Multimedia recommendation method, device, equipment and storage medium

Abstract

The present disclosure provides a multimedia recommendation method, apparatus, device and storage medium, which relate to the field of data processing, especially to the fields of artificial intelligence, intelligent recommendation, etc. The specific implementation scheme is as follows: obtaining M1 first candidate multimedia resources, the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on N1 target external domain interest features; the target external domain interest features are obtained based on the behavioral characteristics of the target object in the source domain; obtaining M2 second candidate multimedia resources, the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target local domain interest features; the target local domain interest features are obtained based on the behavioral characteristics of the target object in the target domain; based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources, obtaining M3 target multimedia resources to be recommended to the target object.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to technical fields such as artificial intelligence and intelligent recommendation. Background Art

[0002] Personalized recommendations, which recommend content of interest to a target user based on their historical behavior, have been widely used in various areas of life. While the target user's historical behavior in the current domain can reflect their current preferences, in practice, recommendations based solely on preferences in the current domain can lead to overly narrow recommendations, creating a cocooning effect and degrading the user experience. Summary of the Invention

[0003] The present disclosure provides a multimedia recommendation method, apparatus, device, and storage medium.

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

[0005] Obtain M1 first candidate multimedia resources, wherein the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are based on the behavioral characteristics of the target object in the source domain; N1 and M1 are both positive integers greater than or equal to 1;

[0006] Obtain M2 second candidate multimedia resources, wherein the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; and N2 and M2 are both positive integers greater than or equal to 1;

[0007] Based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained; M3 is a positive integer greater than or equal to 1.

[0008] According to another aspect of the present disclosure, a multimedia recommendation device is provided, comprising:

[0009] A first resource determination unit is configured to obtain M1 first candidate multimedia resources, wherein the M1 first candidate multimedia resources are multimedia resources to be recommended to a target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are based on behavioral features of the target object in a source domain; and N1 and M1 are both positive integers greater than or equal to 1;

[0010] A second resource determination unit is configured to obtain M2 second candidate multimedia resources, wherein the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; and N2 and M2 are both positive integers greater than or equal to 1;

[0011] The recommendation unit is configured to obtain M3 target multimedia resources to be recommended to the target object based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources; M3 is a positive integer greater than or equal to 1.

[0012] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0013] at least one processor; and

[0014] a memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any method in the embodiments of the present disclosure.

[0016] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any method according to the embodiments of the present disclosure.

[0017] According to another aspect of the present disclosure, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the computer program implements any one of the methods according to the embodiments of the present disclosure.

[0018] In this way, since the target multimedia resources finally recommended to the target object obtained by the disclosed solution are based on the first candidate multimedia resources and the second candidate multimedia resources, it fully refers to the interest characteristics of the target object in the target domain and the source domain. In this way, it effectively avoids the problem of the recommended content being too single or even forming a cocoon effect, thereby laying the foundation for effectively improving the user experience.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0021] Figure 1 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 1 ;

[0022] Figure 2 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 2 ;

[0023] Figure 3 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 3 ;

[0024] Figure 4 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 4 ;

[0025] Figure 5 is a flowchart of a multimedia recommendation method in an example according to an embodiment of the present application;

[0026] Figure 6 is a structural diagram of a multimedia recommendation device according to an embodiment of the present application;

[0027] Figure 7 It is a block diagram of an electronic device for implementing the multimedia recommendation method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope 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.

[0029] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The term "at least one" in this article means any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C, can mean including any one or more elements selected from the set consisting of A, B, and C. The terms "first" and "second" in this article refer to multiple similar technical terms and distinguish them, and do not mean to limit the order or to limit to only two. For example, the first feature and the second feature refer to two categories / two features. The first feature can be one or more, and the second feature can also be one or more.

[0030] In addition, numerous specific details are provided in the following detailed description to better illustrate the present disclosure. Those skilled in the art will appreciate that the present disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main points of the present disclosure.

[0031] Personalized recommendations, which recommend content of interest to a target user based on their historical behavior, have been widely used in various areas of life. Currently, recommendation systems typically employ two primary processing logics: recall and recommendation. Recall is used to generate candidate content for recommendation, aiming to initially retrieve candidate content from multimedia databases that the target user may be interested in. The recall results are typically input to sorting processing logic, which is used to rank the recalled candidate content. For example, the candidate results can be ranked based on the target user's level of interest, aiming to prioritize the content that the user is most interested in.

[0032] In the actual recommendation process, recommendations are often made based on the target object's historical behavior in the target domain. Although the target object's historical behavior in the target domain can reflect the target object's preferences in the current domain, if recommendations are made only based on the preferences in the current domain, the recommended content will inevitably be too monotonous, forming a cocoon effect, and thus reducing the user experience.

[0033] Based on this, the disclosed solution provides a cross-domain interest exploration technology to achieve cross-domain recommendation. Here, cross-domain recommendation (CDR) refers to transferring knowledge learned in the source domain to the target domain to balance the recommendation results.

[0034] Based on this, the cross-domain interest exploration technology of the disclosed solution can effectively solve the problem that the recommendation system in the current immersive video recommendation system over-utilizes the short-term signals of the target domain (for example, historical behavior in the short term), making the recommended content more monotonous and cocooned, thereby affecting the user experience.

[0035] Specifically, Figure 1 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 1 The method may be optionally applied to electronic devices, such as personal computers, servers, server clusters, and other electronic devices.

[0036] Furthermore, the method includes at least part of the following contents. Figure 1 Shown, including:

[0037] Step S101: Obtain M1 first candidate multimedia resources; wherein, the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are obtained based on the behavioral characteristics of the target object in the source domain.

[0038] Here, N1 and M1 are both positive integers greater than or equal to 1.

[0039] Step S102: Obtain M2 second candidate multimedia resources; wherein, the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are obtained based on the behavioral characteristics of the target object in the target domain.

[0040] Here, N2 and M2 are both positive integers greater than or equal to 1.

