Content recommendation model training method, content recommendation method and device

Abstract

This application discloses a content recommendation model training method, a content recommendation method, an apparatus, and an electronic device. The method includes: obtaining sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object; inputting the sample content attribute information, the sample object attribute information, and the sample interest attribute information into an initial content recommendation model to obtain sample behavior prediction information corresponding to the sample behavior content; calculating a target loss based on the sample behavior prediction information and the behavior label corresponding to the sample behavior content; and training the initial content recommendation model based on the target loss to obtain a target content recommendation model. The technical solution provided in the embodiments of this application can at least improve the content recommendation accuracy of the trained content recommendation model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a content recommendation model training method, a content recommendation method, and a content recommendation device. Background Art

[0002] Artificial intelligence software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning. Within machine learning, applications of artificial intelligence software technologies include personalized news recommendations.

[0003] Related personalized news recommendation model training technologies often only focus on users' behavioral information about their favorite content. The sample data used to train the content recommendation model is incomplete, resulting in large object bias when the trained personalized news recommendation model recommends news to the target, greatly affecting the recommendation accuracy of the personalized news recommendation model. Summary of the Invention

[0004] This application provides a content recommendation model training method, content recommendation method, and device, which can at least solve the problem in related technologies where the recommendation accuracy of trained content recommendation models does not meet expectations. The technical solutions of this application are as follows:

[0005] In one aspect, the present application provides a content recommendation model training method, the method comprising:

[0006] Obtaining sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object;

[0007] Inputting the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model for feature extraction to obtain sample object feature information, sample interest feature information, and sample content feature information;

[0008] Inputting the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction, thereby obtaining sample behavior prediction information corresponding to the sample behavior content;

[0009] Calculating target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label;

[0010] The initial content recommendation model is trained based on the target loss to obtain a target content recommendation model.

[0011] In another aspect, the present application provides a content recommendation method, comprising:

[0012] Obtaining content attribute information of each to-be-recommended content in the to-be-recommended content set and interest attribute information corresponding to the target object;

[0013] Inputting the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained by the content recommendation model training method described in any one of the first aspects above to perform feature extraction to obtain the interest feature information and the content feature information of each to-be-recommended content;

[0014] Inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining first behavior prediction information of the target object for each to-be-recommended content;

[0015] According to the first behavior prediction information of each to-be-recommended content, first target recommended content is determined from the to-be-recommended content set, and the first target recommended content is recommended to the target object.

[0016] On the other hand, the present application provides a content recommendation model training device, the device comprising:

[0017] an acquisition module, configured to acquire sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object;

[0018] a feature extraction module, configured to input the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model for feature extraction, thereby obtaining sample object feature information, sample interest feature information, and sample content feature information;

[0019] A behavior prediction module, configured to input the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction, and obtain sample behavior prediction information corresponding to the sample behavior content;

[0020] a calculation module, configured to calculate a target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label;

[0021] A training module is used to train the initial content recommendation model based on the target loss to obtain a target content recommendation model.

[0022] In another aspect, the present application provides a content recommendation device, comprising:

[0023] An acquisition module, configured to acquire content attribute information of each content to be recommended in the content set to be recommended and interest attribute information corresponding to the target object;

[0024] a feature extraction module configured to input the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained by the content recommendation model training method described in any one of the first aspects above, to perform feature extraction, thereby obtaining the interest feature information and the content feature information of each to-be-recommended content;

[0025] A behavior prediction module is configured to input the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining first behavior prediction information of the target subject with respect to each to-be-recommended content;

[0026] The recommendation module is configured to determine a first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, and recommend the first target recommended content to the target object.

[0027] In another aspect, the present application provides an electronic device, comprising:

[0028] processor;

[0029] a memory for storing instructions executable by the processor;

[0030] The processor is configured to execute the instructions to implement the content recommendation model training method or the content recommendation method as described above.

[0031] On the other hand, the present application provides a computer-readable storage medium, which, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the content recommendation model training method described above or implement the content recommendation method described above.

[0032] The content recommendation model training method provided in this application has the following technical effects:

[0033] By obtaining the sample content attribute information of the sample behavior content corresponding to the sample object, the behavior label corresponding to the sample behavior content, the sample object attribute information corresponding to the sample object, and the sample interest attribute information corresponding to the sample object, and using the aforementioned information to train the initial content recommendation model, the content recommendation ability of the trained target content recommendation model can be improved; by extracting the sample object feature information of the sample object attribute information, the sample interest feature information of the sample interest attribute information, and the sample content feature information of the sample content attribute information corresponding to the sample behavior content, and calculating the sample behavior prediction information corresponding to the sample behavior content based on the aforementioned information extracted, the sample situation can be better covered, which is conducive to improving the recommendation accuracy of the trained target content recommendation model.

[0034] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0036] Figure 1 is a schematic diagram of an application environment provided by an embodiment of the present application;

[0037] Figure 2 This is a flowchart of a content recommendation model training method provided in an embodiment of the present application;

[0038] Figure 3 This is a flowchart of a content recommendation method provided by an embodiment of the present application;

[0039] Figure 4 This is a structural diagram of a content recommendation model training device provided in an embodiment of the present application;

[0040] Figure 5 This is a structural diagram of a content recommendation device provided in an embodiment of the present application;

[0041] Figure 6 This is a model block diagram of the initial content recommendation model provided in an embodiment of the present application;

[0042] Figure 7 This is a model block diagram of the target content recommendation model provided in an embodiment of the present application;

[0043] Figure 8This is a block diagram of an electronic device for content recommendation model training or content recommendation provided by an embodiment of the present application;

[0044] Figure 9 This is a block diagram of another electronic device for content recommendation model training or content recommendation provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0046] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0047] See also Figure 1 , Figure 1 is a schematic diagram showing an application environment according to an exemplary embodiment. Figure 1 As shown, the application environment may include a server 01 and a terminal 02 .

[0048] In an optional embodiment, server 01 can be used to train a content recommendation model that can perform content recommendation. Specifically, server 01 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0049] In an optional embodiment, terminal 02 may perform content recommendation processing in conjunction with a content recommendation model trained by server 01. Specifically, terminal 02 may include, but is not limited to, electronic devices such as smartphones, desktop computers, tablet computers, laptop computers, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices. Optionally, the operating system running on the electronic device may include, but is not limited to, Android, iOS, Linux, and Windows.

