Model training method, object recommendation method, device, equipment, medium and product

Abstract

The embodiment of the invention discloses a model training method and device, an object recommendation method and device, equipment, a medium and a product. A specific embodiment of the method comprises the steps of obtaining a target training sample; inputting the user behavior feature information into an initial user interest information generation model to generate initial user interest information; generating initial interest intensity information by utilizing an initial interest intensity information generation model; generating initial object click information; and according to the initial interest intensity information, the initial object click information and the object click tag, performing model training on the initial user interest information generation model and the initial interest intensity information generation model to generate the user interest information generation model and the initial interest intensity information generation model. The embodiment is related to artificial intelligence, the user interest information generation model and the interest intensity information generation model can be utilized, and object recommendation can be more accurately achieved on the basis of the generated interest intensity information.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a model training method, an object recommendation method, an apparatus, a device, a medium, and a product. Background Art

[0002] At present, with the continuous development of artificial intelligence, the application of personalized recommendation based on artificial intelligence is becoming more and more widespread. Personalized recommendation can recommend objects that users are interested in to improve the user experience. For user recommendations, the usual method is to achieve user object recommendation through a deep recall model.

[0003] However, the inventors have found that when the above method is used to recommend items to users, the following technical problems often occur:

[0004] There is no deeper consideration of at least one user interest point corresponding to the user, resulting in the problem that the recommended objects are not accurate enough when recommending objects based on at least one user interest point.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0006] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0007] Some embodiments of the present disclosure propose model training methods, object recommendation methods, devices, equipment, media and products to solve the technical problems mentioned in the above background technology section.

[0008] In a first aspect, some embodiments of the present disclosure provide a model training method, comprising: obtaining a target training sample, wherein the target training sample comprises: recommended object feature information, user behavior feature information and object click labels; inputting the user behavior feature information into an initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is the user's interest feature information under at least one user interest point; using an initial interest intensity information generation model to generate initial interest intensity information corresponding to the at least one user interest point; generating initial object click information based on the recommended object feature information and the initial user interest information; and performing model training on the initial user interest information generation model and the initial interest intensity information generation model based on the initial interest intensity information, the initial object click information and the object click labels to generate a user interest information generation model and an interest intensity information generation model.

[0009] Optionally, the above-mentioned generating initial object click information according to the above-mentioned recommended object feature information and the above-mentioned initial user interest information includes: inputting the above-mentioned recommended object feature information into the initial recommended object feature encoding model to generate object feature encoding information; generating the above-mentioned initial object click information according to the above-mentioned object feature encoding information and the above-mentioned initial user interest information; and the above-mentioned model training of the above-mentioned initial user interest information generation model and the above-mentioned initial interest intensity information generation model according to the above-mentioned initial object click information and the above-mentioned object click label to generate a user interest information generation model and an interest intensity information generation model, including: model training of the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model according to the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0010] Optionally, the above-mentioned initial user interest information generation model is used to vectorize each behavior feature information in the above-mentioned at least one behavior feature information sequence to generate a behavior vector and obtain at least one behavior vector sequence; for each behavior type in the above-mentioned at least one behavior type, the above-mentioned initial user interest information generation model is used to generate a key behavior vector corresponding to the above-mentioned behavior type; and the above-mentioned initial user interest information is generated based on the at least one key behavior vector obtained.

[0011] Optionally, the above-mentioned generation of the key behavior vector corresponding to the above-mentioned behavior type includes: in response to determining that the above-mentioned behavior type is a click behavior type, clustering the behavior vector sequence corresponding to the above-mentioned behavior type to generate a behavior vector cluster set; generating the key behavior vector corresponding to the above-mentioned click behavior type according to the behavior vector cluster center set in the above-mentioned behavior vector cluster set; in response to determining that the above-mentioned behavior type is a value-added behavior type, determining at least one first behavior vector whose time interval with the target time satisfies a preset time interval condition; combining the above-mentioned at least one first behavior vector to generate the key behavior vector corresponding to the above-mentioned value-added behavior type; in response to determining that the above-mentioned behavior type is an item inflow type, determining at least one second behavior vector whose corresponding item inflow times satisfies a preset time condition; and combining the above-mentioned at least one second behavior vector to generate the key behavior vector corresponding to the item inflow type.

[0012] Optionally, the user behavior feature information also includes: a behavior content feature information sequence for the at least one behavior type; and generating the initial user interest information based on the at least one key behavior vector obtained, including: encoding the behavior content feature information in the behavior content feature information sequence to generate a behavior content vector sequence; performing vector fusion on the at least one key behavior vector to generate a key behavior matrix for the at least one behavior type; inputting the key behavior matrix and the behavior content vector sequence into the initial attention mechanism model to generate a similarity matrix, wherein the similarity matrix represents the strength of association between the key behavior and the behavior content. The initial user interest information generation model includes: an initial attention mechanism model; generating the initial user interest information based on the similarity matrix.

[0013] Optionally, the user behavior feature information also includes: a first behavior category feature information sequence for at least one of the behavior types; and generating the initial user interest information based on the similarity matrix, including: encoding the first behavior category feature information in the first behavior category feature information sequence to generate a behavior category vector sequence; masking the elements in the similarity matrix to generate a mask matrix; generating interest point derived content information based on the mask matrix and the behavior content vector sequence; generating interest point derived category information by combining the mask matrix and the behavior category vector sequence; generating the initial user interest information based on the interest point derived content information and the interest point derived category information.

[0014] Optionally, the above-mentioned initial user interest information is generated based on the above-mentioned interest point derived content information and the above-mentioned interest point derived category information, including: inputting the above-mentioned interest point derived content information and the above-mentioned interest point derived category information into an initial interest information extraction model to generate the above-mentioned initial user interest information, wherein the above-mentioned initial user interest information generation model includes: the above-mentioned initial interest information extraction model.

[0015] Optionally, the above-mentioned initial interest intensity information generation model is used to generate the initial interest intensity information corresponding to the above-mentioned at least one user interest point, including: generating a second behavior category feature information sequence for the above-mentioned key behavior matrix; inputting the above-mentioned second behavior category feature information sequence into the above-mentioned initial interest intensity information generation model to generate the above-mentioned initial interest intensity information.

[0016] Optionally, the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label are model-trained to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model, including: generating an initial interest intensity label according to the above-mentioned recommended object feature information and the above-mentioned initial user interest information; generating interest intensity loss information for the above-mentioned initial interest intensity label and the above-mentioned initial interest intensity information; generating click loss information for the above-mentioned initial object click information and the above-mentioned object click label; model-training the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model according to the above-mentioned interest intensity loss information and the above-mentioned click loss information to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0017] In a second aspect, some embodiments of the present disclosure provide a model training device, comprising: a first acquisition unit, configured to acquire target training samples, wherein the target training samples include: recommended object feature information, user behavior feature information and object click labels; a first generation unit, configured to input the user behavior feature information into an initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is the user's interest feature information under at least one user interest point; a second generation unit, configured to generate initial interest strength information corresponding to the at least one user interest point using the initial interest strength information generation model; a third generation unit, configured to generate initial object click information based on the recommended object feature information and the initial user interest information; a training unit, configured to perform model training on the initial user interest information generation model and the initial interest strength information generation model based on the initial interest strength information, the initial object click information and the object click label to generate a user interest information generation model and an interest strength information generation model.

[0018] Optionally, the third generating unit may be configured to: input the recommended object feature information into an initial recommended object feature encoding model to generate object feature encoding information; and generate the initial object click information based on the object feature encoding information and the initial user interest information.

[0019] Optionally, the training unit can be configured to: perform model training on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model according to the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label, so as to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0020] Optionally, the above-mentioned user behavior characteristic information includes: at least one behavior characteristic information sequence corresponding to at least one behavior type; and the first generation unit can be configured to: use the above-mentioned initial user interest information generation model to vectorize each behavior characteristic information in the above-mentioned at least one behavior characteristic information sequence to generate a behavior vector and obtain at least one behavior vector sequence; for each behavior type in the above-mentioned at least one behavior type, use the above-mentioned initial user interest information generation model to generate a key behavior vector corresponding to the above-mentioned behavior type.