[0041] It should be noted that, in a specific example, the target domain may specifically refer to the domain or application that the target object is currently browsing. Accordingly, the source domain refers to a domain or application that is different from the target domain. For example, the target domain may specifically refer to the first application that the target object is currently browsing, while the source domain refers to the second application that the target object has previously browsed. Alternatively, the target domain may specifically refer to the first domain that the target object is currently browsing, while the source domain refers to the second domain that the target object has previously browsed.

[0042] Furthermore, it should be pointed out that the "local domain" and "external domain" described in the present disclosure are a set of relative probabilities. For example, if the domain where the target object is currently located is the target domain, then, relative to the target object, the target domain can be understood as the local domain, and the source domain is the external domain.

[0043] Step S103: Based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources, obtain M3 target multimedia resources to be recommended to the target object; M3 is a positive integer greater than or equal to 1.

[0044] In this way, since the first candidate multimedia resource obtained by the disclosed scheme is based on the target external domain interest characteristics of the target object, and the target external domain interest characteristics are obtained based on the behavioral characteristics of the target object in the source domain (that is, the external domain of the target domain), and since the target multimedia resource finally recommended to the target object is based on the first candidate multimedia resource and the second candidate multimedia resource, the second candidate multimedia resource is based on the target local domain interest characteristics of the target object, and the target local domain interest characteristics are based on the behavioral characteristics of the target object in the target domain (that is, the local domain), the target multimedia resource finally recommended to the target object fully refers to the interest characteristics of the target object in the target domain and the source domain. In this way, the problem of the recommended content being too single or even forming a cocoon effect is effectively avoided, thereby laying the foundation for effectively improving the user experience.

[0045] In a specific example of the disclosed solution, the target external-domain interest feature and / or the target local-domain interest feature may be updated based on the feedback result (e.g., posterior result) of the target object. For example, after step S103, the following steps may be further included:

[0046] Based on the posterior results of the M3 target multimedia resources (for example, feedback results on whether to click, etc.), the N1 target external domain interest features and / or the N2 target local domain interest features are updated. Here, "update" may specifically include: optimization operations such as adding, replacing or deleting interest features. In this way, the recommendation results finally recommended to the target object can be adaptively adjusted based on the real-time feedback of the target object, thereby effectively avoiding overly single recommendations and further improving the user experience.

[0047] In a specific example of the disclosed solution, N1 target external-domain interest features may be obtained in the following manner; for example, before obtaining M1 first candidate multimedia resources (for example, before step S101), the method further includes:

[0048] Step S100 - 1 : Acquire at least one initial external domain explicit interest feature of the target object in the source domain.

[0049] Step S100 - 2 : Acquire at least one extended external domain interest feature of the target object in the source domain.

[0050] In a specific example, interest expansion can be performed in the following manner to obtain an expanded external-domain interest feature, that is, interest expansion is performed on at least one initial external-domain explicit interest feature to obtain at least one expanded external-domain interest feature (for example, step S100-2), specifically including at least one of the following expansion methods:

[0051] Method 1: Determine the mutual information parameter between the initial external domain explicit interest feature and the interest feature to be recommended in the at least one initial external domain explicit interest feature, and use the interest feature to be recommended whose mutual information parameter meets the first preset requirement as the expanded external domain interest feature; wherein the mutual information parameter represents the interest correlation between the initial external domain explicit interest feature and the interest feature to be recommended.

[0052] Here, the mutual information parameter can represent the closeness between two interest features. For details, please refer to the following description and will not be repeated here.

[0053] Method 2: Using the interest features of the target group corresponding to the target object in the source domain as the expanded external domain interest features.

[0054] It is understandable that the above methods can be performed one by one, or both methods can be selected to obtain expanded foreign domain interest features.

[0055] In this way, the disclosed solution provides an expansion method that can determine more external domain interest features, thereby improving the richness of external domain interest features. In this way, the disclosed solution fully refers to the interest features of the source domain, rather than being limited to the interest features of the target domain, laying the foundation for effectively avoiding the problem of recommended content being too single or even forming a cocoon effect.

[0056] Step S100 - 3 : obtaining the N1 target external-domain interest features based at least on the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature.

[0057] For example, in one example, both the initial external-domain explicit interest features and the expanded external-domain interest features mined are directly used as target external-domain interest features. In this case, the total number of target external-domain interest features obtained = the number of initial external-domain explicit interest features + the number of expanded external-domain interest features. Alternatively, in another example, a preset mechanism can be used to filter at least one target external-domain interest feature from the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature.

[0058] It should be noted that "explicit interest features" can be understood as the interests displayed by the target object. Specifically, they can be obtained through explicit mining based on sample data, such as data related to clicks on video resources. These "explicit interest features" have clear physical meanings, for example, "explicit interest features" represent football or basketball. Correspondingly, the "external domain explicit interest features" of the disclosed solution represent the interest features of the target object obtained through explicit mining based on sample data of the target object in the source domain.

[0059] In a specific example, the initial external domain explicit interest features mentioned above can also be specifically referred to as trusted external domain explicit interest features. The trusted external domain explicit interest features can be understood as initial external domain explicit interest features whose confidence meets preset requirements. In this way, it is ensured that the interest features explored in the source domain have a certain correlation with the interest features of the target domain, thereby ensuring that the recommended content is the content that the target object is interested in, laying the foundation for further effectively improving the user experience.

[0060] That is to say, the mining of target external domain interest features in the disclosed solution not only refers to the explicit interest features of the target object in the source domain, but also refers to the expanded external domain interest features. In this way, the breadth of external domain interest features is further enriched, thereby further laying the foundation for effectively avoiding overly single recommendations and avoiding the cocoon effect.

[0061] Figure 2 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 2 The method may be optionally applied to electronic devices, such as personal computers, servers, server clusters, and other electronic devices.

[0062] Furthermore, the method includes at least part of the following contents. Figure 2 Shown, including:

[0063] Step S201: Acquire at least one initial external domain explicit interest feature of the target object in the source domain.

