Content Interaction Prediction Method and Related Devices

Abstract

The present application discloses a content interaction prediction method and related devices. Related embodiments can be applied to various scenarios such as cloud technology, artificial intelligence, intelligent transportation, assisted driving, etc.; it can perform feature analysis processing on the interaction operations of the target content to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target sharing operation feature; based on the target first operation feature, determine the initial predicted interaction result of the target content on the first interaction operation; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced operation feature; predict the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; correct the initial predicted interaction result based on the negative noise information to obtain the target predicted interaction result. The present application can improve the prediction accuracy of the interaction result on the first interaction operation.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a content interaction prediction method and related devices. Background Art

[0002] With the rapid development of artificial intelligence technology, more and more application scenarios use artificial intelligence technology to recommend personalized content for users to improve the user interaction experience.

[0003] In the process of recommending content to users, related technologies generally predict the click probability and reading duration corresponding to each candidate content through a multi-task model, and then select the target content to be recommended to the user from the candidate content according to the predicted click probability and reading duration. However, current multi-task models generally focus too much on tasks with rich supervision signals (such as the prediction of click probability), resulting in poor prediction effects on tasks with fewer supervision signals (such as the prediction of reading duration). Summary of the Invention

[0004] Embodiments of this application provide a content interaction prediction method and related devices. The related devices may include a content interaction prediction device, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the prediction accuracy of the interaction result on the first interaction operation.

[0005] Embodiments of this application provide a content interaction prediction method, including:

[0006] Obtain target content; and perform feature analysis processing on the target content for an interaction operation to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature of the target content on a second interaction operation, and a target shared operation feature, where the target shared operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends;

[0007] Based on the target first operation feature, determine an initial predicted interaction result of the target content on the first interaction operation;

[0008] Perform feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation;

[0009] According to the enhanced operation feature, predict negative noise information generated by the second interaction operation on the first interaction operation;

[0010] Based on the negative noise information, correct the initial predicted interaction result to obtain a target predicted interaction result of the target content on the first interaction operation.

[0011] Correspondingly, an embodiment of the present application provides a content interaction prediction device, including:

[0012] An acquisition unit, configured to acquire target content; and perform feature analysis processing on the target content in terms of interaction operations, to obtain a target first operation feature of the target content in a first interaction operation, a target second operation feature of the target content in a second interaction operation, and a target sharing operation feature, where the target sharing operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends;

[0013] A determination unit, configured to determine an initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature;

[0014] A strengthening unit, configured to perform feature strengthening processing on the target second operation feature and the target sharing operation feature, to obtain a strengthened operation feature for the second interaction operation;

[0015] A prediction unit, configured to predict negative noise information generated by the second interaction operation on the first interaction operation according to the strengthened operation feature;

[0016] A correction unit, configured to correct the initial predicted interaction result based on the negative noise information, to obtain a target predicted interaction result of the target content in the first interaction operation.

[0017] Optionally, in some embodiments of the present application, the acquisition unit may include a feature extraction subunit and a feature interaction subunit, as follows:

[0018] The feature extraction subunit is configured to extract a first operation feature of the target content in a first interaction operation, a second operation feature of the target content in a second interaction operation, and a sharing operation feature;

[0019] The feature interaction subunit is configured to perform feature interaction processing on the first operation feature, the second operation feature, and the sharing operation feature, to obtain a target first operation feature corresponding to the first interaction operation, a target second operation feature corresponding to the second interaction operation, and a target sharing operation feature.

[0020] Optionally, in some embodiments of the present application, the feature interaction subunit may specifically be configured to fuse the first operation feature and the shared operation feature, and update the first operation feature based on the fused feature; fuse the second operation feature and the shared operation feature, and update the second operation feature based on the fused feature; fuse the first operation feature, the second operation feature, and the shared operation feature, and update the shared operation feature based on the fused feature; return to execute the step of fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature, until a target shared operation feature that meets the preset feature interaction condition, a target first operation feature corresponding to the first interaction operation, and a target second operation feature corresponding to the second interaction operation are obtained.

[0021] Optionally, in some embodiments of the present application, the strengthening unit may include a first fusion subunit, a first fully connected subunit, and a second fusion subunit, as follows:

[0022] The first fusion subunit is configured to fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0023] The first fully connected subunit is configured to perform a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation;

[0024] The second fusion subunit is configured to fuse the target feature information with the target shared operation feature to obtain a strengthened operation feature for the second interaction operation.

[0025] Optionally, in some embodiments of the present application, the second fusion subunit may specifically be configured to perform feature selection processing on the target shared operation feature and the second fused feature information respectively to obtain first feature information and second feature information; fuse the first feature information, the second feature information, and the target feature information to obtain a strengthened operation feature for the second interaction operation.

[0026] Optionally, in some embodiments of the present application, the determination unit may include a third fusion subunit, a second fully connected subunit, and a first determination subunit, as follows:

[0027] The third fusion subunit is configured to fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0028] A second fully-connected sub-unit, configured to perform a fully-connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation;

[0029] A first determination sub-unit, configured to determine an initial predicted interaction result of the target content on the first interaction operation based on the target feature information, the target shared operation feature, and the target first operation feature.