[0021] Optionally, the first generating unit may be configured to: in response to determining that the above-mentioned behavior type is a click behavior type, cluster the behavior vector sequence corresponding to the above-mentioned behavior type to generate a behavior vector cluster set; generate a key behavior vector corresponding to the above-mentioned click behavior type based on the behavior vector cluster center set in the above-mentioned behavior vector cluster set; in response to determining that the above-mentioned behavior type is a value-added behavior type, determine at least one first behavior vector whose time interval with the target time satisfies a preset time interval condition; combine the above-mentioned at least one first behavior vector to generate a key behavior vector corresponding to the above-mentioned value-added behavior type; in response to determining that the above-mentioned behavior type is an item inflow type, determine at least one second behavior vector whose corresponding item inflow times satisfies a preset time condition; and combine the above-mentioned at least one second behavior vector to generate a key behavior vector corresponding to the item inflow type.

[0022] Optionally, the above-mentioned user behavior feature information also includes: a behavior content feature information sequence for the above-mentioned at least one behavior type; and the first generation unit can be configured to: encode the behavior content feature information in the above-mentioned behavior content feature information sequence to generate a behavior content vector sequence; perform vector fusion on the above-mentioned at least one key behavior vector to generate a key behavior matrix for the above-mentioned at least one behavior type; input the above-mentioned key behavior matrix and the above-mentioned behavior content vector sequence into the initial attention mechanism model to generate a similarity matrix, wherein the above-mentioned similarity matrix represents the correlation strength between the key behavior and the behavior content, and the above-mentioned initial user interest information generation model includes: an initial attention mechanism model; based on the above-mentioned similarity matrix, generate the above-mentioned initial user interest information.

[0023] Optionally, the above-mentioned user behavior feature information also includes: a first behavior category feature information sequence for the above-mentioned at least one behavior type; and the first generation unit can be configured to: encode the first behavior category feature information in the above-mentioned first behavior category feature information sequence to generate a behavior category vector sequence; mask the elements in the above-mentioned similarity matrix to generate a mask matrix; generate interest point derived content information based on the above-mentioned mask matrix and the above-mentioned behavior content vector sequence; generate interest point derived category information by combining the above-mentioned mask matrix and the above-mentioned behavior category vector sequence; generate the above-mentioned initial user interest information based on the above-mentioned interest point derived content information and the above-mentioned interest point derived category information.

[0024] Optionally, the first generating unit may be configured to: input the above-mentioned interest point derived content information and the above-mentioned interest point derived category information into the initial interest information extraction model to generate the above-mentioned initial user interest information, wherein the above-mentioned initial user interest information generation model includes: the above-mentioned initial interest information extraction model.

[0025] Optionally, the second generating unit may be configured to: generate a second behavior category feature information sequence for the key behavior matrix; and input the second behavior category feature information sequence into the initial interest intensity information generating model to generate the initial interest intensity information.

[0026] Optionally, the training unit can be configured to: generate an initial interest intensity label based on the above-mentioned recommended object feature information and the above-mentioned initial user interest information; generate interest intensity loss information for the above-mentioned initial interest intensity label and the above-mentioned initial interest intensity information; generate click loss information for the above-mentioned initial object click information and the above-mentioned object click label; and perform model training on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model based on the above-mentioned interest intensity loss information and the above-mentioned click loss information to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0027] In a third aspect, some embodiments of the present disclosure provide an object recommendation method, comprising: obtaining user feature information corresponding to a target user as target user feature information; generating user interest information for the target user using a pre-trained user interest information generation model based on the target user feature information, wherein the user interest information generation model is generated based on a model training method; generating an initial recommended object information set for at least one user interest point based on the user interest information; generating interest intensity information for the at least one user interest point using a pre-trained interest intensity information generation model, wherein the interest intensity information generation model is generated based on a model training method; and filtering out a recommended object information set from the initial recommended object information set based on the interest intensity information to perform object recommendation for the recommended object information set.

[0028] Optionally, the above-mentioned filtering out a recommended object information set from the above-mentioned initial recommended object information set based on the above-mentioned interest strength information includes: for each user interest point in at least one user interest point, executing the following first generation step: determining the initial recommended object information corresponding to the above-mentioned user interest point in the above-mentioned initial recommended object information set, and obtaining at least one initial recommended object information; determining the number of recommended objects corresponding to the above-mentioned user interest point based on the above-mentioned interest strength information; filtering out the initial recommended object information whose corresponding recommendation similarity meets the preset similarity condition from the above-mentioned at least one initial recommended object information as the recommended object information, and obtaining at least one recommended object information corresponding to the above-mentioned user interest point, wherein the number of information corresponding to the above-mentioned at least one recommended object information is the same as the above-mentioned number of recommended objects.

[0029] Optionally, the above method also includes: for each user interest point among the above at least one user interest point, executing the following second generation step: determining the recall result corresponding to the above user interest point and the corresponding interest intensity value in the above interest intensity information; based on the above recall result and the above interest intensity value, determining the click rate corresponding to the above user interest point.

[0030] In a fourth aspect, some embodiments of the present disclosure provide an object recommendation device, comprising: a second acquisition unit, configured to acquire user feature information corresponding to a target user as target user feature information; a fourth generation unit, configured to generate user interest information for the target user based on the target user feature information and using a pre-trained user interest information generation model, wherein the user interest information generation model is generated based on a model training method; a fifth generation unit, configured to generate an initial recommended object information set for at least one user interest point based on the user interest information; a sixth generation unit, configured to generate interest intensity information for the at least one user interest point based on a pre-trained interest intensity information generation model, wherein the interest intensity information generation model is generated based on a model training method; a recommendation unit, configured to filter out a recommended object information set from the initial recommended object information set based on the interest intensity information, so as to perform object recommendation for the recommended object information set.

[0031] Optionally, the recommendation unit can be configured to: for each user interest point in at least one user interest point, perform the following first generation step: determine the initial recommended object information corresponding to the above-mentioned user interest point in the above-mentioned initial recommended object information set, and obtain at least one initial recommended object information; determine the number of recommended objects corresponding to the above-mentioned user interest point based on the above-mentioned interest intensity information; filter out the initial recommended object information whose corresponding recommendation similarity meets the preset similarity condition from the above-mentioned at least one initial recommended object information as the recommended object information, and obtain at least one recommended object information corresponding to the above-mentioned user interest point, wherein the number of information corresponding to the above-mentioned at least one recommended object information is the same as the above-mentioned number of recommended objects.

[0032] Optionally, the device also includes: for each user interest point among the at least one user interest point mentioned above, executing the following second generation step: determining the recall result corresponding to the above user interest point and the corresponding interest intensity value in the above interest intensity information; and determining the click rate corresponding to the above user interest point based on the above recall result and the above interest intensity value.

[0033] In a fifth aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any one of the implementation methods in the first and third aspects.

[0034] In a sixth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any one of the implementation modes of the first aspect and the third aspect is implemented.

[0035] In a seventh aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which, when executed by a processor, implements the method described in any one of the implementation modes of the first and third aspects above.

[0036] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model training method of some embodiments of the present disclosure, the user interest information generation model and the interest intensity information generation model can be used, and based on the generated interest intensity information, the object recommendation can be more accurately implemented. Specifically, the reason for the inaccuracy of the relevant object recommendation is that: there is no deeper consideration of at least one user interest point corresponding to the user, resulting in the object recommendation based on at least one user interest point, and there is a problem that the recommended object is not accurate enough. Based on this, the model training method of some embodiments of the present disclosure first obtains the target training sample, wherein the above-mentioned target training sample includes: recommended object feature information, user behavior feature information and object click label. Here, the obtained target training sample is used for model training. In addition, the recommended object feature information can characterize the feature information of the recommended object, so as to facilitate the subsequent generation of initial object click information. The user behavior feature information can fully reflect the feature information of the user in each behavior, so as to be used for the subsequent generation of the user's interest feature information under at least one user interest point. The object click label is used as the real data of the object click for the subsequent model training. Then, the user behavior feature information is input into the initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is the interest feature information of the user under at least one user interest point. Here, by using the initial user interest information generation model, the interest feature content of the user under at least one user interest point can be fully grasped, so as to facilitate the subsequent model training. Then, the initial interest strength information generation model is used to generate the initial interest strength information corresponding to the at least one user interest point. Here, the initial interest strength information can characterize the user's interest in at least one user interest point. In the model training process, by generating the initial interest strength information, the importance of the user under at least one user interest point can be fully considered, so that in the subsequent object recommendation, the adaptive object recommendation adjustment can be made according to the interest strength, so that the recommended items are more accurate and reasonable. Furthermore, according to the recommended object feature information and the initial user interest information, the initial object click information is generated as the predicted object click information, and the corresponding click loss is formed with the object click label, which is convenient for the subsequent model training. Finally, based on the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label, the above-mentioned initial user interest information generation model and the above-mentioned initial interest intensity information generation model are trained to generate a user interest information generation model and an interest intensity information generation model.In summary, during the model training process, by considering the user's interest feature information and the corresponding interest intensity information at at least one user interest point, the user interest information generation model and the interest intensity information generation model can be subsequently used to more accurately achieve object recommendation based on the generated interest intensity information. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0038] Figure 1 is a schematic diagram of an application scenario of a model training method according to some embodiments of the present disclosure;