[0064] Step S202: Obtain at least one extended external domain interest feature of the target object in the source domain.

[0065] Step S203: Acquire at least one initial external domain implicit interest feature of the target object in the source domain.

[0066] It is understandable that the steps of obtaining the initial external domain explicit interest features, expanding the external domain interest features, and the initial external domain implicit interest features can be swapped. For example, step S203 can be executed first, and then step S201 and step S202 can be executed. The present disclosure does not limit this.

[0067] Step S204: obtaining the N1 target external-domain interest features based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature.

[0068] For example, in one example, the initial external-domain explicit interest features, the expanded external-domain interest features, and the initial external-domain implicit interest features mined are all used as target external-domain interest features. In this case, the total number of target external-domain interest features obtained = the number of initial external-domain explicit interest features + the number of expanded external-domain interest features + the number of initial external-domain implicit interest features. Alternatively, in another example, a preset mechanism can be used to filter out at least one target external-domain interest feature from the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature.

[0069] Here, it should be noted that "implicit interest features" can be understood as interests that are difficult for the target object to express directly but have potential tendencies. Specifically, they can be obtained by implicit mining based on sample data, such as data from clicking on video resources. This "implicit interest feature" has no clear physical meaning. For example, "implicit interest features" are expressed as vector expressions of multiple interest features. In this case, the vector expression has no clear interest meaning. Correspondingly, the "external domain implicit interest features" of the disclosed solution represent the interest features of the target object obtained by implicit mining based on the sample data of the target object in the source domain.

[0070] That is to say, the mining of target external domain interest features in the disclosed scheme not only refers to the explicit interest features and implicit interest features of the target object in the source domain, but also refers to the expanded external domain interest features, so as to further enrich the breadth of external domain interest features, and further lay the foundation for effectively avoiding overly single recommendations and avoiding the cocoon effect.

[0071] In one specific example, the initial external-domain implicit interest features described above can also be specifically referred to as trusted external-domain implicit interest features. These trusted external-domain implicit interest features can be understood as initial external-domain implicit interest features whose confidence level meets preset requirements. This ensures that the interest features discovered in the source domain are correlated with those in the target domain, thereby ensuring that the recommended content is of interest to the target audience, laying the foundation for further improving the user experience.

[0072] In a specific example, N1 target external-domain interest features may be obtained in the following manner. Specifically, the above-described process of obtaining the N1 target external-domain interest features based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature (e.g., step S204) specifically includes:

[0073] Step S204-1: based at least on the activity level of each element in the first initial external-domain interest set, delete at least some elements from the first initial external-domain interest set to obtain a target initial external-domain interest set.

[0074] Here, the first initial external-domain interest set includes at least one initial external-domain explicit interest feature, at least one extended external-domain interest feature, and at least one initial external-domain implicit interest feature.

[0075] In a specific example, the "activity level" mentioned above can be specifically expressed by relative entropy. In this way, relative entropy is used to measure the relative activity of each interest feature. On the basis of effectively avoiding the problem of recommended content being too single or even forming a cocoon effect, it ensures that the recommended content is the content that the target object is interested in, thereby laying the foundation for effectively improving the user experience.

[0076] Step S204 - 2 : adopting a preset sampling rule (eg, Thompson sampling) to sample from the target initial external-domain interest set to obtain the N1 target external-domain interest features.

[0077] In this way, since the disclosed solution can screen interest features based on the activity level of each element (for example, the activity level of the initial external domain explicit interest features, the activity level of the expanded external domain interest features, and the activity level of the initial external domain implicit interest features), and then sample interest features from the set obtained after screening, and obtain the target external domain interest features required for the recall process, it can effectively avoid the problem of recommended content being too single or even forming a cocoon effect, and ensure that the recommended content is the content that the target object is interested in, thereby laying the foundation for effectively improving the user experience.

[0078] Furthermore, in a specific example, during the screening process, an exit mechanism for interest features (e.g., a posteriori results, exit time, etc.) can also be considered to effectively avoid the monotony of recommendations and the information cocoon problem. Specifically, the above-described process of removing at least some elements from the constructed first initial external interest set based on the activity level of each element to obtain a target initial external interest set (e.g., step S204-1) specifically includes:

[0079] Based on the activity level of each element in the first initial external domain interest set and based on at least one of the following information, at least some elements are deleted from the first initial external domain interest set to obtain a target initial external domain interest set: a posterior result of the element, and an exit time of the element.

[0080] That is to say, in the process of screening interest features, the disclosed solution not only fully considers the activity level of each element, but also fully considers the exit factors of each element. In this way, it can effectively avoid the problem of too much reverse order of sorting results and the inability to effectively exit interest features, which leads to the recommendation content being too single, thereby laying the foundation for further improving the user experience.

[0081] Step S205: Obtain M1 first candidate multimedia resources; wherein, the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on the N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are obtained based on the behavioral characteristics of the target object in the source domain; N1 and M1 are both positive integers greater than or equal to 1.

[0082] Step S206: Obtain M2 second candidate multimedia resources; wherein, the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; N2 and M2 are both positive integers greater than or equal to 1.

[0083] In a specific example, before step S206, the target domain interest feature can also be obtained in the following manner, for example:

[0084] Method 1: Acquire at least one initial local domain explicit interest feature of the target object in the target domain, and obtain a target local domain interest feature based on the at least one initial local domain explicit interest feature of the target object in the target domain.

[0085] Method 2: Obtain at least one initial local domain explicit interest feature of the target object in the target domain, expand the interest of at least one initial local domain explicit interest feature, and obtain at least one expanded local domain interest feature; based on the at least one initial local domain explicit interest feature and the at least one expanded local domain interest feature, obtain the N2 target local domain interest features.

[0086] Method three: obtain at least one initial local domain explicit interest feature of the target object in the target domain, expand the interest of at least one initial local domain explicit interest feature, and obtain at least one expanded local domain interest feature; obtain at least one initial local domain implicit interest feature of the target object in the target domain; and then obtain the N2 target local domain interest features based on the at least one initial local domain explicit interest feature, the at least one expanded local domain interest feature, and the at least one initial local domain implicit interest feature.