[0050] In addition, it should be noted that Figure 1 What is shown is only one application environment provided by this application. In actual applications, other application environments may also be included. For example, the training of a target content recommendation model may also be implemented on the terminal 02.

[0051] In the embodiments of this specification, the server 01 and the terminal 02 may be connected directly or indirectly via wired or wireless communication, and this application does not impose any limitation thereto.

[0052] The following describes a content recommendation model training method for this application. Figure 2 It is a flow chart of a content recommendation model training method provided by an embodiment of the present application. This specification provides the method operation steps as described in the embodiment or flow chart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many steps, and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiment or the accompanying drawings. Specifically, Figure 2 As shown, the content recommendation model training method can be applied to electronic devices such as servers, terminals, and edge computing nodes. The method may include:

[0053] S201: Acquire sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object.

[0054] In the embodiments of this specification, the sample objects may be at least two objects using a content recommendation platform, and the at least two objects may include, but are not limited to, users and organizational structures using the content recommendation platform. The content recommendation platform may be a media content recommendation platform, including, but not limited to, a news recommendation platform, a music recommendation platform, and a video recommendation platform.

[0055] In the embodiments of this specification, the sample behavior content corresponding to the sample object may be historical sample behavior content that the content recommendation platform responds to the request of the terminal corresponding to the sample object and exposes (recommends) to the sample object, and may include, but is not limited to, historical news, historical music, and historical videos.

[0056] In the embodiments of this specification, the sample object attribute information corresponding to the sample object may include information that can characterize the object attributes of the sample object. Optionally, the sample object attribute information may include portrait attribute information and behavior attribute information of the sample object. Among them, the portrait attribute information may include, but is not limited to, age (19 years old), gender (female), and device number, and the behavior attribute information may include, but is not limited to, the historical click records and historical blocking records of the sample object.

[0057] In the embodiments of this specification, the sample interest attribute information corresponding to the sample object may include information that can characterize the interest attributes of the sample object. Optionally, the sample interest attribute information may include, but is not limited to, coarse-grained or fine-grained interest points such as title interest, topic interest, classification interest, and media interest, such as sports competitions (title interest), celebrity gossip (topic interest), and constellations (classification interest), etc.

[0058] In the embodiments of this specification, the sample content attribute information of the sample behavior content may include information that can characterize the theme of the sample behavior content. Optionally, the sample behavior content information may include, but is not limited to, the topics, tags, themes, titles, and classifications corresponding to the sample behavior content, such as "#Martial Arts Novel A#" (topic), "The Ten Big Pitfalls of Writers" (title), etc.

[0059] In the embodiments of this specification, the behavior label corresponding to the sample behavior content represents whether the sample object clicks on the sample behavior content. Among them, the sample behavior content clicked by the sample object is positive sample behavior content, and its behavior label is 1; the sample behavior content not clicked by the sample object is negative sample behavior content, and its behavior label is 0.

[0060] In practical applications, the content recommendation platform responds to the requests of the terminals corresponding to at least two sample objects, and exposes a certain number (for example, 500 pieces) of recommended contents to at least two objects, so that at least two objects can click on the recommended contents according to at least one of their interests. In the case where the content recommendation platform responds to the request of any object again, the above at least two objects are used as sample objects, the recommended contents exposed by the content recommendation platform to the sample objects are used as the sample behavior content corresponding to the sample objects, and the above at least one interest point is used as at least one historical interest point of the sample objects.

[0061] In an optional embodiment, in order to obtain the sample behavior content corresponding to the sample object, the sample object includes multiple objects, and the above method further includes:

[0062] Obtain at least one historical sample behavior content corresponding to each object;

[0063] Obtaining sample content attribute information of each historical sample behavior content;

[0064] Based on the sample object attribute information, sample interest attribute information and corresponding sample content attribute information of each object, at least one historical sample behavior content corresponding to each object is screened to obtain the sample behavior content corresponding to each object.

[0065] In an optional embodiment, the filtering of at least one historical sample behavior content corresponding to each object based on the sample object attribute information, sample interest attribute information, and corresponding sample content attribute information of each object can be achieved by:

[0066] From at least one historical sample behavior content corresponding to each object, historical sample behavior content having content attribute information that is at least partially identical, similar, or associated with the sample object attribute information and sample interest attribute information of each object is filtered out.

[0067] For example, the sample object attribute information for object C includes age 19, gender female, and martial arts novel enthusiast. The sample interest attribute information corresponding to object C includes martial arts novel author B, books, and Korean dramas. The content attribute information for a historical sample behavior content corresponding to object C includes martial arts novel A and TV series. It can be understood that since martial arts enthusiast, martial arts novel author B, and martial arts novel A are related attribute information, this historical sample behavior content can be used as the sample behavior content corresponding to object C.

[0068] In the embodiments of this specification, by screening at least one historical sample behavior content corresponding to each object in the sample objects, the validity of the sample behavior content corresponding to the obtained sample behavior objects can be ensured, thereby improving the training efficiency of the content recommendation model.

[0069] In the embodiments of this specification, by obtaining sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object, an initial content recommendation model is trained using the aforementioned information to improve the content recommendation capability of a trained target content recommendation model.

[0070] S203: Input the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in the initial content recommendation model for feature extraction to obtain sample object feature information, sample interest feature information, and sample content feature information.

[0071] In the embodiments of this specification, the feature extraction network may include an object feature extraction network, an interest feature extraction network, and a content feature extraction network.

[0072] In the embodiments of this specification, feature extraction methods may include but are not limited to sum-pooling, average-pooling, and gated network methods.

[0073] In a specific embodiment, inputting the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model for feature extraction includes:

[0074] Inputting the sample object attribute information into the object feature extraction network to perform feature extraction to obtain the sample object feature information;

[0075] Inputting the sample interest attribute information into the interest feature extraction network to perform feature extraction to obtain the sample interest feature information;

[0076] The sample content attribute information is input into the content feature extraction network to perform feature extraction to obtain the sample content feature information.

[0077] In an embodiment of the present specification, the object feature extraction network may include an input layer, an intermediate layer, and an output layer; wherein the parameters of the input layer are the weights of each sample object attribute in the sample object attribute information, and the weights of the above-mentioned sample object attributes are determined based on the importance of each object attribute; wherein the intermediate layer is used to extract features of the sample object attribute information according to the parameters of the input layer; wherein the output layer is used to output the extracted sample object feature information.