[0039] Figure 2 is a flowchart of some embodiments of the model training method according to the present disclosure;

[0040] Figure 3 is a flowchart of other embodiments of the model training method according to the present disclosure;

[0041] Figure 4 is a flow chart of some embodiments of the object recommendation method according to the present disclosure;

[0042] Figure 5 is a schematic diagram of the structure of some embodiments of the model training device according to the present disclosure;

[0043] Figure 6 is a schematic diagram of the structure of some embodiments of the object recommendation device according to the present disclosure;

[0044] Figure 7 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0045] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0046] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0047] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0048] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0049] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0050] With regard to the collection, storage, and use of user personal information (for example, user behavior characteristic information) involved in this disclosure, before performing the corresponding operations, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, fulfilling the obligation to inform the personal information subject, and obtaining the authorization and consent of the personal information subject in advance.

[0051] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0052] Figure 1 It is a schematic diagram of an application scenario of the model training method according to some embodiments of the present disclosure.

[0053] exist Figure 1In the application scenario, first, the electronic device 101 can obtain the target training sample 102. The target training sample 102 includes: recommended object feature information 103, user behavior feature information 104 and object click label 105. Then, the electronic device 101 can input the user behavior feature information 104 into the initial user interest information generation model 108 to generate initial user interest information 106. The initial user interest information 106 is the interest feature information of the user under at least one user interest point. In this application scenario, the at least one user interest point may include: a first user interest point, a second user interest point and a third user interest point. Then, the electronic device 101 can use the initial interest strength information generation model 109 to generate the initial interest strength information 107 corresponding to the at least one user interest point. In this application scenario, the initial interest strength information 107 includes: a value "0.2" corresponding to the first user interest point, a value "0.4" corresponding to the second user interest point, and a value "0.4" corresponding to the third user interest point. Furthermore, the electronic device 101 may generate initial object click information 110 based on the recommended object feature information 103 and the initial user interest information 106. Finally, the electronic device 101 may perform model training on the initial user interest information generation model 108 and the initial interest intensity information generation model 109 based on the initial interest intensity information 107, the initial object click information 110 and the object click label 105 to generate a user interest information generation model 111 and an interest intensity information generation model 112.

[0054] It should be noted that the electronic device 101 can be hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or it can be implemented as a single server or a single terminal device. When the electronic device is embodied as software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, for example, or it can be implemented as a single software or software module. No specific limitation is made here.

[0055] It should be understood that Figure 1 The number of electronic devices in the embodiment is only for illustration. Any number of electronic devices may be provided according to implementation requirements.

[0056] Continue to refer Figure 2 , shows a process 200 of some embodiments of the model training method according to the present disclosure. The model training method comprises the following steps:

[0057] Step 201, obtaining target training samples.

[0058] In some embodiments, the execution entity of the above model training method (for example Figure 1 The electronic device 101 shown can obtain the target training sample through a wired connection or a wireless connection. Among them, the target training sample can be a sample used for subsequent model training. Among them, the above-mentioned target training sample includes: recommended object feature information, user behavior feature information and object click label. Recommended object feature information can be feature content of recommended object related features. Recommended object can be an object that supports recommendation. In practice, the recommended object can be for different recommendation scenarios and can be items of different categories. For example, for the recommendation scenario on a news website, the corresponding recommended object can be news. For the recommendation scenario on an e-commerce website, the corresponding recommended object can be a commodity. For the short video recommendation scenario, the corresponding recommended object can be a video. The recommended object related features can include but are not limited to at least one of the following: recommended object views, recommended object theme, recommended object forwarding volume, recommended object like volume. User behavior feature information can be behavior feature information of the user's behavior on the recommended item in a historical time period. The user corresponding to the user behavior feature information is the recommended user corresponding to the recommended item. In practice, the user behavior feature information can include: user click behavior feature information, user collection behavior feature information, and user download behavior feature information. The object click label may be an actual label of a user performing a click operation on the recommended object in a future time period.

[0059] Step 202: input the above user behavior characteristic information into an initial user interest information generation model to generate initial user interest information.

[0060] In some embodiments, the execution subject may input the user behavior feature information into the initial user interest information generation model to generate initial user interest information. The initial user interest information is the user's interest feature information under at least one user interest point. The initial user interest information generation model may be a neural network model that has not been trained yet and is used to generate initial user interest information. The user interest point may be a feature point that the user is interested in. For example, the user interest point may be but is not limited to at least one of the following: a first brand of mobile phone, a second brand of car, and a third brand of fishing rod. The interest feature information may be an interest feature corresponding to the user interest point, or may be the behavior feature information under the user interest point. That is, the interest feature information may characterize how the user likes to perform behavior operations on the user interest point. For example, the behavior feature information may include: behavior information characterizing browsing of the first brand of mobile phone, behavior information characterizing browsing of the second brand of car, and behavior information characterizing purchasing of the third brand of fishing rod. In practice, the user interest information generation model may be a Transformer model.

[0061] In some optional implementations of some embodiments, the user behavior characteristic information includes: at least one behavior characteristic information sequence corresponding to at least one behavior type. At least one behavior type may be a characterization of different behavior operations. In practice, at least one behavior type may be pre-set for the corresponding recommendation scenario. For example, for the commodity recommendation scenario, the corresponding at least one behavior type may include: click behavior type, add-to-cart behavior type, purchase behavior type, and collection behavior type. The behavior type in at least one behavior type corresponds one-to-one to the behavior characteristic information sequence in at least one behavior characteristic information sequence. The behavior characteristic information may be the characteristic information generated by the user performing the corresponding behavior in a historical time period. That is, the behavior characteristic information in the behavior characteristic information sequence is sorted in chronological order. For example, for the behavior type being a user click behavior type, the corresponding behavior characteristic information sequence may be a characteristic information sequence of the user performing a click operation on the recommended object in a historical time period.

[0062] Optionally, the execution subject may input the user behavior feature information into an initial user interest information generation model to generate initial user interest information, including:

[0063] In the first step, each behavior feature information in the at least one behavior feature information sequence is transformed into a vector using the initial user interest information generation model to generate a behavior vector and obtain at least one behavior vector sequence. The behavior vector may be a vector that represents the semantics of the behavior feature.

[0064] As an example, the execution entity may input each behavior feature information in at least one behavior feature information sequence into an embedding lookup layer included in the initial user interest information generation model to generate a behavior vector, thereby obtaining at least one behavior vector sequence.

[0065] In the second step, for each of the at least one behavior type, the initial user interest information generation model is used to generate a key behavior vector corresponding to the behavior type. The key behavior vector can represent the behavior feature semantic content of the key behavior (important behavior, i.e., the behavior that the user likes very much) performed by the user under the behavior corresponding to the behavior type.

[0066] As an example, for each behavior type of the at least one behavior type, a behavior vector sequence corresponding to the behavior type is input into an initial attention mechanism model corresponding to the behavior type to generate a key behavior vector.

[0067] The third step is to generate the above-mentioned initial user interest information according to the obtained at least one key behavior vector.

[0068] As an example, the execution subject may input at least one key behavior vector into an initial attention mechanism model to generate the initial user interest information. The initial user interest information generation model includes: an initial attention mechanism model.

[0069] In some optional implementations of some embodiments, the execution subject may generate a key behavior vector corresponding to the behavior type, including:

[0070] In the first step, in response to determining that the above behavior type is a click behavior type, clustering is performed on the behavior vector sequence corresponding to the above behavior type to generate a behavior vector cluster set. The click behavior type is a behavior type related to click behavior. Each behavior vector in the behavior vector cluster has the same behavior feature information.

[0071] As an example, the execution entity may use a K-means clustering algorithm to perform clustering processing on the behavior vector sequences corresponding to the behavior types to generate behavior vector clusters.