[0087] The three methods described above can be executed in any one of the above methods, and the present disclosure does not limit this.

[0088] Here, the "local domain explicit interest features" of the disclosed solution refer to the interest features of the target object obtained by explicitly mining the sample data of the target object in the target domain. Correspondingly, the "local domain implicit interest features" of the disclosed solution refer to the interest features of the target object obtained by implicitly mining the sample data of the target object in the target domain.

[0089] Step S207: Based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources, obtain M3 target multimedia resources to be recommended to the target object; M3 is a positive integer greater than or equal to 1.

[0090] In this way, the disclosed solution can not only effectively enrich the breadth of external domain interest features, but also effectively ensure the correlation between external domain interest features and local domain interest features. In this way, while effectively avoiding the problem of recommended content being too single or even forming a cocoon effect, it can expand the breadth of recommended content. At the same time, it can also ensure that the recommended content is content that the target object is interested in, thereby laying the foundation for effectively improving user experience.

[0091] Figure 3 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 3 The method may be optionally applied to electronic devices, such as personal computers, servers, server clusters, and other electronic devices.

[0092] Furthermore, the method includes at least part of the following contents. Figure 3 Shown, including:

[0093] Step S301: Based on N1 target external domain interest features, M1 first candidate multimedia resources are recalled from the multimedia resource library of the source domain.

[0094] It is understandable that, in this example, the above-mentioned related methods can be used to obtain N1 target out-of-domain interest features, which will not be described in detail here.

[0095] Step S302: input the N1 target external-domain interest features and the M1 first candidate multimedia resources into an external-domain ranking model to obtain the ranked M1 first candidate multimedia resources.

[0096] Here, the external domain sorting model is obtained by training the first preset sorting model based on the first sample data of the source domain; the first sample data at least represents the association relationship between multimedia resources (for example, multimedia resources in the multimedia database of the source domain) and the interest features of the object in the source domain.

[0097] That is to say, the external domain ranking model of the disclosed solution is trained based on sample data of the source domain, rather than sample fusion data of the target domain and the source domain. In this way, in order to effectively avoid the problems such as the uneven distribution of samples in the target domain and the source domain, which leads to the weakening of certain interest features and the difficulty of stably exposing external domain interest features, it is ensured that the external domain interest features can be learned losslessly.

[0098] In one example, the output result of the external domain ranking model can be specifically the click probability of each input first candidate multimedia resource. In this way, based on the click probability, M1 first candidate multimedia resources can be sorted. For example, based on the click probability, sorting is performed by size to obtain M1 first candidate multimedia resources.

[0099] Step S303: Obtain M2 second candidate multimedia resources.

[0100] Here, the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on the N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; N2 and M2 are both positive integers greater than or equal to 1.

[0101] Step S304: Based on the sorted M1 first candidate multimedia resources and the M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained.

[0102] In a specific example, M3 target multimedia resources may be obtained in the following manner, that is, the above-mentioned step S304 specifically includes:

[0103] Selecting the first candidate multimedia resource ranked first from the sorted M1 first candidate multimedia resources; for example, sorting them from largest to smallest based on click probability. In this case, the first candidate multimedia resource ranked first is the content with the highest click probability;

[0104] The first candidate multimedia resource ranked first is inserted into the M2 second candidate multimedia resources to obtain M3 target multimedia resources. At this time, the multimedia resources to be exposed to the target object include content recommended based on external domain interest features.

[0105] Alternatively, the above-mentioned step S304 specifically includes:

[0106] Selecting the first candidate multimedia resource ranked first from the sorted M1 first candidate multimedia resources, for example, sorting based on click probability from large to small. In this case, the first candidate multimedia resource ranked first is the content with the highest click probability;

[0107] Randomly remove one from the M2 second candidate multimedia resources, and insert the first candidate multimedia resource ranked first into the position of the removed second candidate multimedia resource to obtain M3 target multimedia resources. At this time, the multimedia resources to be exposed to the target object include content recommended based on external domain interest features.

[0108] In this way, the disclosed solution uses the external domain ranking model obtained by training sample data based on the source domain to rank the candidate multimedia resources (i.e., M1 first candidate multimedia resources) recalled based on the target external domain interest features, and then combines the candidate multimedia resources recalled based on the target domain interest features (i.e., M2 first candidate multimedia resources) to obtain the M3 target multimedia resources finally recommended to the target object. In this way, the problem of uneven sample distribution between the target domain and the source domain, which leads to the weakening of certain interest features and the difficulty of stable exposure of external domain interest features, is effectively avoided, thereby ensuring that the external domain interest features can be learned losslessly, and further effectively avoiding the problem of the recommended content being too single or even forming a cocoon effect, further laying the foundation for effectively improving the user experience.

[0109] Figure 4 This is a schematic flow chart of a multimedia recommendation method according to an embodiment of the present application. Figure 4 The method may be optionally applied to electronic devices, such as personal computers, servers, server clusters, and other electronic devices.

[0110] Furthermore, the method includes at least part of the following contents. Figure 4 Shown, including:

[0111] Step S401: Based on N1 target external domain interest features, M1 first candidate multimedia resources are recalled from the multimedia resource library of the source domain.

[0112] It is understandable that, in this example, the above-mentioned related methods can be used to obtain N1 target out-of-domain interest features, which will not be described in detail here.

[0113] Step S402: inputting the N1 target external-domain interest features and the M1 first candidate multimedia resources into an external-domain ranking model to obtain the ranked M1 first candidate multimedia resources.

[0114] Here, the external domain sorting model is obtained by training the first preset sorting model based on the first sample data of the source domain; the first sample data at least represents the association relationship between multimedia resources (for example, multimedia resources in the multimedia database of the source domain) and the interest features of the object in the source domain.

[0115] Here, the relevant content about the external domain sorting model can be found in the above description and will not be repeated here.