[0078] Specifically, the parameters of the input layer can be represented by vector embedding. For example, the sample object attributes include age 5 (weight 0.5), gender female (weight 0.3), and occupation writer (weight 0.2). The above sample object attributes can be represented in vector form as follows: and Perform vector concatenation on the three vectors above to obtain That is, the parameters of the input layer are expressed in the form of embedding. This representation makes it easier for the middle layer to extract the feature information of the sample object based on the parameters of the input layer.

[0079] Specifically, the middle layer extracts the features of the sample object attribute information according to the parameters of the input layer, obtains the initial sample object feature information, and transmits the initial sample object feature information represented in the form of embedding to the output layer.

[0080] The output layer projects the initial sample object feature information, represented in the form of embeddings, into the target dimensional space, obtaining the sample object feature information represented in the form of embeddings. Representing the sample object feature information in the form of embeddings facilitates the processing of the sample object feature information by the behavior prediction network.

[0081] In an embodiment of the present specification, the interest feature extraction network may include an input layer, an intermediate layer, and an output layer; wherein the parameters of the input layer are the weights of each sample interest attribute (sample interest point) in the sample interest attribute information, and the weights of each sample interest attribute are determined based on the importance of each sample interest attribute.

[0082] Specifically, the parameters of the input layer can be expressed in embedding form. The intermediate layer extracts features from the sample interest attribute information based on the input layer parameters and transmits the obtained initial sample interest feature information expressed in embedding form to the output layer. The output layer maps the initial sample interest feature information expressed in embedding form to the target dimensional space to obtain sample interest feature information expressed in embedding form. Expressing the input layer parameters in embedding form facilitates the intermediate layer to extract features from the sample interest attribute information based on the input layer parameters. Expressing the sample interest feature information in embedding form facilitates the processing of the sample interest feature information by the behavior prediction network.

[0083] In an embodiment of the present specification, the content feature extraction network may include an input layer, an intermediate layer, and an output layer; wherein the parameters of the input layer are the weights of each sample content attribute in the sample interest attribute information, and the weights of each sample content attribute are determined based on the importance of each sample content attribute.

[0084] Specifically, the parameters of the input layer can be expressed in embedding form. The intermediate layer extracts features from the sample content attribute information based on the input layer parameters and transmits the obtained initial sample content feature information expressed in embedding form to the output layer. The output layer maps the initial sample content feature information expressed in embedding form to the target dimensional space to obtain sample content feature information expressed in embedding form. Expressing the input layer parameters in embedding form facilitates the intermediate layer to extract features from the sample content attribute information based on the input layer parameters. Expressing the sample content feature information in embedding form also facilitates the processing of the sample content feature information by the behavior prediction network.

[0085] In the embodiments of this specification, various information obtained during the training process is represented in the form of embedding, which can simplify information processing, cover more information, and enhance the application value of the trained target content recommendation model.

[0086] In some embodiments, the implementation methods of the sample object feature extraction network, the sample interest feature extraction network and the content feature extraction network include but are not limited to fully connected (forward connected) neural networks, convolutional neural networks and variants, recurrent neural networks and variants, transformer networks (transformer), recursive neural networks and bidirectional neural networks and other network structures.

[0087] In some embodiments, the sample object feature extraction network, the sample interest feature extraction network, and the content feature extraction network may be networks that employ a self-attention mechanism.

[0088] In the embodiments of this specification, the sample object feature extraction network, the sample interest feature extraction network, and the content feature extraction network each have at least one intermediate layer. It will be appreciated that the greater the number of intermediate layers, the higher the computational accuracy and complexity. The embodiments of this specification do not limit the number of intermediate layers; in actual applications, the number of intermediate layers used in the network will depend on actual needs.

[0089] In the embodiments of this specification, by extracting sample object feature information of sample object attribute information, sample interest feature information of sample interest attribute information, and sample content feature information of sample content attribute information corresponding to sample behavior content, the sample behavior prediction information corresponding to the sample behavior content is calculated based on the aforementioned extracted feature information, which can better cover the sample situation and is conducive to improving the content recommendation accuracy of the trained target content recommendation model.

[0090] S205: Inputting the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction, and obtaining sample behavior prediction information corresponding to the sample behavior content.

[0091] In the embodiment of this specification, the behavior prediction network can calculate the semantic similarity of different feature information, including the semantic similarity between the sample object feature information and the sample interest feature information, and the semantic similarity between the sample interest feature information and the sample content feature information.

[0092] In the embodiment of this specification, the sample behavior prediction information represents the probability of the sample object predicted by the behavior prediction network clicking on the corresponding sample behavior content.

[0093] In a specific embodiment, inputting the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction, and obtaining sample behavior prediction information corresponding to the sample behavior content, includes:

[0094] Inputting the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model, calculating the semantic similarity between the sample object feature information and the sample interest feature information, and obtaining a first sample similarity;

[0095] Calculating the semantic similarity between the sample interest feature information and the sample content feature information to obtain a second sample similarity;

[0096] Determine sample behavior prediction information corresponding to the sample behavior content according to the sum of the first sample similarity and the second sample similarity.

[0097] In the embodiment of this specification, a method for calculating the first sample similarity and the second sample similarity may include, but is not limited to, a cosine distance formula, a Euclidean distance formula, and a Manhattan distance formula.

[0098] In practical applications, the sample object feature information, sample interest feature information, and sample content feature information represented in embedding form are mapped to the same dimensional space (target dimensional space). Based on this, preferably, the first sample similarity can be obtained by calculating the cosine distance between the sample object feature information and the sample interest feature information. Preferably, the second sample similarity can be obtained by calculating the cosine distance between the sample interest feature information and the sample content feature information.

[0099] It can be understood that the first sample similarity represents the degree of preference of the sample object predicted by the behavior prediction network for the sample interest attribute information, and the second sample similarity represents the similarity between the sample interest attribute information predicted by the prediction network and the sample content attribute information.

[0100] In an optional embodiment, determining the sample behavior prediction information corresponding to the sample behavior content according to the sum of the first sample similarity and the second sample similarity includes:

[0101] The sum of the similarity of the first sample and the similarity of the second sample is calculated, and the above sum is used as the independent variable of the activation function to obtain behavior prediction information.