[0072] The second step is to generate key behavior vectors corresponding to the click behavior types according to the behavior vector cluster center set in the behavior vector cluster set, wherein the behavior vector cluster centers in the behavior vector cluster center set and the behavior vector clusters in the behavior vector cluster set have a one-to-one correspondence.

[0073] As an example, in the first step, for each behavior vector cluster center in the behavior vector cluster center set, first, determine the behavior vector cluster corresponding to the above behavior vector cluster center as the target behavior vector cluster. Then, select the behavior vectors whose similarity with the above behavior vector cluster center is in the first target number from the target behavior vector cluster as the key behavior vectors, and obtain at least one key behavior vector. In the second step, vector combination is performed on the obtained multiple key behavior vectors and the behavior vector cluster center set to generate a combination matrix as the key behavior vector.

[0074] As another example, the execution subject may directly combine the behavior vector cluster centers in the behavior vector cluster center set to generate a combination matrix as the key behavior vector.

[0075] In the third step, in response to determining that the above-mentioned behavior type is a value-added behavior type, determine at least one first behavior vector whose time interval with the target time satisfies a preset time interval condition. The value-added behavior type may be a behavior type related to the value-added behavior. In practice, in an e-commerce scenario, the value-added behavior may be an add-to-cart behavior. The target time may be the predicted time of the object click corresponding to the object click label. The preset time interval condition may be that the first behavior vector is a behavior vector in the corresponding behavior vector sequence, and the time interval between the corresponding time and the target is less than the target duration.

[0076] The fourth step is to combine the at least one first behavior vector to generate a key behavior vector corresponding to the value-added behavior type.

[0077] As an example, the execution entity may perform vector concatenation on at least one first behavior vector to generate a key behavior vector corresponding to the value-added behavior type.

[0078] In step 5, in response to determining that the above-mentioned behavior type is an item inflow type, at least one second behavior vector corresponding to the number of item inflows that meets a preset number condition is determined. The item inflow type may be a behavior type related to the item inflow behavior. In practice, in an e-commerce scenario, the item inflow behavior may be a purchase behavior. The preset number condition may be a second behavior vector corresponding to the number of item inflows that is within the previous target number.

[0079] In a sixth step, the at least one second behavior vector is combined to generate a key behavior vector corresponding to the item inflow type.

[0080] As an example, the execution entity may perform vector concatenation on at least one second behavior vector to generate a key behavior vector corresponding to the item inflow type.

[0081] In some optional implementations of some embodiments, the user behavior feature information further includes: a behavior content feature information sequence for at least one of the above-mentioned behavior types. The behavior content feature information in the behavior content feature information sequence is arranged according to the corresponding generation time. The behavior content feature information may be content feature information related to the recommended object in the behavior feature information under the corresponding behavior type. For example, the behavior content feature information may include but is not limited to at least one of the following: a recommended object identifier, a recommended object category, a recommended object brand, and a recommended object keyword.

[0082] In practice, first, the behavior content feature information corresponding to each behavior feature information in the at least one behavior feature information sequence is determined to obtain a behavior content feature information set. Then, the behavior content feature information in the behavior content feature information set is sorted according to the generation time corresponding to each behavior content feature information to generate a behavior content feature information sequence.

[0083] Optionally, the execution subject may generate the initial user interest information according to the obtained at least one key behavior vector, including the following steps:

[0084] The first step is to encode the behavior content feature information in the above behavior content feature information sequence to generate a behavior content vector sequence.

[0085] As an example, for each behavioral content feature information, first, the above-mentioned execution entity can input each content feature information (for example, recommended object identifier, recommended object category, recommended object brand, recommended object keyword, etc.) in the above-mentioned behavioral content feature information into the Embedding Lookup layer to generate each content feature vector. Then, each content feature vector is vector-concatenated to generate a concatenated vector. Next, the concatenated vector is input into the MLP (Multi-LayerPerceptron) layer for encoding to generate an encoded vector. Finally, each element in the encoded vector is normalized to generate a behavioral content vector.

[0086] The second step is to perform vector fusion on the at least one key behavior vector to generate a key behavior matrix for the at least one behavior type.

[0087] As an example, the execution entity may merge and remove duplicates of at least one key behavior vector to generate a key behavior matrix.

[0088] The third step is to input the above key behavior matrix and the above behavior content vector sequence into the initial attention mechanism model to generate a similarity matrix. The above similarity matrix represents the correlation strength between the key behavior and the behavior content, and the above initial user interest information generation model includes: an initial attention mechanism model. The initial attention mechanism model can be an attention mechanism model that has not been trained. In practice, the attention mechanism model can be a multi-head attention mechanism model. The elements in the similarity matrix can represent the degree of correlation between the corresponding behavior and the behavior content feature information. The elements are values ​​between 0 and 1. The higher the value, the higher the correlation between the corresponding behavior and the behavior content.

[0089] The fourth step is to generate the initial user interest information based on the similarity matrix.

[0090] As an example, first, the execution subject can filter out elements whose corresponding element values ​​are greater than a preset value from the similarity matrix to obtain an element set. Then, the behavior content corresponding to each element in the element set is determined to obtain a behavior content set. Finally, according to the behavior content feature information set corresponding to the behavior content set, the relevant interest feature information for at least one user interest point is extracted to obtain an interest feature information set as the initial user interest information.

[0091] In some optional implementations of some embodiments, the user behavior feature information further includes: a first behavior category feature information sequence for the at least one behavior type. The behavior category feature information in the behavior category feature information sequence is arranged according to the corresponding generation time. The behavior category feature information may be the category feature information related to the recommended object in the behavior feature information under the corresponding behavior type. For example, the behavior category feature information may include but is not limited to at least one of the following: behavior type, behavior time, stay duration, number of behaviors, and internal page actions.

[0092] Optionally, the execution entity may generate the initial user interest information according to the similarity matrix, including:

[0093] The first step is to encode the first behavior category feature information in the first behavior category feature information sequence to generate a behavior category vector sequence.

[0094] As an example, the execution entity may input the first behavior category feature information in the first behavior category feature information sequence into the Embedding Lookup layer to generate a behavior category vector sequence.

[0095] The second step is to perform masking on the elements in the similarity matrix to generate a mask matrix, wherein the element value in the mask matrix can be one of the following: value 0, value 1.

[0096] As an example, for each row of similarity vectors in the similarity matrix, first, the similarity soft threshold corresponding to the similarity vector is determined using the MLP layer. Each row of similarity vectors represents the content similarity between each user's interest point and each behavior content. Then, the elements in each row of similarity vectors that are less than the similarity soft threshold are determined as a value of 0, and the elements that are greater than or equal to the similarity soft threshold are determined as a value of 1 to generate a determined vector. Finally, the determined vectors corresponding to each row are combined to generate a mask matrix.

[0097] The third step is to generate interest point derived content information based on the mask matrix and the behavior content vector sequence. The interest point derived content information may be behavior content information derived from the user's interest point. That is, the interest point derived content information may represent derived content related to the user's interest point that the user may be interested in.

[0098] As an example, the execution subject may multiply the mask matrix and the behavior content matrix corresponding to the behavior content vector sequence to generate a multiplication matrix as the interest point derived content information.

[0099] The fourth step is to generate interest point derived category information by combining the mask matrix and the behavior category vector sequence. The interest point derived category information may be behavior category information derived from the user's interest point. That is, the interest point derived category information may represent derived categories related to the user's interest point that the user may be interested in.

[0100] As an example, the execution subject may multiply the mask matrix and the behavior category matrix corresponding to the behavior category vector sequence to generate a multiplication matrix as the interest point derived category information.

[0101] The fifth step is to generate the initial user interest information according to the interest point derived content information and the interest point derived category information.

[0102] As an example, the execution entity may combine the interest point derived content information and the interest point derived category information to generate initial user interest information.

[0103] In some optional implementations of some embodiments, the execution entity may input the above-mentioned interest point derived content information and the above-mentioned interest point derived category information into the initial interest information extraction model to generate the above-mentioned initial user interest information. The above-mentioned initial user interest information generation model includes: the above-mentioned initial interest information extraction model. In practice, the initial interest information extraction model may be a time series neural network model that has not yet been trained and is used to extract user interest information. In practice, the initial interest information extraction model may be a recurrent neural network (RNN) model.

[0104] Step 203: using the initial interest intensity information generation model, generate initial interest intensity information corresponding to the at least one user interest point.