[0116] Step S403: Based on the N2 target domain interest features, M2 second candidate multimedia resources are recalled from the multimedia resource library of the target domain.

[0117] Step S404: input the N2 target local domain interest features and the M2 second candidate multimedia resources into a local domain ranking model to obtain ranked M2 second candidate multimedia resources.

[0118] Here, the local domain ranking model is obtained by training a second preset ranking model based on second sample data of the target domain; the second sample data at least includes the association relationship between multimedia resources and interest features of objects in the target domain.

[0119] In one example, the output result of the domain ranking model can also be specifically the click probability of each input second candidate multimedia resource. In this way, M2 sorted second candidate multimedia resources can be obtained based on the click probability. For example, based on the click probability, sorting is performed by size to obtain M2 sorted second candidate multimedia resources.

[0120] It should be noted that the first preset sorting model and the second preset sorting model mentioned above may be the same or different, and the present disclosure does not impose any limitation on this.

[0121] In addition, it should be noted that the steps of obtaining the sorted M1 first candidate multimedia resources and the steps of obtaining the sorted M2 second candidate multimedia resources can be swapped, that is, first execute step S403 and step S404, and then execute step S401 and step S402. The present disclosure does not limit this.

[0122] That is to say, the external domain ranking model of the disclosed solution is obtained by training based on the sample data of the source domain, rather than the sample fusion data of the target domain and the source domain. The local domain ranking model of the disclosed solution is obtained by training based on the sample data of the target domain, rather than the sample fusion data of the target domain and the source domain. In this way, compared with the ranking model trained with the fusion sample data, the disclosed solution can effectively avoid the problems such as the uneven distribution of samples in the target domain and the source domain, which leads to the weakening of certain interest features and the difficulty in stably exposing external domain interest features, thereby ensuring that the external domain interest features can be learned losslessly.

[0123] Step S405: Based on the sorted M1 first candidate multimedia resources and the sorted M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained.

[0124] In a specific example, M3 target multimedia resources may be obtained in the following manner, that is, the above-mentioned step S304 specifically includes:

[0125] Selecting the first candidate multimedia resource ranked first from the sorted M1 first candidate multimedia resources; for example, sorting them from largest to smallest based on click probability. In this case, the first candidate multimedia resource ranked first is the content with the highest click probability;

[0126] The first candidate multimedia resource ranked first is inserted into the sorted M2 second candidate multimedia resources to obtain sorted M3 target multimedia resources. At this time, the multimedia resources to be exposed to the target object include content recommended based on external domain interest features.

[0127] Alternatively, the above-mentioned step S304 specifically includes:

[0128] Selecting the first candidate multimedia resource ranked first from the sorted M1 first candidate multimedia resources, for example, sorting based on click probability from large to small. In this case, the first candidate multimedia resource ranked first is the content with the highest click probability;

[0129] Randomly remove one from the sorted M2 second candidate multimedia resources, and insert the first candidate multimedia resource ranked first into the position of the removed second candidate multimedia resource to obtain sorted M3 target multimedia resources. At this time, the multimedia resources to be exposed to the target object include content recommended based on external domain interest features.

[0130] In this way, the disclosed scheme uses the external domain ranking model obtained by training based on the sample data of the source domain to rank the candidate multimedia resources (i.e., M1 first candidate multimedia resources) recalled based on the target external domain interest features. At the same time, the local domain ranking model obtained by training based on the sample data of the target domain is used to rank the candidate multimedia resources (i.e., M2 second candidate multimedia resources) recalled based on the target local domain interest features. The ranking results are then fused to obtain the final ranked M3 target multimedia resources to be exposed to the target object. In this way, the problems such as the uneven distribution of samples in the target domain and the source domain, which leads to the weakening of certain interest features and the difficulty in stably exposing external domain interest features, are effectively avoided, thereby ensuring that the external domain interest features can be learned losslessly, and further effectively avoiding the problem of the recommended content being too single or even forming a cocoon effect, further laying the foundation for effectively improving the user experience.

[0131] The following is combined with Figure 5 The disclosed solution is further described in detail. Specifically, Figure 5As shown, the disclosed solution provides a practical and complete cross-domain interest exploration solution from the aspects of cross-domain interest mining, resource matching, resource sorting, resource exposure, etc. This solution can decouple various processing logics, thereby improving reusability and scalability. Therefore, the disclosed solution can be applied in different scenarios, expanding the scope of application.

[0132] It should be noted that, based on the correlation retention attribution, secondary category entropy is highly correlated with target object retention. The more secondary categories a target object displays and the more balanced the category distribution, the higher the user retention. Based on this, cross-domain interest mining can also be conducted from the perspective of secondary categories.

[0133] 1. Cross-domain interest mining

[0134] 1. Mining external trust interests

[0135] Explicit interest mining, for example, obtains relevant features of sample data in the source domain based on causal inference (Double Machine Learning, DML) analysis, identifies positive samples, and performs confidence interest mining based on positive sample unlabeled learning (PU Learning) to obtain explicit interest features, such as the initial external domain explicit interest features mentioned above.

[0136] Implicit interest mining, for example, based on the elastic orthogonal network, learns the multiple interest expressions of the target object in the source domain to obtain implicit interest features, such as the initial external domain implicit interest features mentioned above.

[0137] (2) Interest Development

[0138] Specific expansion methods include:

[0139] Extension method 1: Interest expansion based on mutual information (that is, the mutual information parameter mentioned above)

[0140] Mutual information is used to characterize the closeness between two interest features. In a specific example, the mutual information between two interest features can be obtained based on the following formula, and the target object's confidence interest is used as a judgment criterion (for example, based on the target object's confidence interest, the first preset requirement is determined), so as to select the expanded interest feature that meets the first preset requirement. In this way, it can be effectively ensured that the new interest explored will not deviate too much from the target object's existing interest.