[0102] In the embodiments of this specification, the activation function may include but is not limited to a sigmoid function, a tanh function, a Relu activation function, and an ELU activation function.

[0103] In an optional embodiment, the sum of the first sample similarity and the second sample similarity is calculated, and the sum is used as the independent variable of the activation function to obtain the behavior prediction information, which can be combined with the following formula:

[0104] s2=s(u,t)+s(t,d);

[0105] p = sigmoid(s2);

[0106] Among them, s2 is the sum of the first sample similarity and the second sample similarity, s(u,t) is the first sample similarity, s(t,d) is the second sample similarity, and p is the behavior prediction information.

[0107] Specifically, the derivation process of the above formula is:

[0108] Based on the product of the first sample similarity and the second sample similarity, the behavior prediction information corresponding to the sample behavior content in a narrow sense can be calculated. The specific formula is as follows:

[0109] s1=s(u,t)×s(t,d);

[0110] p=sigmoid(s1).

[0111] According to the monotonically increasing characteristic of the exponential function of e, the exponent of e is taken for the first sample similarity s(u,t) to obtain e s(u ,t) , take the exponent of e for the similarity of the second sample to get e s(t,d) It can be understood that s(u,t)×s(t,d) and e s(u,t) ×e s(t,d) The monotonicity of is consistent, and e s(u,t) ×e s(t,d) =es(u,t)+s(t,d) .

[0112] From this, we can see that in terms of data monotonicity, s(u,t)×s(t,d) is equivalent to s(u,t)+s(t,d), and the data growth rate of s(u,t)+s(t,d) is lower than that of s(u,t)×s(t,d).

[0113] Based on this, the embodiment of this specification calculates behavior prediction information based on s(u,t)+s(t,d), which can cover more sample situations, greatly reduce the prediction deviation of the trained content recommendation model on the behavior of the object clicking on the recommended content, and improve the accuracy of content recommendation.

[0114] S107: Calculate target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label.

[0115] In the embodiment of this specification, the target loss represents the difference between the probability of the sample behavior content corresponding to the sample object click predicted by the behavior prediction network and the sample behavior label corresponding to the sample behavior content (actual click or no click); wherein, the actual click is recorded as 1 and the actual no click is recorded as 0.

[0116] In the embodiments of this specification, the loss function used to calculate the target loss may include but is not limited to a negative log-likelihood loss function, a cross-entropy loss function, an exponential loss function, and a square loss function.

[0117] Preferably, in the embodiments of this specification, a cross entropy loss function is used to process the sample behavior prediction information and behavior labels corresponding to the sample behavior content to calculate the target loss.

[0118] In practical applications, the sample interest feature information includes a sample interest feature corresponding to at least one sample interest point; the second sample similarity includes the semantic similarity between the sample interest feature of each sample interest point and the sample behavior content feature information. Based on the first similarity and the semantic similarity between the sample interest feature of each sample interest point and the sample behavior content feature information, the sample behavior sub-prediction information corresponding to each sample interest point can be calculated in the scheme for calculating the sample behavior prediction information.

[0119] Specifically, the calculating the target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label includes:

[0120] According to the sample behavior sub-prediction information corresponding to each sample interest point and the label of the sample behavior content, the target sub-loss (L i );

[0121] Determine the target sub-loss (L i) is the target loss (L).

[0122] Specifically, according to the calculation target sub-loss (L i ) is used to calculate the target loss (L), which can be referred to Formula 3:

[0123]

[0124] Among them, N is the number of sample interest points, i represents the current traversal to the i-th sample interest point, L i is the target sub-loss corresponding to the currently traversed sample interest point, p i is the sample behavior prediction sub-information corresponding to the i-th sample interest point, y i It is the behavior label corresponding to the sample behavior content. A value of 1 indicates an actual click, and a value of 0 indicates no actual click.

[0125] In the embodiments of this specification, in the scheme for calculating the target loss corresponding to the sample behavior content, by traversing each sample interest point, calculating the target sub-loss corresponding to each sample interest point, and calculating the target loss based on the target sub-loss corresponding to each sample interest point, the calculation accuracy of the target loss can be improved, so as to more effectively train the content recommendation model based on the higher-precision target loss.

[0126] In the embodiments of this specification, by calculating the target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label, the prediction deviation of the initial content recommendation model for the behavior of the sample object clicking on the sample behavior content can be effectively determined, which facilitates adjusting the model parameters of the initial content recommendation model according to the prediction deviation to obtain a content recommendation model that can be used for content recommendation.

[0127] S109: Training the initial content recommendation model based on the target loss to obtain a target content recommendation model.

[0128] In the embodiment of this specification, based on the target loss, the model parameters of the initial content recommendation model are updated until the target loss meets the preset conditions, the training is stopped, and the current target content recommendation model is used as the target content recommendation model.

[0129] Wherein, the model parameters include network parameters of the object feature extraction network, the interest feature extraction network and the content feature extraction network; wherein, the network parameters of the object feature extraction network include parameters of the input layer, parameters of the intermediate layer and parameters of the output layer of the object feature extraction network, the network parameters of the interest feature extraction network include parameters of the input layer, parameters of the intermediate layer and parameters of the output layer of the interest feature extraction network, and the network parameters of the content feature extraction network include parameters of the input layer, parameters of the intermediate layer and parameters of the output layer of the content feature extraction network.

[0130] In a specific embodiment, the training of the initial content recommendation model based on the target loss to obtain the target content recommendation model includes:

[0131] When the target loss does not meet a preset condition, updating network parameters of the object feature extraction network, the interest feature extraction network, and the content feature extraction network;

[0132] The target loss is updated based on the network parameters of the updated object feature extraction network, the interest feature extraction network, and the content feature extraction network until the target loss meets the preset conditions, and the current content recommendation model is used as the target content recommendation model.

[0133] In the embodiment of this specification, the preset condition may include: the target loss is less than a preset first threshold or the number of training times reaches a preset second threshold.

[0134] In an embodiment of the present specification, the target content recommendation model is obtained by training the initial content recommendation model based on the target loss, and the training is stopped when the target loss meets the preset conditions, thereby ensuring that the content recommendation ability of the target content recommendation model meets expectations, and training the network parameters of the three feature extraction networks in the initial content recommendation model based on the target loss, thereby realizing multi-task learning of the initial content recommendation model.