[0105] In some embodiments, the execution subject may generate initial interest intensity information corresponding to the at least one user interest point using an initial interest intensity information generation model. The initial interest intensity information generation model may be a neural network model for generating interest intensity information that has not been trained. The interest intensity information may characterize the user's interest level in each of the at least one user interest point. In practice, the at least one user interest point may include: a first brand mobile phone, a second brand car, and a third brand fishing rod. The initial interest intensity information may include: the interest level corresponding to the "first brand mobile phone", the interest level corresponding to the "second brand car", and the interest level corresponding to the "third brand fishing rod". The interest level corresponding to each user interest point is followed by a value "1". For example, the initial interest intensity information may include: the interest level "0.2" corresponding to the "first brand mobile phone", the interest level "0.4" corresponding to the "second brand car", and the interest level "0.4" corresponding to the "third brand fishing rod". In practice, the initial interest intensity information generation model may be a Transformer model.

[0106] As an example, the execution entity may input the initial user interest information into the initial interest intensity information generation model to generate the initial interest intensity information corresponding to the at least one user interest point.

[0107] In some optional implementations of some embodiments, the execution subject may generate the initial interest intensity information corresponding to the at least one user interest point by using the initial interest intensity information generation model, including:

[0108] The first step is to generate a second behavior category feature information sequence for the above key behavior matrix.

[0109] The second step is to input the second behavior category feature information sequence into the initial interest intensity information generation model to generate the initial interest intensity information.

[0110] Step 204: Generate initial object click information based on the recommended object feature information and the initial user interest information.

[0111] In some embodiments, the execution entity may generate initial object click information based on the recommended object feature information and the initial user interest information. The initial object click information may be a prediction of the user's click on the recommended object at a target time. For example, the target time may be a predetermined time point after a historical time period. In practice, the initial object click information may be an object click rate.

[0112] As an example, first, the execution subject may encode the recommended object feature information to generate a recommended object feature vector, and then determine the Cosine distance metric similarity between the recommended object feature vector and the information vector corresponding to the initial user interest information as the initial object click information.

[0113] Step 205, based on the initial interest strength information, the initial object click information and the object click label, the initial user interest information generation model and the initial interest strength information generation model are trained to generate a user interest information generation model and an initial interest strength information generation model.

[0114] In some embodiments, the above-mentioned execution entity can perform model training on the above-mentioned initial user interest information generation model and the above-mentioned initial interest intensity information generation model based on the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label to generate a user interest information generation model and an interest intensity information generation model.

[0115] As an example, first, the execution subject may obtain an interest intensity label for at least one user interest point. The interest intensity label may be an interest level label for at least one user interest point that is marked by the user. Then, the interest intensity loss information between the interest intensity label and the initial interest intensity information is determined. Next, the click loss information between the initial object click information and the object click label is determined. Next, the interest intensity loss information and the click loss information are weighted summed to generate weighted sum loss information. Finally, in response to determining that the corresponding value of the loss information sequence corresponding to the weighted sum loss information tends to be stable, the initial user interest information generation model and the initial interest intensity information generation model are respectively determined as the user interest information generation model and the interest intensity information generation model. In response to determining that it does not tend to be stable, the model parameters of the initial user interest information generation model and the initial interest intensity information generation model are updated according to the weighted sum loss information to continue to obtain training samples for model training until the loss information finally tends to be stable.

[0116] In some optional implementations of some embodiments, the execution subject may perform model training on the initial user interest information generation model and the initial interest intensity information generation model according to the initial interest intensity information, the initial object click information, and the object click label to generate a user interest information generation model and an interest intensity information generation model, which may include the following steps:

[0117] The first step is to generate an initial interest intensity label based on the above-mentioned recommended object feature information and the above-mentioned initial user interest information.

[0118] As an example, first, the execution entity may perform similarity multiplication on the recommendation object feature information corresponding matrix and the initial user interest information corresponding matrix to generate at least one similarity for at least one user interest point. Then, at least one of the largest similarities is set to a value of 1, and the remaining similarities are set to 0 to generate a numerical matrix as an initial interest intensity label. The initial user interest information corresponding matrix includes: an interest point vector corresponding to at least one user interest point. The similarity in the at least one similarity may be the result of multiplying the interest point vector and the recommendation object feature information corresponding matrix.

[0119] The second step is to generate interest intensity loss information for the initial interest intensity label and the initial interest intensity information, wherein the interest intensity loss information may represent interest intensity difference information between the initial interest intensity label and the initial interest intensity information at at least one user interest point.

[0120] As an example, the execution entity may generate interest intensity loss information for the initial interest intensity label and the initial interest intensity information using the Cross Entropy loss function.

[0121] The third step is to generate click loss information for the initial object click information and the object click label, wherein the click loss information can represent the click difference between the initial object click information and the object click label.

[0122] In the fourth step, based on the above interest intensity loss information and the above click loss information, the above initial user interest information generation model, the above initial interest intensity information generation model and the above initial recommended object feature encoding model are trained to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0123] As an example, first, the execution subject may perform a weighted summation of the interest intensity loss information and the click loss information to generate weighted sum loss information. Then, based on the weighted sum loss information, the initial user interest information generation model, the initial interest intensity information generation model, and the initial recommended object feature encoding model are trained to generate a user interest information generation model, an initial interest intensity information generation model, and a recommended object feature encoding model.

[0124] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model training method of some embodiments of the present disclosure, the user interest information generation model and the interest intensity information generation model can be used, and based on the generated interest intensity information, the object recommendation can be more accurately implemented. Specifically, the reason for the inaccuracy of the relevant object recommendation is that: there is no deeper consideration of at least one user interest point corresponding to the user, resulting in the object recommendation based on at least one user interest point, and there is a problem that the recommended object is not accurate enough. Based on this, the model training method of some embodiments of the present disclosure first obtains the target training sample, wherein the above-mentioned target training sample includes: recommended object feature information, user behavior feature information and object click label. Here, the obtained target training sample is used for model training. In addition, the recommended object feature information can characterize the feature information of the recommended object, so as to facilitate the subsequent generation of initial object click information. The user behavior feature information can fully reflect the feature information of the user in each behavior, so as to be used for the subsequent generation of the user's interest feature information under at least one user interest point. The object click label is used as the real data of the object click for the subsequent model training. Then, the user behavior feature information is input into the initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is the interest feature information of the user under at least one user interest point. Here, by using the initial user interest information generation model, the interest feature content of the user under at least one user interest point can be fully grasped, so as to facilitate the subsequent model training. Then, the initial interest strength information generation model is used to generate the initial interest strength information corresponding to the at least one user interest point. Here, the initial interest strength information can characterize the user's interest in at least one user interest point. In the model training process, by generating the initial interest strength information, the importance of the user under at least one user interest point can be fully considered, so that in the subsequent object recommendation, the adaptive object recommendation adjustment can be made according to the interest strength, so that the recommended items are more accurate and reasonable. Furthermore, according to the recommended object feature information and the initial user interest information, the initial object click information is generated as the predicted object click information, and the corresponding click loss is formed with the object click label, which is convenient for the subsequent model training. Finally, based on the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label, the above-mentioned initial user interest information generation model and the above-mentioned initial interest intensity information generation model are trained to generate a user interest information generation model and an interest intensity information generation model.In summary, during the model training process, by considering the user's interest feature information and the corresponding interest intensity information at at least one user interest point, the user interest information generation model and the interest intensity information generation model can be subsequently used to more accurately achieve object recommendation based on the generated interest intensity information.

[0125] Further references Figure 3 , shows a process 300 of other embodiments of the model training method according to the present disclosure. The model training method comprises the following steps:

[0126] Step 301, obtaining target training samples.

[0127] Step 302: input the above user behavior characteristic information into an initial user interest information generation model to generate initial user interest information.

[0128] Step 303: using the initial interest intensity information generation model, generate initial interest intensity information corresponding to the at least one user interest point.

[0129] In some embodiments, the specific implementation of steps 301-303 and the technical effects thereof can be referred to in Figure 2 The steps 201-203 in the corresponding embodiment are not described in detail here.

[0130] Step 304: input the recommended object feature information into an initial recommended object feature encoding model to generate object feature encoding information.

[0131] In some embodiments, an execution entity (e.g. Figure 1 The electronic device 101 shown in the figure can input the above-mentioned recommended object feature information into the initial recommended object feature coding model to generate object feature coding information. Among them, the initial recommended object feature coding model can be a recommended object feature coding model that has not yet been trained and is used to perform information coding processing on the recommended object feature information. The object feature coding information can be information that characterizes the semantics of the feature content corresponding to the recommended object feature information. Specifically, the object feature coding information can be information in the form of a vector. In practice, the recommended object feature coding model can include: an MLP layer and a normalization layer.