[0141] Here, the specific expression of mutual information is:

[0142]

[0143] here, represents the probability of the occurrence of the feature of interest x; represents the probability of the occurrence of the feature of interest y; It represents the probability that interest feature x and interest feature y appear at the same time.

[0144] Furthermore, in the process of obtaining the expanded external domain interest features described above, x can be specifically the initial external domain explicit interest features; correspondingly, y can be specifically the interest features to be recommended. In this way, based on the above formula, the expanded external domain interest features whose mutual information meets the first preset requirement can be selected.

[0145] Expansion method 2: Expansion based on the generalization of interests among the population

[0146] Select the target group corresponding to the target object, for example, select the target group indicated by the high target group index (TGI) corresponding to the target object, and use the interest characteristics of the target group as the expanded interest characteristics, for example, as the expanded external domain interest characteristics described above.

[0147] It is understandable that, in a specific example, the above-mentioned solution can also be used to mine and obtain the domain interest features of the target domain.

[0148] 2. Resource Matching

[0149] Here, it should be pointed out that the resource matching process can specifically correspond to the recall process described above. For example, the following method can be used to recall the M1 first candidate multimedia resources described above; similarly, the following method can be used to recall the M2 second candidate multimedia resources described above; further, the methods used to recall the M1 first candidate multimedia resources and the M2 second candidate multimedia resources can be the same or different, and the present disclosure does not impose any restrictions on this.

[0150] For example, recall is performed based on the following methods:

[0151] Key Opinion Leader (KOL) - User-based Collaborative Filtering (UCF) propagation: Based on the target object's posterior performance and propagation effect, KOL users in each first-level category are mined and used as a candidate set. Similar KOLs to the target object are obtained online, and multimedia data is voted on. Based on the voting results, multimedia data that satisfies the target object is recalled.

[0152] Explicit term recall: Recalling multimedia data by explicitly matching the target object based on its known confidence.

[0153] ICF (item-based collaboration filter) recall: uses the historical click records of the target object as the trigger key to recall similar multimedia data.

[0154] Implicit interest recall: Recall similar multimedia data based on the mined implicit interest features (such as vector representation).

[0155] It is understandable that the recall method described above is only an example, and other recall methods can be used in actual applications, and the present disclosure does not limit this.

[0156] 3. Resource Sorting

[0157] For example, a Deep Structured Semantic Model (DSSM) is trained based on sample data from the external domain. The labels can be specifically click probabilities, and the model obtained after training is used as the external domain ranking model. Here, the features used for training can be aligned with the features used by the ranking model in the training target domain.

[0158] It should be noted that the training method of the local domain ranking model can be the same as the training method of the external domain ranking model, except that the sample data of the two are different.

[0159] 4. Resource Exposure

[0160] Exposure timing judgment, for example, can calculate the fatigue coefficient of interest characteristics based on Beta distribution and expose with a certain probability.

[0161] Exit judgment, for example, can be based on multiple dimensions such as the target object's posterior results and time. Here, the multi-dimensional exit mechanism can effectively reduce the reverse order of results.

[0162] For example, interest selection can be performed by mining underdeveloped interests in the target domain based on relative entropy, and after excluding underdeveloped interests, Thompson sampling is used to select the target interest features used for the final exposure.

[0163] Resource selection: The multimedia data is sorted using the labels output by the aforementioned ranking models (such as the external domain ranking model or the local domain ranking model). The top-ranked multimedia resource is then selected as the final exposed multimedia resource.

[0164] Position selection: The exploration slot is guaranteed by the pit setting method, and the final location of the exploration resource exposure is determined by the a posteriori results. This makes it easier to expose the exploration resources in a pit setting manner.

[0165] In this way, the disclosed solution ensures the relevance of the explored interest features to the target objects by mining external domain confidence interests, thereby solving the problem of incomparable results caused by inconsistent distribution of local and external domain interests; at the same time, by constructing an independent external domain ranking model, it avoids the problem of weakening local domain interests after the fusion of external domain samples and local domain samples, and the independently constructed external domain ranking model is conducive to model iteration, and at the same time, it can effectively ensure that external domain interests are stably exposed.

[0166] The disclosed solution also provides a multimedia recommendation device, such as Figure 6 Shown, including:

[0167] The first resource determination unit 601 is configured to obtain M1 first candidate multimedia resources, wherein the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are based on the behavioral characteristics of the target object in the source domain; N1 and M1 are both positive integers greater than or equal to 1;

[0168] The second resource determination unit 602 is configured to obtain M2 second candidate multimedia resources, wherein the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; and N2 and M2 are both positive integers greater than or equal to 1;

[0169] The recommendation unit 603 is configured to obtain M3 target multimedia resources to be recommended to the target object based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources; M3 is a positive integer greater than or equal to 1.

[0170] In a specific example of the disclosed solution, the first resource determining unit is further configured to:

[0171] Acquire at least one initial external domain explicit interest feature of the target object in the source domain;

[0172] Acquire at least one extended external domain interest feature of the target object in the source domain;

[0173] The N1 target external-domain interest features are obtained based at least on the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature.

[0174] In a specific example of the disclosed solution, the first resource determination unit is specifically configured to do at least one of the following:

[0175] Determining a mutual information parameter between an initial external-domain explicit interest feature and the interest feature to be recommended in the at least one initial external-domain explicit interest feature, and using the interest feature to be recommended whose mutual information parameter meets a first preset requirement as an expanded external-domain interest feature; wherein the mutual information parameter represents the interest correlation between the initial external-domain explicit interest feature and the interest feature to be recommended;

[0176] The interest features of the target group corresponding to the target object in the source domain are used as the expanded external domain interest features.

[0177] In a specific example of the disclosed solution, the first resource determining unit is further configured to:

[0178] Acquire at least one initial external domain implicit interest feature of the target object in the source domain;

[0179] The N1 target external-domain interest features are obtained based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature.