[0135] The block diagram of the initial content recommendation model involved in the above solution can be referred to Figure 6 .

[0136] It can be seen from the technical solutions provided in the above embodiments of this specification that, in the embodiments of this specification, by obtaining the sample content attribute information of the sample behavior content corresponding to the sample object, the behavior label corresponding to the sample behavior content, the sample object attribute information corresponding to the sample object, and the sample interest attribute information corresponding to the sample object, rich training samples are obtained, which can improve the prediction ability of the trained target content recommendation model for recommended content; by performing feature extraction on the sample object attribute information, the sample interest attribute information, and the sample content attribute information, and performing behavior prediction based on the extracted sample object feature information, the sample feature attribute information, and the sample feature attribute information, behavior prediction information covering more sample situations can be obtained, which can improve the global prediction ability of the trained target content recommendation model; by calculating the target loss based on the behavior prediction information and the behavior label corresponding to the sample behavior content, the prediction deviation of the content recommendation model during the training process can be accurately obtained; by training the initial content recommendation model based on the target loss, and stopping the training when the target loss meets the preset conditions to obtain the target content recommendation model, it can ensure that the content recommendation model is fully trained, thereby improving the prediction ability of the target content recommendation model for recommended content.

[0137] Based on the embodiment of the content recommendation model training method described above, the following is a flow chart of a content recommendation method provided by this application, with reference to Figure 3 ,This method can be applied to electronic devices such as servers, terminals, edge computing nodes, etc., and includes the following steps.

[0138] S301: Obtain content attribute information of each content to be recommended in the content set to be recommended and interest attribute information corresponding to the target object.

[0139] In the embodiments of this specification, the content to be recommended may be the content that the content recommendation platform needs to recommend to the target object. The content attribute information of the content to be recommended may include information that can characterize the subject of the content to be recommended. The content attribute information may include but is not limited to the topic, title, theme and category of the content.

[0140] S303: Input the interest attribute information and the content attribute information of each content to be recommended into the feature extraction network of the target content recommendation model trained according to the above-mentioned content recommendation model training method to perform feature extraction, and obtain interest feature information and the content feature information of each content to be recommended.

[0141] Specifically, the feature extraction step includes:

[0142] Inputting the interest attribute information into the interest feature extraction network to perform feature extraction to obtain the interest feature information;

[0143] The content attribute information of each content to be recommended is input into the content feature extraction network for feature extraction to obtain the content feature information of each content to be recommended.

[0144] S305: Inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, and obtaining first behavior prediction information of the target object for each to-be-recommended content.

[0145] Specifically, the step of acquiring the first behavior prediction information includes:

[0146] Inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model, calculating the language similarity between the interest feature information and the content feature information of each to-be-recommended content, and obtaining a second similarity corresponding to each to-be-recommended content;

[0147] The second similarity corresponding to each to-be-recommended content is determined as first behavior prediction information of the target object with respect to each to-be-recommended content.

[0148] S307: Determine first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, and recommend the first target recommended content to the target object.

[0149] In the embodiment of this specification, the first behavior prediction information represents the probability of the target object clicking on each content to be recommended.

[0150] Specifically, the step of determining the first target recommended content includes:

[0151] Determine the recommendation priority of each to-be-recommended content according to the first behavior prediction information of each to-be-recommended content, sort each to-be-recommended content according to the recommendation priority, and determine the to-be-recommended content ranked in the top K positions as the first target recommended content.

[0152] In an optional embodiment, the method further includes:

[0153] Obtaining object attribute information corresponding to the target object;

[0154] Inputting the object attribute information into the feature extraction network of the target content recommendation model to perform feature extraction to obtain object feature information;

[0155] Inputting the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into a behavior prediction network of a target content recommendation model to perform behavior prediction, thereby obtaining second behavior prediction information of the target object with respect to each to-be-recommended content;

[0156] After determining the first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, the method further includes:

[0157] According to the second behavior prediction information, second target recommended content is determined from the first target recommended content, and the second target recommended content is recommended to the target object.

[0158] Specifically, the feature extraction step includes:

[0159] Inputting the object attribute information into the feature extraction network of the target content recommendation model to perform feature extraction to obtain object feature information;

[0160] Inputting the interest attribute information into the interest feature extraction network to perform feature extraction to obtain the interest feature information;

[0161] The content attribute information of each content to be recommended is input into the content feature extraction network for feature extraction to obtain the content feature information of each content to be recommended.

[0162] Specifically, the step of obtaining the second behavior prediction information includes:

[0163] Inputting the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model, calculating the semantic similarity between the object feature information and the interest feature information to obtain a first similarity; calculating the semantic similarity between the interest feature information and the content feature information of each to-be-recommended content, to obtain a second similarity corresponding to each to-be-recommended content;

[0164] The product of the first similarity and the second similarity corresponding to each to-be-recommended content is determined as the second behavior prediction information corresponding to each to-be-recommended content.

[0165] The block diagram of the target content recommendation model involved in the above scheme can be referred to Figure 7 .

[0166] In the embodiments of this specification, the target content recommendation model obtained by training according to the above-mentioned content recommendation model training method can predict whether the target user will click on the recommended content in the recommended content set, thereby accurately inferring the target user's preference for the recommended content, ensuring that the target recommended content meets the user's expectations, and improving the target user's experience in using the content recommendation platform.

[0167] Figure 4 This is a structural diagram of a content recommendation model training device provided in an embodiment of the present application. Figure 4, the device comprises:

[0168] An acquisition module 401 is configured to acquire sample content attribute information of a sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object;

[0169] A feature extraction module 403 is configured to input the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model to perform feature extraction, thereby obtaining sample object feature information, sample interest feature information, and sample content feature information;

[0170] The behavior prediction module 405 is configured to input the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction and obtain sample behavior prediction information corresponding to the sample behavior content;

[0171] A calculation module 407 is configured to calculate a target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label;

[0172] The training module 409 is configured to train the initial content recommendation model based on the target loss to obtain a target content recommendation model.

