate resources, a target satisfaction partial order relationship between the two candidate resources according to a first satisfaction level and a second satisfaction level of the two candidate resources in the resource pair; The selection unit is used to select the resource to be recommended from the candidate resources according to the target satisfaction partial order relationship between two candidate resources in each resource pair.

20. The device according to claim 19, wherein The determining unit is specifically configured to: Determining a first weight corresponding to the ranking model and a second weight corresponding to the large language model; Performing weighted processing on the first satisfaction degree and the second satisfaction degree of the candidate resources in the resource pair according to the first weight and the second weight to obtain a target satisfaction degree of the candidate resources in the resource pair; According to the target satisfaction levels of two candidate resources in the resource pair, a target satisfaction partial order relationship between the two candidate resources is determined.

21. The device according to claim 19, wherein The determining unit is specifically configured to: Determine a first satisfactory partial order relationship between the two candidate resources according to first satisfaction levels of the two candidate resources in the resource pair; Determine a second satisfactory partial order relationship between the two candidate resources according to the second satisfaction levels of the two candidate resources in the resource pair; A target satisfactory partial order relationship between the two candidate resources is determined according to the first satisfactory partial order relationship and the second satisfactory partial order relationship.

22. The device according to claim 21, wherein The determining unit is further specifically configured to: In the case where the first satisfactory partial order relationship is inconsistent with the second satisfactory partial order relationship, determining satisfaction deviation values ​​of two candidate resources in the resource pair according to the second satisfactory partial order relationship; According to the satisfaction degree deviation values ​​of the two candidate resources, adjusting the first satisfaction degrees of the two candidate resources to obtain adjusted first satisfaction degrees; A target satisfaction partial order relationship between the two candidate resources is determined according to the adjusted first satisfaction levels of the two candidate resources.

23. The device according to claim 22, wherein: The determining unit is further specifically configured to: When the first satisfactory partial order relationship is consistent with the second satisfactory partial order relationship, the first satisfactory partial order relationship or the second satisfactory partial order relationship is determined as the target satisfactory partial order relationship.

24. The device according to claim 19, wherein The target satisfaction partial order relationship between the two candidate resources in the resource pair indicates the satisfaction relationship between the two candidate resources in the resource pair; the selection unit is specifically used to: Determine a first candidate resource according to a target satisfaction partial order relationship between two candidate resources in each resource pair; the satisfaction level of the first candidate resource is greater than the satisfaction levels of a preset number of non-first candidate resources; The resource to be recommended is selected from the first candidate resources.

25. The device according to claim 17, wherein: The candidate resource is a resource in a candidate resource list; The candidate resource list is a resource list obtained by performing resource query processing on the target recommendation object based on the target query statement; or, it is a resource list triggered after the target recommendation object selects a resource entry.

26. The device according to claim 17, wherein: The candidate resource is a resource of at least one of the following types: a video resource, a dynamic resource, and a graphic resource.

27. The device according to claim 17 or 26, wherein: The candidate resource features include text-related content of the candidate resource; The text-related content includes at least one of the following: a resource title, a resource introduction, and content obtained by performing text recognition on a video and / or image in the resource.

28. A training device for a large language model for resource recommendation, the device comprising: An acquisition module, used to acquire a large language model to be trained and training data of the large language model; The training data includes: sample resource characteristics of sample resources, sample object characteristics of sample recommended objects, sample historical resource characteristics of sample historically satisfied resources of the sample recommended objects, and sample satisfaction of the sample resources; The training module is used to train the large language model using the sample resource characteristics, the sample object characteristics, the sample historical resource characteristics and the sample satisfaction, and obtain the trained large language model for resource recommendation processing.

29. The device according to claim 28, wherein The training module is specifically used for: Inputting the sample resource feature, the sample object feature and the sample historical resource feature into the large language model to obtain the predicted satisfaction level output by the large language model; According to the sample satisfaction, the predicted satisfaction and the first loss function of the large language model, parameter adjustment processing is performed on the large language model to obtain a trained large language model.

30. The device according to claim 28, wherein The training data also includes: a sample satisfaction partial order relationship between at least two sample resources; The sample satisfaction partial order relationship between at least two of the sample resources is determined by combining the sample satisfaction of at least two of the sample resources; the sample satisfaction of the sample resources is the satisfaction of the sample object with respect to the sample resources.

31. The device according to claim 30, wherein The training module is specifically used for: Inputting sample resource features of at least two of the sample resources, the sample object features, and the sample historical resource features into the large language model to obtain a predicted satisfactory partial order relationship output by the large language model; According to the sample satisfaction partial order relationship, the predicted satisfaction partial order relationship and the second loss function of the large language model, parameter adjustment processing is performed on the large language model to obtain a trained large language model.

32. The apparatus of claim 28, wherein: The sample resource characteristics include text-related content of the sample resource; The text-related content includes at least one of the following: a resource title, a resource introduction, and content obtained by performing text recognition on a video and / or image in the resource.

33. 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 execute the method of any one of claims 1 to 11; or, execute the method of any one of claims 12 to 16.

34. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 11; or, to execute the method according to any one of claims 12 to 16.

35. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 11; or, implements the method according to any one of claims 12 to 16.

Images