[0180] In a specific example of the disclosed solution, the first resource determining unit is specifically configured to:

[0181] At least based on the activity levels of each element in a first initial external interest set, at least some of the elements are deleted from the first initial external interest set to obtain a target initial external interest set; wherein the first initial external interest set includes at least one initial external explicit interest feature, at least one expanded external interest feature, and at least one initial external implicit interest feature;

[0182] The N1 target external-domain interest features are obtained by sampling from the target initial external-domain interest set using a preset sampling rule.

[0183] In a specific example of the disclosed solution, the first resource determining unit is specifically configured to:

[0184] A target initial external-domain interest set is obtained by deleting at least some elements from the first initial external-domain interest set based on the activity level of each element in the first initial external-domain interest set and based on at least one of the following: a posterior result of the element, and an exit time of the element.

[0185] In a specific example of the disclosed solution, the first resource determination unit is specifically configured to recall M1 first candidate multimedia resources from a multimedia resource library in a source domain based on the N1 target external domain interest features; input the N1 target external domain interest features and the M1 first candidate multimedia resources into an external domain ranking model to obtain the sorted M1 first candidate multimedia resources; wherein the external domain ranking model is obtained by training a first preset ranking model based on first sample data in the source domain; the first sample data at least indicates an association relationship between multimedia resources and the interest features of objects in the source domain;

[0186] The recommendation unit is specifically configured to obtain M3 target multimedia resources to be recommended to the target object based on the sorted M1 first candidate multimedia resources and the M2 second candidate multimedia resources.

[0187] In a specific example of the disclosed solution, the second resource determination unit is specifically configured to recall M2 second candidate multimedia resources from a multimedia resource library of a target domain based on the N2 target local domain interest features; input the N2 target local domain interest features and the M2 second candidate multimedia resources into a local domain ranking model to obtain the sorted M2 second candidate multimedia resources; wherein the local domain ranking model is obtained by training a second preset ranking model based on second sample data of the target domain; and the second sample data at least includes an association relationship between multimedia resources and the interest features of objects in the target domain;

[0188] The recommendation unit is specifically configured to obtain M3 target multimedia resources to be recommended to the target object based on the sorted M1 first candidate multimedia resources and the sorted M2 second candidate multimedia resources.

[0189] In a specific example of the disclosed solution, the recommendation unit is further configured to:

[0190] Based on the a posteriori results of the M3 target multimedia resources, the N1 target external-domain interest features and / or the N2 target local-domain interest features are updated.

[0191] For the description of specific functions and examples of each unit of the device in the embodiment of the present disclosure, please refer to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0192] In the technical solutions disclosed herein, the acquisition, storage, and application of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0193] 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.

[0194] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, 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.

[0195] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform 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. RAM 703 may also store various programs and data required for the operation of device 700. Computing unit 701, ROM 702, and RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to bus 704.

[0196] Various components in device 700 are connected to I / O interface 705, including an input unit 706, such as a keyboard, mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, optical disk, etc.; and a communication unit 709, such as a network card, modem, wireless communication transceiver, etc. The 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.

[0197] The computing unit 701 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as the multimedia recommendation method. For example, in some embodiments, the multimedia recommendation method may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the multimedia recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the multimedia recommendation method in any other suitable manner (e.g., via firmware).

[0198] Various embodiments of the systems and techniques described 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), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system comprising 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.

[0199] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0200] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0201] 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).

[0202] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0203] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0204] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0205] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A multimedia recommendation method, comprising: Obtain M1 first candidate multimedia resources, wherein the M1 first candidate multimedia resources are multimedia resources to be recommended to the target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are based on the behavioral characteristics of the target object in the source domain; N1 and M1 are both positive integers greater than or equal to 1; Obtain M2 second candidate multimedia resources, wherein the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; and N2 and M2 are both positive integers greater than or equal to 1; Based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained; M3 is a positive integer greater than or equal to 1.

2. The method according to claim 1, further comprising: Acquire at least one initial external domain explicit interest feature of the target object in the source domain; Acquire at least one extended external domain interest feature of the target object in the source domain; The N1 target external-domain interest features are obtained based at least on the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature.

3. The method according to claim 2, wherein: The acquiring of at least one extended external domain interest feature of the target object in the source domain includes at least one of the following: Determining a mutual information parameter between an initial external-domain explicit interest feature and the interest feature to be recommended in the at least one initial external-domain explicit interest feature, and using the interest feature to be recommended whose mutual information parameter meets a first preset requirement as an expanded external-domain interest feature; wherein the mutual information parameter represents the interest correlation between the initial external-domain explicit interest feature and the interest feature to be recommended; The interest features of the target group corresponding to the target object in the source domain are used as the expanded external domain interest features.

4. The method according to claim 2 or 3, further comprising: Acquire at least one initial external domain implicit interest feature of the target object in the source domain; The obtaining of the N1 target external-domain interest features based at least on the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature includes: The N1 target external-domain interest features are obtained based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature.

5. The method according to claim 4, wherein The obtaining the N1 target external-domain interest features based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature includes: At least based on the activity levels of each element in a first initial external interest set, at least some of the elements are deleted from the first initial external interest set to obtain a target initial external interest set; wherein the first initial external interest set includes at least one initial external explicit interest feature, at least one expanded external interest feature, and at least one initial external implicit interest feature; The N1 target external-domain interest features are obtained by sampling from the target initial external-domain interest set using a preset sampling rule.

6. The method according to claim 5, wherein: The step of deleting at least some elements from the first initial external-domain interest set based at least on the activity levels of the elements in the first initial external-domain interest set to obtain a target initial external-domain interest set includes: A target initial external-domain interest set is obtained by deleting at least some elements from the first initial external-domain interest set based on the activity level of each element in the first initial external-domain interest set and based on at least one of the following: a posterior result of the element, and an exit time of the element.