[0173] Optionally, the behavior prediction module 405 includes:

[0174] a first sample similarity calculation unit, configured to input the sample object feature information, the sample interest feature information, and the sample content feature information into a behavior prediction network of the initial content recommendation model, calculate the semantic similarity between the sample object feature information and the sample interest feature information, and obtain a first sample similarity;

[0175] a second sample similarity calculation unit, calculating the semantic similarity between the sample interest feature information and the sample content feature information to obtain a second sample similarity;

[0176] A behavior prediction unit is configured to determine sample behavior prediction information corresponding to the sample behavior content according to the sum of the first sample similarity and the second sample similarity.

[0177] Optionally, the feature extraction module 403 includes:

[0178] an object feature extraction unit, configured to input the sample object attribute information into the object feature extraction network for feature extraction to obtain the sample object feature information;

[0179] an interest feature extraction unit, configured to input the sample interest attribute information into the interest feature extraction network for feature extraction to obtain the sample interest feature information;

[0180] The content feature extraction unit is configured to input the sample content attribute information into the content feature extraction network to perform feature extraction to obtain the sample content feature information.

[0181] Optionally, the training module 409 includes:

[0182] an updating unit, configured to update network parameters of the object feature extraction network, the interest feature extraction network, and the content feature extraction network when the target loss does not meet a preset condition;

[0183] A training unit is used to update the target loss based on the network parameters of the updated object feature extraction network, the interest feature extraction network, and the content feature extraction network until the target loss meets the preset conditions, and use the current content recommendation model as the target content recommendation model.

[0184] Optionally, the acquisition module 401 includes:

[0185] A first acquisition unit is configured to acquire at least one historical sample behavior content corresponding to each object;

[0186] A second acquiring unit is used to acquire sample content attribute information of each historical sample behavior content;

[0187] The screening unit is configured to screen at least one historical sample behavior content corresponding to each object based on the sample object attribute information, sample interest attribute information and corresponding sample content attribute information of each object to obtain the sample behavior content corresponding to each object.

[0188] The device and method embodiments in the device embodiments are based on the same inventive concept.

[0189] Figure 5 This is a structural diagram of a content recommendation device provided in an embodiment of the present application. Figure 5 , the device comprises:

[0190] An acquisition module 501 is used to acquire content attribute information of each content to be recommended in the content set to be recommended and interest attribute information corresponding to the target object;

[0191] A feature extraction module 503 is configured to input the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained by the content recommendation model training method described in any one of the first aspects above, to perform feature extraction, thereby obtaining interest feature information and the content feature information of each to-be-recommended content;

[0192] A behavior prediction module 505 is configured to input the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining first behavior prediction information of the target subject with respect to each to-be-recommended content;

[0193] The recommendation module 507 is configured to determine a first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, and recommend the first target recommended content to the target object.

[0194] Optionally, the acquisition module 501 is further configured to acquire object attribute information corresponding to the target object;

[0195] Optionally, the feature extraction module 503 is further configured to input the object attribute information into a feature extraction network of the target content recommendation model to perform feature extraction to obtain object feature information;

[0196] Optionally, the behavior prediction module 505 is further configured to input the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into a behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining second behavior prediction information of the target object with respect to each to-be-recommended content;

[0197] Optionally, the recommendation module 507 is further configured to determine second target recommended content from the first target recommended content according to the second behavior prediction information, and recommend the second target recommended content to the target object.

[0198] Optionally, the feature extraction module 503 includes:

[0199] an interest feature extraction unit, configured to input the interest attribute information into the interest feature extraction network for feature extraction to obtain the interest feature information;

[0200] The content feature extraction unit is used to input the content attribute information of each content to be recommended into the content feature extraction network to perform feature extraction to obtain the content feature information of each content to be recommended.

[0201] Optionally, the feature extraction module 503 further includes:

[0202] The object feature extraction unit is used to input the object attribute information into the object feature extraction network to perform feature extraction to obtain the object feature information.

[0203] Optionally, the behavior prediction module 505 includes:

[0204] a second similarity calculation unit, configured to input the interest feature information and the content feature information of each to-be-recommended content into a behavior prediction network of the target content recommendation model, calculate the semantic similarity between the interest feature information and the content feature information of each to-be-recommended content, and obtain a second similarity corresponding to each to-be-recommended content;

[0205] The first behavior prediction unit is configured to determine the second similarity corresponding to each to-be-recommended content as first behavior prediction information of the target object with respect to each to-be-recommended content.

[0206] Optionally, the behavior prediction module 505 further includes:

[0207] a first similarity calculation unit, configured to input the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into a behavior prediction network of the target content recommendation model, calculate the semantic similarity between the object feature information and the interest feature information, and obtain a first similarity;

[0208] The second similarity calculation unit is further configured to calculate the language similarity between the interest feature information and the content feature information of each to-be-recommended content, to obtain a second similarity corresponding to each to-be-recommended content;

[0209] The second behavior prediction unit is configured to determine a product of the first similarity and the second similarity corresponding to each content to be recommended as second behavior prediction information corresponding to each content to be recommended.

[0210] The device and method embodiments in the device embodiments are based on the same inventive concept.

[0211] An embodiment of the present application also provides a computer-readable storage medium. When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the content recommendation model training method described above or implement the content recommendation method described above.

[0212] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0213] Figure 8 This is a block diagram of an electronic device for content recommendation model training or content recommendation according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 8 As shown. The electronic device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for determining an image defect filling network or processing an image defect is implemented. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the electronic device, or an external keyboard, touchpad or mouse.

[0214] Figure 9 This is a block diagram of an electronic device for content recommendation model training or content recommendation according to an exemplary embodiment. The electronic device may be a server, and its internal structure diagram may be as shown in FIG. Figure 9 As shown. The electronic device includes a processor, a memory and a network interface connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a content recommendation model training or a content recommendation method.

[0215] Those skilled in the art will understand that Figure 8 and Figure 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0216] In an exemplary embodiment, an electronic device is also provided, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the content recommendation model training method or content recommendation method as described in the embodiments of this specification.

[0217] In an exemplary embodiment, a storage medium is also provided. When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the content recommendation model training method or content recommendation method in the embodiment of this specification.

[0218] In an exemplary embodiment, a computer program product including instructions is also provided. When the computer program product is run on a computer, the computer is caused to execute the content recommendation model training method or the content recommendation method in the embodiments of this specification.

[0219] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, which can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0220] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0221] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

[0222] An embodiment of the present application also provides a storage medium, which can be set in a server to store at least one instruction, at least one program, code set or instruction set for implementing a content recommendation model training method or a content recommendation method in a method embodiment. The at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement the content recommendation model training method or content recommendation processing method provided by the above-mentioned method embodiment.