[0132] Step 305: Generate the initial object click information according to the object feature coding information and the initial user interest information.

[0133] In some embodiments, the execution entity may generate the initial object click information according to the object feature coding information and the initial user interest information.

[0134] Step 306, based on the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label, the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model are trained to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0135] In some embodiments, the above-mentioned execution entity may perform model training on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model based on the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0136] As an example, first, an initial interest intensity label is generated based on the above-mentioned recommended object feature information and the above-mentioned initial user interest information. Then, interest intensity loss information is generated for the above-mentioned initial interest intensity label and the above-mentioned initial interest intensity information. Next, click loss information is generated for the above-mentioned initial object click information and the above-mentioned object click label. Furthermore, the above-mentioned interest intensity loss information and the above-mentioned click loss information are weighted and processed to generate weighted sum loss information. Finally, based on the above-mentioned weighted sum loss information, model training is performed on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0137] from Figure 3 It can be seen that Figure 2 Compared with the description of some corresponding embodiments, Figure 3 In the process 300 of the model training method in some corresponding embodiments, by utilizing the trained recommended object feature encoding model, more accurate encoding of recommended object feature information can be achieved, so that subsequent object recommendations are more accurate.

[0138] Continue to refer Figure 4 , shows a process 400 of some embodiments of the object recommendation method according to the present disclosure. The object recommendation method comprises the following steps:

[0139] Step 401: Acquire user characteristic information corresponding to a target user as target user characteristic information.

[0140] In some embodiments, the execution subject (e.g., electronic device) of the object recommendation method can obtain user feature information corresponding to the target user through a wired connection or a wireless connection as the target user feature information. The target user can be the user to be recommended an object. The target user feature information can be information under various features related to the target user.

[0141] Step 402: Based on the target user feature information, a pre-trained user interest information generation model is used to generate user interest information for the target user.

[0142] In some embodiments, the execution subject may generate user interest information for the target user based on the target user feature information using a pre-trained user interest information generation model. The user interest information generation model is trained based on a model training method. The user interest information is the interest feature information of the target user under at least one user interest point.

[0143] As an example, first, the execution subject may extract behavior feature information from the target user feature information, and then input the behavior feature information into the user interest information generation model to generate user interest information for the target user.

[0144] Step 403: Generate an initial recommendation object information set for at least one user interest point based on the user interest information.

[0145] In some embodiments, the execution entity may generate an initial recommendation object information set for at least one user interest point based on the user interest information.

[0146] As an example, the recommended object information with the largest Cosine similarity with the user interest vector corresponding to the user interest information is retrieved on the nearest neighbor retrieval tree constructed based on the recommended object vectors of the entire library of recommended objects as the initial recommended object information to obtain the initial recommended object information set.

[0147] Step 404: Generate interest intensity information for the at least one user interest point using a pre-trained interest intensity information generation model.

[0148] In some embodiments, the execution subject may generate interest intensity information for the at least one user interest point using a pre-trained interest intensity information generation model, wherein the interest intensity information generation model may be generated based on a model training method.

[0149] As an example, first, the execution subject may generate a second behavior category feature information sequence for the key behavior matrix, and then input the second behavior category feature information sequence into the initial interest intensity information generation model to generate the interest intensity information.

[0150] Step 405 , based on the interest strength information, filter out a recommended object information set from the initial recommended object information set to perform object recommendation for the recommended object information set.

[0151] In some embodiments, the execution subject may filter out a recommended object information set from the initial recommended object information set according to the interest strength information, so as to perform object recommendation for the recommended object information set.

[0152] In some optional implementations of some embodiments, the filtering of the recommended object information set from the initial recommended object information set according to the interest strength information may include the following steps:

[0153] For each user interest point in the at least one user interest point, the following first generating step is performed:

[0154] Sub-step 1: determine the initial recommended object information corresponding to the user's interest point in the initial recommended object information set to obtain at least one initial recommended object information.

[0155] Sub-step 2: determining the number of recommended objects corresponding to the user's interest points based on the interest strength information.

[0156] As an example, first, the execution subject may normalize the intensity value corresponding to at least one user interest point in the interest intensity information to generate each normalized intensity value. Then, the normalized intensity value corresponding to the user interest point is selected from each normalized intensity value as the target intensity value. Finally, the target intensity value is multiplied by the total number of recommended items to generate the number of recommended objects.

[0157] Sub-step 3, selecting the initial recommended object information whose corresponding recommendation similarity satisfies the preset similarity condition from the at least one initial recommended object information as the recommended object information, and obtaining at least one recommended object information corresponding to the user's interest point, wherein the number of information corresponding to the at least one recommended object information is the same as the number of recommended objects. The preset similarity condition may be that the similarity size is within the previous target number.

[0158] In some optional implementations of some embodiments, after step 406, the steps further include:

[0159] For each user interest point in the at least one user interest point, the following second generating step is performed:

[0160] Sub-step 1: Determine the recall result corresponding to the user's interest point and the corresponding interest strength value in the interest strength information. The recall result may be the recall result of at least one recommended object information corresponding to the user's interest point. The interest strength value may be the interest strength value corresponding to the user's interest point in the interest strength information.

[0161] Sub-step 2: determining the click rate corresponding to the user's interest point based on the recall result and the interest strength value.

[0162] As an example, first, determine the similarity value between the recall result and the user's interest point. Then, multiply the similarity value by the first adjustable parameter to generate a first multiplied value. Next, add the first multiplied value and the second adjustable parameter to generate an added value. Further, use the added value as a base and the third adjustable parameter as an exponential value to generate an exponential value. Finally, multiply the exponential value by the interest intensity value to generate a click-through rate.

[0163] Here, by determining the click rate in this way, we can introduce a tendency that the recall results of strong interest points are better than those of weak interest points, and we can make adjustments based on the recall effect to ensure the optimal efficiency of the online recall results.

[0164] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: the object recommendation method of some embodiments of the present disclosure utilizes a user interest information generation model and an interest intensity information generation model, and can implement object recommendation more accurately based on the generated interest intensity information.

[0165] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a model training device, and these device embodiments are Figure 2 Corresponding to the method embodiments shown, the model training device can be specifically applied to various electronic devices.

[0166] like Figure 5As shown, a model training device 500 includes: a first acquisition unit 501, a first generation unit 502, a second generation unit 503, a third generation unit 504 and a training unit 505. Among them, the first acquisition unit 501 is configured to acquire target training samples, wherein the above-mentioned target training samples include: recommended object feature information, user behavior feature information and object click labels; the first generation unit 502 is configured to input the above-mentioned user behavior feature information into the initial user interest information generation model to generate initial user interest information, wherein the above-mentioned initial user interest information is the user's interest feature information under at least one user interest point; the second generation unit 503 is configured to use the initial interest strength information generation model to generate the initial interest strength information corresponding to the above-mentioned at least one user interest point; the third generation unit 504 is configured to generate initial object click information according to the above-mentioned recommended object feature information and the above-mentioned initial user interest information; the training unit 505 is configured to perform model training on the above-mentioned initial user interest information generation model and the above-mentioned initial interest strength information generation model according to the above-mentioned initial interest strength information, the above-mentioned initial object click information and the above-mentioned object click label to generate a user interest information generation model and an interest strength information generation model.

[0167] In some optional implementations of some embodiments, the third generation unit 504 can be further configured to: input the above-mentioned recommended object feature information into the initial recommended object feature coding model to generate object feature coding information; and generate the above-mentioned initial object click information based on the above-mentioned object feature coding information and the above-mentioned initial user interest information.

[0168] In some optional implementations of some embodiments, the training unit 505 can be further configured to: perform model training on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model according to the above-mentioned initial interest intensity information, the above-mentioned initial object click information and the above-mentioned object click label, so as to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0169] In some optional implementations of some embodiments, the above-mentioned user behavior characteristic information includes: at least one behavior characteristic information sequence corresponding to at least one behavior type; and the first generation unit 502 can be further configured to: use the above-mentioned initial user interest information generation model to vectorize each behavior characteristic information in the above-mentioned at least one behavior characteristic information sequence to generate a behavior vector and obtain at least one behavior vector sequence; for each behavior type in the above-mentioned at least one behavior type, use the above-mentioned initial user interest information generation model to generate a key behavior vector corresponding to the above-mentioned behavior type; and generate the above-mentioned initial user interest information based on the obtained at least one key behavior vector.