7. The method according to any one of claims 1 to 3, wherein: The acquiring M1 first candidate multimedia resources includes: Based on the N1 target external domain interest features, recall M1 first candidate multimedia resources from the multimedia resource library of the source domain; Inputting the N1 target external domain interest features and the M1 first candidate multimedia resources into an external domain ranking model to obtain the ranked M1 first candidate multimedia resources; wherein the external domain ranking model is obtained by training a first preset ranking model based on first sample data of the source domain; the first sample data at least indicates an association relationship between the multimedia resource and the object's interest feature in the source domain; The step of obtaining M3 target multimedia resources to be recommended to the target object based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources includes: Based on the sorted M1 first candidate multimedia resources and the M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained.

8. The method according to claim 7, wherein: The acquiring M2 second candidate multimedia resources includes: Recalling M2 second candidate multimedia resources from the multimedia resource library of the target domain based on the N2 target domain interest features; Inputting the N2 target domain interest features and the M2 second candidate multimedia resources into a local domain ranking model to obtain ranked M2 second candidate multimedia resources; wherein the local domain ranking model is obtained by training a second preset ranking model based on second sample data of the target domain; the second sample data at least includes an association relationship between the multimedia resources and the object's interest features in the target domain; Wherein, based on the sorted M1 first candidate multimedia resources and M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained, including: Based on the sorted M1 first candidate multimedia resources and the sorted M2 second candidate multimedia resources, M3 target multimedia resources to be recommended to the target object are obtained.

9. The method according to any one of claims 1 to 3, further comprising: Based on the a posteriori results of the M3 target multimedia resources, the N1 target external-domain interest features and / or the N2 target local-domain interest features are updated.

10. A multimedia recommendation device, comprising: A first resource determination unit is configured to obtain M1 first candidate multimedia resources, wherein the M1 first candidate multimedia resources are multimedia resources to be recommended to a target object based on N1 target external domain interest features; the target external domain interest features in the N1 target external domain interest features are based on behavioral features of the target object in a source domain; and N1 and M1 are both positive integers greater than or equal to 1; A second resource determination unit is configured to obtain M2 second candidate multimedia resources, wherein the M2 second candidate multimedia resources are multimedia resources to be recommended to the target object based on N2 target domain interest features; the target domain interest features in the N2 target domain interest features are based on the behavioral characteristics of the target object in the target domain; and N2 and M2 are both positive integers greater than or equal to 1; The recommendation unit is configured to obtain M3 target multimedia resources to be recommended to the target object based on the M1 first candidate multimedia resources and the M2 second candidate multimedia resources; M3 is a positive integer greater than or equal to 1.

11. The device according to claim 10, wherein The first resource determining unit is further configured to: Acquire at least one initial external domain explicit interest feature of the target object in the source domain; Acquire at least one extended external domain interest feature of the target object in the source domain; The N1 target external-domain interest features are obtained based at least on the at least one initial external-domain explicit interest feature and the at least one expanded external-domain interest feature.

12. The device according to claim 11, wherein The first resource determination unit is specifically configured to do at least one of the following: Determining a mutual information parameter between an initial external-domain explicit interest feature and the interest feature to be recommended in the at least one initial external-domain explicit interest feature, and using the interest feature to be recommended whose mutual information parameter meets a first preset requirement as an expanded external-domain interest feature; wherein the mutual information parameter represents the interest correlation between the initial external-domain explicit interest feature and the interest feature to be recommended; The interest features of the target group corresponding to the target object in the source domain are used as the expanded external domain interest features.

13. The device according to claim 11 or 12, wherein: The first resource determining unit is further configured to: Acquire at least one initial external domain implicit interest feature of the target object in the source domain; The N1 target external-domain interest features are obtained based on the at least one initial external-domain explicit interest feature, the at least one expanded external-domain interest feature, and the at least one initial external-domain implicit interest feature.

14. The device according to claim 13, wherein The first resource determination unit is specifically configured to: At least based on the activity levels of each element in a first initial external interest set, at least some of the elements are deleted from the first initial external interest set to obtain a target initial external interest set; wherein the first initial external interest set includes at least one initial external explicit interest feature, at least one expanded external interest feature, and at least one initial external implicit interest feature; The N1 target external-domain interest features are obtained by sampling from the target initial external-domain interest set using a preset sampling rule.

15. The device according to claim 14, wherein The first resource determination unit is specifically configured to: A target initial external-domain interest set is obtained by deleting at least some elements from the first initial external-domain interest set based on the activity level of each element in the first initial external-domain interest set and based on at least one of the following: a posterior result of the element, and an exit time of the element.

16. The device according to any one of claims 10 to 12, wherein: The first resource determination unit is specifically configured to recall M1 first candidate multimedia resources from a multimedia resource library in a source domain based on the N1 target external domain interest features; input the N1 target external domain interest features and the M1 first candidate multimedia resources into an external domain ranking model to obtain the sorted M1 first candidate multimedia resources; wherein the external domain ranking model is obtained by training a first preset ranking model based on first sample data in the source domain; the first sample data at least indicates an association relationship between multimedia resources and the interest features of objects in the source domain; The recommendation unit is specifically configured to obtain M3 target multimedia resources to be recommended to the target object based on the sorted M1 first candidate multimedia resources and the M2 second candidate multimedia resources.

17. The device according to claim 16, wherein The second resource determination unit is specifically configured to recall M2 second candidate multimedia resources from the multimedia resource library of the target domain based on the N2 target domain interest features; input the N2 target domain interest features and the M2 second candidate multimedia resources into a local domain ranking model to obtain the sorted M2 second candidate multimedia resources; wherein the local domain ranking model is obtained by training a second preset ranking model based on second sample data of the target domain; the second sample data at least includes an association relationship between the multimedia resources and the interest features of the object in the target domain; The recommendation unit is specifically configured to obtain M3 target multimedia resources to be recommended to the target object based on the sorted M1 first candidate multimedia resources and the sorted M2 second candidate multimedia resources.

18. The device according to any one of claims 10 to 12, wherein: The recommendation unit is further configured to: Based on the a posteriori results of the M3 target multimedia resources, the N1 target external-domain interest features and / or the N2 target local-domain interest features are updated.

19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

20. 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 9.

21. 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 9.

Images