[0223] Optionally, in the embodiments of this specification, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0224] It can be seen from the embodiments of the content recommendation model training method, content recommendation method, device, server or storage medium provided by the present application that, in the present application, by obtaining the sample content attribute information of the sample behavior content corresponding to the sample object, the behavior label corresponding to the sample behavior content, the sample object attribute information corresponding to the sample object and the sample interest attribute information corresponding to the sample object, and using the obtained aforementioned information to train the initial content recommendation model, the content recommendation ability of the trained target content recommendation model can be improved; by extracting the sample object feature information of the sample object attribute information, the sample interest feature information of the sample interest attribute information and the sample content feature information of the sample content attribute information corresponding to the sample behavior content, and calculating the sample behavior prediction information corresponding to the sample behavior content based on the extracted aforementioned information, the sample situation can be better covered, which is conducive to improving the recommendation accuracy of the trained target content recommendation model.

[0225] It should be noted that the order of the embodiments of the present application described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. The above description is of specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0226] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. In particular, the device embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0227] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0228] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A content recommendation model training method, characterized in that: The method comprises: Obtaining sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object; the sample object includes multiple objects, and the sample interest attribute information includes at least one interest point corresponding to each object; Inputting the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model for feature extraction, thereby obtaining sample object feature information, sample interest feature information, and sample content feature information; the feature extraction network includes an object feature extraction network, an interest feature extraction network, and a content feature extraction network; Inputting the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model, calculating the semantic similarity between the sample object feature information and the sample interest feature information, and obtaining a first sample similarity; Calculating the semantic similarity between the sample interest feature information and the sample content feature information to obtain a second sample similarity; Determining sample behavior prediction information corresponding to the sample behavior content according to the sum of the first sample similarity and the second sample similarity; Calculating target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label; The initial content recommendation model is trained based on the target loss to obtain a target content recommendation model.

2. The method according to claim 1, characterized in that The step of inputting the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model to perform feature extraction to obtain sample object feature information, sample interest feature information, and sample content feature information includes: Inputting the sample object attribute information into the object feature extraction network to perform feature extraction to obtain the sample object feature information; Inputting the sample interest attribute information into the interest feature extraction network to perform feature extraction to obtain the sample interest feature information; The sample content attribute information is input into the content feature extraction network to perform feature extraction to obtain the sample content feature information.

3. The method according to claim 1, characterized in that The step of training the initial content recommendation model based on the target loss to obtain a target content recommendation model includes: When the target loss does not meet a preset condition, updating network parameters of the object feature extraction network, the interest feature extraction network, and the content feature extraction network; The target loss is updated based on the network parameters of the updated object feature extraction network, the interest feature extraction network, and the content feature extraction network until the target loss meets the preset conditions, and the current content recommendation model is used as the target content recommendation model.

4. The method according to claim 1, wherein The sample object includes a plurality of objects, and the method further includes: Obtain at least one historical sample behavior content corresponding to each object; Obtaining sample content attribute information of each historical sample behavior content; At least one historical sample behavior content corresponding to each object is screened based on the sample object attribute information, the sample interest attribute information and the corresponding sample content attribute information of each object to obtain the sample behavior content corresponding to each object.

5. A content recommendation method, characterized in that: The method comprises: Obtaining content attribute information of each to-be-recommended content in the to-be-recommended content set and interest attribute information corresponding to the target object; Inputting the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained by any content recommendation model training method according to claims 1 to 4 to perform feature extraction to obtain the interest feature information and the content feature information of each to-be-recommended content; Inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining first behavior prediction information of the target object for each to-be-recommended content; According to the first behavior prediction information of each to-be-recommended content, first target recommended content is determined from the to-be-recommended content set, and the first target recommended content is recommended to the target object.

6. The method according to claim 5, characterized in that The method further comprises: Obtaining object attribute information corresponding to the target object; Inputting the object attribute information into the feature extraction network of the target content recommendation model to perform feature extraction to obtain object feature information; Inputting the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into a behavior prediction network of a target content recommendation model to perform behavior prediction, thereby obtaining second behavior prediction information of the target object with respect to each to-be-recommended content; After determining the first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, the method further includes: According to the second behavior prediction information, second target recommended content is determined from the first target recommended content, and the second target recommended content is recommended to the target object.

7. The method according to claim 5, characterized in that The step of inputting the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained according to any one of the content recommendation model training methods of claims 1 to 4 to perform feature extraction, thereby obtaining the interest feature information and the content feature information of each to-be-recommended content, comprises: Inputting the interest attribute information into the interest feature extraction network to perform feature extraction to obtain the interest feature information; The content attribute information of each content to be recommended is input into the content feature extraction network for feature extraction to obtain the content feature information of each content to be recommended.

8. The method according to claim 6, characterized in that The step of inputting the object attribute information into the feature extraction network of the target content recommendation model to extract features and obtain object feature information includes: The object attribute information is input into the object feature extraction network to perform feature extraction to obtain the object feature information.

9. The method according to claim 5, characterized in that The step of inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, and obtaining first behavior prediction information of the target object for each to-be-recommended content, includes: Inputting the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model, calculating the semantic similarity between the interest feature information and the content feature information of each to-be-recommended content, and obtaining a second similarity corresponding to each to-be-recommended content; The second similarity corresponding to each to-be-recommended content is determined as first behavior prediction information of the target object with respect to each to-be-recommended content.

10. The method according to claim 6, characterized in that The second behavior prediction information is obtained by: Inputting the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model, calculating the semantic similarity between the object feature information and the interest feature information to obtain a first similarity; calculating the semantic similarity between the interest feature information and the content feature information of each to-be-recommended content, to obtain a second similarity corresponding to each to-be-recommended content; The product of the first similarity and the second similarity corresponding to each to-be-recommended content is determined as the second behavior prediction information corresponding to each to-be-recommended content.