[0170] In some optional implementations of some embodiments, the first generation unit 502 may be further configured to: in response to determining that the above-mentioned behavior type is a click behavior type, clustering the behavior vector sequence corresponding to the above-mentioned behavior type to generate a behavior vector cluster set; generating a key behavior vector corresponding to the above-mentioned click behavior type based on the behavior vector cluster center set in the above-mentioned behavior vector cluster set; in response to determining that the above-mentioned behavior type is a value-added behavior type, determining at least one first behavior vector whose time interval with the target time satisfies a preset time interval condition; combining the above-mentioned at least one first behavior vector to generate a key behavior vector corresponding to the above-mentioned value-added behavior type; in response to determining that the above-mentioned behavior type is an item inflow type, determining at least one second behavior vector whose corresponding item inflow times satisfies a preset time condition; combining the above-mentioned at least one second behavior vector to generate a key behavior vector corresponding to the item inflow type.

[0171] In some optional implementations of some embodiments, the above-mentioned user behavior feature information also includes: a behavior content feature information sequence for the above-mentioned at least one behavior type; and the first generation unit 502 can be further configured to: encode the behavior content feature information in the above-mentioned behavior content feature information sequence to generate a behavior content vector sequence; perform vector fusion on the above-mentioned at least one key behavior vector to generate a key behavior matrix for the above-mentioned at least one behavior type; input the above-mentioned key behavior matrix and the above-mentioned behavior content vector sequence into the initial attention mechanism model to generate a similarity matrix, wherein the above-mentioned similarity matrix represents the correlation strength between the key behavior and the behavior content, and the above-mentioned initial user interest information generation model includes: an initial attention mechanism model; based on the above-mentioned similarity matrix, generate the above-mentioned initial user interest information.

[0172] In some optional implementations of some embodiments, the above-mentioned user behavior feature information also includes: a first behavior category feature information sequence for the above-mentioned at least one behavior type; and the first generation unit 502 can be further configured to: encode the first behavior category feature information in the above-mentioned first behavior category feature information sequence to generate a behavior category vector sequence; mask the elements in the above-mentioned similarity matrix to generate a mask matrix; generate interest point derived content information based on the above-mentioned mask matrix and the above-mentioned behavior content vector sequence; generate interest point derived category information by combining the above-mentioned mask matrix and the above-mentioned behavior category vector sequence; generate the above-mentioned initial user interest information based on the above-mentioned interest point derived content information and the above-mentioned interest point derived category information.

[0173] In some optional implementations of some embodiments, the first generation unit 502 can be further configured to: input the above-mentioned interest point derived content information and the above-mentioned interest point derived category information into the initial interest information extraction model to generate the above-mentioned initial user interest information, wherein the above-mentioned initial user interest information generation model includes: the above-mentioned initial interest information extraction model.

[0174] In some optional implementations of some embodiments, the second generation unit 503 may be further configured to: generate a second behavior category feature information sequence for the above-mentioned key behavior matrix; input the above-mentioned second behavior category feature information sequence into the above-mentioned initial interest intensity information generation model to generate the above-mentioned initial interest intensity information.

[0175] In some optional implementations of some embodiments, the training unit 505 can be further configured to: generate an initial interest intensity label based on the above-mentioned recommended object feature information and the above-mentioned initial user interest information; generate interest intensity loss information for the above-mentioned initial interest intensity label and the above-mentioned initial interest intensity information; generate click loss information for the above-mentioned initial object click information and the above-mentioned object click label; and perform model training on the above-mentioned initial user interest information generation model, the above-mentioned initial interest intensity information generation model and the above-mentioned initial recommended object feature encoding model based on the above-mentioned interest intensity loss information and the above-mentioned click loss information to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

[0176] It is understandable that the units recorded in the model training device 500 are similar to the reference Figure 2 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the model training device 500 and the units contained therein, and will not be repeated here.

[0177] Further references Figure 6As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an object recommendation device. These device embodiments are similar to Figure 4 Corresponding to the method embodiments shown, the object recommendation device can be specifically applied to various electronic devices.

[0178] like Figure 6 As shown, an object recommendation device 600 includes: a second acquisition unit 601, a fourth generation unit 602, a fifth generation unit 603, a sixth generation unit 604 and a recommendation unit 605. The second acquisition unit 601 is configured to acquire user feature information corresponding to a target user as target user feature information; the fourth generation unit 602 is configured to generate user interest information for the target user based on the target user feature information and using a pre-trained user interest information generation model, wherein the user interest information generation model is generated based on a model training method; the fifth generation unit 603 is configured to generate an initial recommended object information set for at least one user interest point based on the user interest information; the sixth generation unit 604 is configured to generate interest intensity information for at least one user interest point based on a pre-trained interest intensity information generation model, wherein the interest intensity information generation model is generated based on a model training method; the recommendation unit 605 is configured to filter out a recommended object information set from the initial recommended object information set based on the interest intensity information, so as to perform object recommendation for the recommended object information set.

[0179] In some optional implementations of some embodiments, the recommendation unit 605 may be further configured to: for each user interest point in at least one user interest point, perform the following first generation step: determine the initial recommendation object information corresponding to the above-mentioned user interest point in the above-mentioned initial recommendation object information set, and obtain at least one initial recommendation object information; determine the number of recommended objects corresponding to the above-mentioned user interest point based on the above-mentioned interest intensity information; filter out the initial recommendation object information whose corresponding recommendation similarity meets the preset similarity condition from the above-mentioned at least one initial recommendation object information as the recommended object information, and obtain at least one recommended object information corresponding to the above-mentioned user interest point, wherein the number of information corresponding to the above-mentioned at least one recommended object information is the same as the above-mentioned number of recommended objects.

[0180] In some optional implementations of some embodiments, the device further includes: an execution unit. The execution unit is configured to: for each user interest point of the at least one user interest point, perform the following second generation step: determine a recall result corresponding to the user interest point and a corresponding interest intensity value in the interest intensity information; and determine a click rate corresponding to the user interest point according to the recall result and the interest intensity value.

[0181] It can be understood that the units described in the object recommendation device 600 are similar to those described in the reference Figure 4 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the object recommendation device 600 and the units included therein, and will not be described in detail here.

[0182] Reference below Figure 7 , which shows an electronic device (eg, Figure 1 Schematic diagram of the structure of the electronic device 101)700. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0183] like Figure 7 As shown, the electronic device 700 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory 702 or a program loaded from a storage device 708 into a random access memory 703. Various programs and data required for the operation of the electronic device 700 are also stored in the random access memory 703. The processing device 701, the read-only memory 702, and the random access memory 703 are connected to each other via a bus 704. An input / output interface 705 is also connected to the bus 704.

[0184] Typically, the following devices may be connected to the input / output interface 705: an input device 706 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 708 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 709. The communication device 709 may allow the electronic device 700 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 7 The electronic device 700 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 7 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0185] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 709, or installed from the storage device 708, or installed from the read-only memory 702. When the computer program is executed by the processing device 701, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0186] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0187] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with 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"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0188] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a target training sample, wherein the target training sample includes: recommended object feature information, user behavior feature information, and object click labels; inputs the user behavior feature information into the initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is the user's interest feature information under at least one user interest point; generates initial interest strength information corresponding to the at least one user interest point using the initial interest strength information generation model; generates initial object click information based on the recommended object feature information and the initial user interest information; performs model training on the initial user interest information generation model and the initial interest strength information generation model based on the initial interest strength information, the initial object click information, and the object click label to generate a user interest information generation model and an interest strength information generation model. Obtain user characteristic information corresponding to the target user as the target user characteristic information; based on the target user characteristic information, generate user interest information for the target user using a pre-trained user interest information generation model, wherein the user interest information generation model is generated based on a model training method; based on the user interest information, generate an initial recommended object information set for at least one user interest point; use a pre-trained interest intensity information generation model to generate interest intensity information for at least one user interest point, wherein the interest intensity information generation model is generated based on a model training method; based on the interest intensity information, filter out a recommended object information set from the initial recommended object information set to perform object recommendation for the recommended object information set.

[0189] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0190] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0191] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be provided in a processor, for example, may be described as: a processor including a first acquisition unit, a first generation unit, a second generation unit, a third generation unit and a training unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the first acquisition unit may also be described as a "unit for acquiring target training samples".

[0192] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0193] Some embodiments of the present disclosure also provide a computer program product, including a computer program, which implements any of the above-mentioned model training methods or object recommendation methods when executed by a processor.

[0194] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A model training method, comprising: Obtaining a target training sample, wherein the target training sample includes: recommended object feature information, user behavior feature information, and object click label; Inputting the user behavior characteristic information into an initial user interest information generation model to generate initial user interest information, wherein the initial user interest information is interest characteristic information of the user under at least one user interest point; Generate initial interest intensity information corresponding to the at least one user interest point using an initial interest intensity information generation model; generating initial object click information according to the recommended object feature information and the initial user interest information; According to the initial interest intensity information, the initial object click information and the object click label, the initial user interest information generation model and the initial interest intensity information generation model are trained to generate a user interest information generation model and an interest intensity information generation model.

2. The method according to claim 1, wherein: The generating initial object click information according to the recommended object feature information and the initial user interest information includes: Inputting the recommended object feature information into an initial recommended object feature encoding model to generate object feature encoding information; generating the initial object click information according to the object feature coding information and the initial user interest information; and The performing model training on the initial user interest information generation model and the initial interest intensity information generation model according to the initial interest intensity information, the initial object click information and the object click label to generate a user interest information generation model and an interest intensity information generation model, including: According to the initial interest intensity information, the initial object click information and the object click label, the initial user interest information generation model, the initial interest intensity information generation model and the initial recommended object feature encoding model are trained to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

3. The method according to claim 2, wherein: The user behavior characteristic information includes: at least one behavior characteristic information sequence corresponding to at least one behavior type; and The step of inputting the user behavior feature information into an initial user interest information generation model to generate initial user interest information includes: Using the initial user interest information generation model, each behavior feature information in the at least one behavior feature information sequence is vectorized to generate a behavior vector, thereby obtaining at least one behavior vector sequence; For each behavior type in the at least one behavior type, generating a key behavior vector corresponding to the behavior type by using the initial user interest information generation model; The initial user interest information is generated according to the obtained at least one key behavior vector.

4. The method according to claim 3, wherein: The generating a key behavior vector corresponding to the behavior type includes: In response to determining that the behavior type is a click behavior type, clustering the behavior vector sequence corresponding to the behavior type to generate a behavior vector cluster set; generating a key behavior vector corresponding to the click behavior type according to the behavior vector cluster center set in the behavior vector cluster set; In response to determining that the behavior type is a value-added behavior type, determining at least one first behavior vector whose time interval with the target time satisfies a preset time interval condition; combining the at least one first behavior vector to generate a key behavior vector corresponding to the value-added behavior type; In response to determining that the behavior type is an item inflow type, determining at least one second behavior vector corresponding to the item inflow number that satisfies a preset number condition; The at least one second behavior vector is combined to generate a key behavior vector corresponding to the item inflow type.

5. The method according to claim 3, wherein: The user behavior feature information further includes: a behavior content feature information sequence for the at least one behavior type; and The generating the initial user interest information according to the obtained at least one key behavior vector includes: Encoding the behavior content feature information in the behavior content feature information sequence to generate a behavior content vector sequence; Performing vector fusion on the at least one key behavior vector to generate a key behavior matrix for the at least one behavior type; The key behavior matrix and the behavior content vector sequence are input into the initial attention mechanism model to generate a similarity matrix, wherein the similarity matrix represents the correlation strength between the key behavior and the behavior content. The initial user interest information generation model includes: an initial attention mechanism model; The initial user interest information is generated according to the similarity matrix.

6. The method according to claim 5, wherein: The user behavior feature information further includes: a first behavior category feature information sequence for the at least one behavior type; and The generating the initial user interest information according to the similarity matrix includes: encoding the first behavior category feature information in the first behavior category feature information sequence to generate a behavior category vector sequence; Performing masking on the elements in the similarity matrix to generate a mask matrix; Generate interest point derived content information according to the mask matrix and the behavior content vector sequence; The mask matrix and the behavior category vector sequence are used to generate interest point derived category information; The initial user interest information is generated according to the interest point derived content information and the interest point derived category information.

7. The method according to claim 6, wherein: The generating the initial user interest information according to the interest point derived content information and the interest point derived category information includes: The interest point derived content information and the interest point derived category information are input into an initial interest information extraction model to generate the initial user interest information, wherein the initial user interest information generation model includes: the initial interest information extraction model.

8. The method according to claim 5, wherein: The generating model using the initial interest intensity information to generate the initial interest intensity information corresponding to the at least one user interest point includes: generating a second behavior category feature information sequence for the key behavior matrix; The second behavior category feature information sequence is input into the initial interest intensity information generation model to generate the initial interest intensity information.

9. The method according to claim 1, wherein: The performing model training on the initial user interest information generation model and the initial interest intensity information generation model according to the initial interest intensity information, the initial object click information and the object click label to generate a user interest information generation model and an interest intensity information generation model, including: Generate an initial interest strength label according to the recommended object feature information and the initial user interest information; generating interest intensity loss information for the initial interest intensity tag and the initial interest intensity information; Generate click loss information for the initial object click information and the object click label; According to the interest intensity loss information and the click loss information, model training is performed on the initial user interest information generation model, the initial interest intensity information generation model and the initial recommended object feature encoding model to generate a user interest information generation model, an initial interest intensity information generation model and a recommended object feature encoding model.

10. An object recommendation method, comprising: Obtaining user characteristic information corresponding to the target user as target user characteristic information; According to the target user feature information, using a pre-trained user interest information generation model, generating user interest information for the target user, wherein the user interest information generation model is generated based on the method according to any one of claims 1 to 9; Generating an initial recommended object information set for at least one user interest point according to the user interest information; Generate interest intensity information for the at least one user interest point using a pre-trained interest intensity information generation model, wherein the interest intensity information generation model is generated based on the method according to any one of claims 1 to 9; According to the interest strength information, a recommended object information set is screened out from the initial recommended object information set to perform object recommendation for the recommended object information set.

11. The method according to claim 10, wherein: The step of selecting a recommended object information set from the initial recommended object information set according to the interest strength information includes: For each user interest point in the at least one user interest point, the following first generating step is performed: Determine the initial recommended object information corresponding to the user's interest point in the initial recommended object information set to obtain at least one initial recommended object information; Determining the number of recommended objects corresponding to the user's interest point according to the interest strength information; Initial recommended object information whose corresponding recommendation similarity meets the preset similarity condition is filtered out from the at least one initial recommended object information as the recommended object information, and at least one recommended object information corresponding to the user interest point is obtained, wherein the number of information corresponding to the at least one recommended object information is the same as the number of recommended objects.

12. The method according to claim 11, wherein: The method further comprises: For each user interest point in the at least one user interest point, the following second generating step is performed: Determine a recall result corresponding to the user's interest point and a corresponding interest intensity value in the interest intensity information; The click rate corresponding to the user's interest point is determined according to the recall result and the interest strength value.

13. A model training device, comprising: A first acquisition unit is configured to acquire a target training sample, wherein the target training sample includes: feature information of a recommended object, feature information of a user behavior, and a click label of an object; A first generating unit is configured to input the user behavior feature information into an initial user interest information generating model to generate initial user interest information, wherein the initial user interest information is interest feature information of the user under at least one user interest point; A second generating unit is configured to generate initial interest intensity information corresponding to the at least one user interest point by using the initial interest intensity information generating model; A third generating unit is configured to generate initial object click information according to the recommended object feature information and the initial user interest information; The training unit is configured to perform model training on the initial user interest information generation model and the initial interest intensity information generation model according to the initial interest intensity information, the initial object click information and the object click label to generate a user interest information generation model and an interest intensity information generation model.

14. An object recommendation device, comprising: A second acquisition unit is configured to acquire user characteristic information corresponding to a target user as target user characteristic information; a fourth generating unit, configured to generate user interest information for the target user based on the target user feature information and using a pre-trained user interest information generating model, wherein the user interest information generating model is generated based on the method according to any one of claims 1 to 9; a fifth generating unit, configured to generate an initial recommended object information set for at least one user interest point according to the user interest information; a sixth generating unit, configured to generate interest intensity information for the at least one user interest point using a pre-trained interest intensity information generating model, wherein the interest intensity information generating model is generated based on the method according to any one of claims 1 to 9; The recommendation unit is configured to filter out a recommended object information set from the initial recommended object information set according to the interest strength information, so as to perform object recommendation for the recommended object information set.

15. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 12.

16. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

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

Images