11. A content recommendation model training device, characterized in that: The device comprises: an acquisition module, configured to acquire sample content attribute information of sample behavior content corresponding to a sample object, a behavior label corresponding to the sample behavior content, sample object attribute information corresponding to the sample object, and sample interest attribute information corresponding to the sample object; the sample object includes a plurality of objects, and the sample interest attribute information includes at least one interest point corresponding to each object; a feature extraction module, configured to input the sample content attribute information, the sample object attribute information, and the sample interest attribute information into a feature extraction network in an initial content recommendation model for feature extraction, thereby obtaining sample object feature information, sample interest feature information, and sample content feature information; the feature extraction network includes an object feature extraction network, an interest feature extraction network, and a content feature extraction network; A behavior prediction module, configured to input the sample object feature information, the sample interest feature information, and the sample content feature information into the behavior prediction network of the initial content recommendation model to perform behavior prediction, and obtain sample behavior prediction information corresponding to the sample behavior content; a calculation module, configured to calculate a target loss based on the sample behavior prediction information corresponding to the sample behavior content and the behavior label; A training module, configured to train the initial content recommendation model based on the target loss to obtain a target content recommendation model; The behavior prediction module includes: a first sample similarity calculation unit, configured to input the sample object feature information, the sample interest feature information, and the sample content feature information into a behavior prediction network of the initial content recommendation model, calculate the semantic similarity between the sample object feature information and the sample interest feature information, and obtain a first sample similarity; A second sample similarity calculation unit is used to calculate the semantic similarity between the sample interest feature information and the sample content feature information to obtain a second sample similarity; A behavior prediction unit is configured to determine sample behavior prediction information corresponding to the sample behavior content according to the sum of the first sample similarity and the second sample similarity.

12. The device according to claim 11, characterized in that The feature extraction module includes: an object feature extraction unit, configured to input the sample object attribute information into the object feature extraction network for feature extraction to obtain the sample object feature information; an interest feature extraction unit, configured to input the sample interest attribute information into the interest feature extraction network for feature extraction to obtain the sample interest feature information; The content feature extraction unit is configured to input the sample content attribute information into the content feature extraction network to perform feature extraction to obtain the sample content feature information.

13. The device according to claim 11, characterized in that The training module includes: an updating unit, configured to update network parameters of the object feature extraction network, the interest feature extraction network, and the content feature extraction network when the target loss does not meet a preset condition; A training unit is used to update the target loss based on the network parameters of the updated object feature extraction network, the interest feature extraction network, and the content feature extraction network until the target loss meets the preset conditions, and use the current content recommendation model as the target content recommendation model.

14. The device according to claim 11, characterized in that The sample object includes multiple objects, and the acquisition module includes: A first acquisition unit is configured to acquire at least one historical sample behavior content corresponding to each object; A second acquiring unit is used to acquire sample content attribute information of each historical sample behavior content; The screening unit is configured to screen at least one historical sample behavior content corresponding to each object based on the sample object attribute information, sample interest attribute information and corresponding sample content attribute information of each object to obtain the sample behavior content corresponding to each object.

15. A content recommendation device, characterized in that: The device comprises: An acquisition module, configured to acquire content attribute information of each content to be recommended in the content set to be recommended and interest attribute information corresponding to the target object; a feature extraction module configured to input the interest attribute information and the content attribute information of each to-be-recommended content into a feature extraction network of a target content recommendation model trained by the content recommendation model training method according to any one of claims 1 to 4, to perform feature extraction, thereby obtaining the interest feature information and the content feature information of each to-be-recommended content; A behavior prediction module is configured to input the interest feature information and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining first behavior prediction information of the target subject with respect to each to-be-recommended content; The recommendation module is configured to determine a first target recommended content from the set of contents to be recommended based on the first behavior prediction information of each content to be recommended, and recommend the first target recommended content to the target object.

16. The device according to claim 15, characterized in that The acquisition module is further configured to acquire object attribute information corresponding to the target object; The feature extraction module is further configured to input the object attribute information into the feature extraction network of the target content recommendation model to perform feature extraction to obtain object feature information; The behavior prediction module is further configured to input the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into the behavior prediction network of the target content recommendation model to perform behavior prediction, thereby obtaining second behavior prediction information of the target object with respect to each to-be-recommended content; The recommendation module is further configured to determine second target recommended content from the first target recommended content based on the second behavior prediction information, and recommend the second target recommended content to the target object.

17. The device according to claim 15, characterized in that The feature extraction module includes: an interest feature extraction unit, configured to input the interest attribute information into the interest feature extraction network for feature extraction to obtain the interest feature information; The content feature extraction unit is used to input the content attribute information of each content to be recommended into the content feature extraction network to perform feature extraction to obtain the content feature information of each content to be recommended.

18. The device according to claim 16, characterized in that The feature extraction module also includes: The object feature extraction unit is used to input the object attribute information into the object feature extraction network to perform feature extraction to obtain the object feature information.

19. The device according to claim 15, characterized in that The behavior prediction module includes: a second similarity calculation unit, configured to input the interest feature information and the content feature information of each to-be-recommended content into a behavior prediction network of the target content recommendation model, calculate the semantic similarity between the interest feature information and the content feature information of each to-be-recommended content, and obtain a second similarity corresponding to each to-be-recommended content; The first behavior prediction unit is configured to determine the second similarity corresponding to each to-be-recommended content as first behavior prediction information of the target object with respect to each to-be-recommended content.

20. The device according to claim 16, wherein The behavior prediction module also includes: a first similarity calculation unit, configured to input the object feature information, the interest feature information, and the content feature information of each to-be-recommended content into a behavior prediction network of the target content recommendation model, calculate the semantic similarity between the object feature information and the interest feature information, and obtain a first similarity; The second similarity calculation unit is further configured to calculate the language similarity between the interest feature information and the content feature information of each to-be-recommended content, to obtain a second similarity corresponding to each to-be-recommended content; The second behavior prediction unit is configured to determine a product of the first similarity and the second similarity corresponding to each content to be recommended as second behavior prediction information corresponding to each content to be recommended.

21. An electronic device, characterized in that: include: processor; memory for storing computer programs; The processor is configured to execute the computer program to implement the content recommendation model training method described in any one of claims 1 to 4 or the content recommendation method described in any one of claims 5 to 10.

22. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the content recommendation model training method described in any one of claims 1 to 4 or the content recommendation method described in any one of claims 5 to 10 is implemented.

23. A computer program product comprising a computer program, characterized in that When the computer program is executed by an electronic device, it implements the content recommendation model training method described in any one of claims 1 to 4 or the content recommendation method described in any one of claims 5 to 10.

Images