[0030] Optionally, in some embodiments of the present application, the first determination sub-unit may specifically be configured to fuse the target shared operation feature and the target first operation feature to obtain first fused feature information; perform feature selection processing on the second fused feature information to obtain second feature information; and determine an initial predicted interaction result of the target content on the first interaction operation according to the target feature information, the first fused feature information, and the second feature information.

[0031] Optionally, in some embodiments of the present application, the obtaining unit may specifically be configured to perform feature analysis processing on the target content in terms of interaction operations through a content interaction prediction model to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature on a second interaction operation, and a target shared operation feature.

[0032] Optionally, in some embodiments of the present application, the content interaction prediction device may further include a training unit, and the training unit is configured to train a content interaction prediction model; specifically, the training unit may include a data acquisition sub-unit, a feature analysis sub-unit, a second determination sub-unit, a reinforcement sub-unit, a prediction sub-unit, and an adjustment sub-unit, as follows:

[0033] The data acquisition sub-unit is configured to acquire training data, where the training data includes sample content, a first expected interaction result of the sample content on the first interaction operation, and a second expected interaction result on the second interaction operation;

[0034] The feature analysis sub-unit is configured to perform feature analysis processing on the sample content in terms of interaction operations through a content interaction prediction model to obtain a target first operation feature of the sample content on a first interaction operation, a target second operation feature on a second interaction operation, and a target shared operation feature;

[0035] The second determination sub-unit is configured to respectively determine an initial first actual interaction result of the sample content on the first interaction operation and a second actual interaction result on the second interaction operation based on the target first operation feature and the target second operation feature;

[0036] An intensifying subunit, configured to perform feature intensification processing on the target second operation feature and the target shared operation feature to obtain an intensified sample operation feature for the second interaction operation;

[0037] A prediction subunit, configured to predict, according to the intensified sample operation feature, the actual negative noise information generated by the second interaction operation on the first interaction operation; and based on the actual negative noise information, correct the initial first actual interaction result to obtain a target first actual interaction result;

[0038] An adjustment subunit, configured to adjust the parameters of the content interaction prediction model according to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result, to obtain a trained content interaction prediction model.

[0039] Optionally, in some embodiments of the present application, the adjustment subunit may specifically be configured to calculate a first loss value between the initial first actual interaction result and the first expected interaction result; calculate a second loss value between the target first actual interaction result and the first expected interaction result; calculate a third loss value between the second actual interaction result and the second expected interaction result; and adjust the parameters of the content interaction prediction model according to the first loss value, the second loss value, and the third loss value, to obtain a trained content interaction prediction model.

[0040] Optionally, in some embodiments of the present application, the step of "calculating the second loss value between the target first actual interaction result and the first expected interaction result" may include:

[0041] Based on the first loss value and the third loss value, calculate weight information corresponding to the sample content;

[0042] Perform loss calculation on the target first actual interaction result and the first expected interaction result to obtain an initial second loss value;

[0043] Fuse the weight information and the initial second loss value to obtain a second loss value.

[0044] An electronic device provided in an embodiment of the present application includes a processor and a memory. The memory stores multiple instructions, and the processor loads the instructions to execute the steps in the content interaction prediction method provided in the embodiment of the present application.

[0045] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the content interaction prediction method provided in the embodiment of the present application are implemented.

[0046] In addition, an embodiment of the present application further provides a computer program product, including a computer program or instruction, which, when executed by a processor, implements the steps in the content interaction prediction method provided by the embodiment of the present application.

[0047] An embodiment of the present application provides a content interaction prediction method and related devices, which can obtain target content; perform feature analysis processing on the interaction operations of the target content to obtain a target first operation feature of the target content in a first interaction operation, a target second operation feature of the target content in a second interaction operation, and a target sharing operation feature, where the target sharing operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; determine an initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain an enhanced operation feature for the second interaction operation; predict negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; and correct the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content in the first interaction operation. The present application can correct the predicted interaction result of the target content in the first interaction operation by capturing the negative noise information generated by the second interaction operation on the first interaction operation, thereby improving the prediction accuracy of the interaction result in the first interaction operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained without creative efforts based on these drawings.

[0049] Figure 1a is a schematic diagram of the scenario of the content interaction prediction method provided by the embodiment of the present application;

[0050] Figure 1b is a flowchart of the content interaction prediction method provided by the embodiment of the present application;

[0051] Figure 1c is a model architecture diagram of the content interaction prediction method provided by the embodiment of the present application;

[0052] Figure 1d is an explanatory diagram of the content interaction prediction method provided by the embodiment of the present application;

[0053] Figure 2 It is another flowchart of the content interaction prediction method provided by the embodiments of the present application;

[0054] Figure 3 It is a schematic structural diagram of the content interaction prediction device provided by the embodiments of the present application;

[0055] Figure 4 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application. Detailed implementation manners

[0056] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0057] The embodiments of the present application provide a content interaction prediction method and related devices. The related devices may include a content interaction prediction device, an electronic device, a computer-readable storage medium, and a computer program product. The content interaction prediction device may be specifically integrated in the electronic device, and the electronic device may be a device such as a terminal or a server.

[0058] It can be understood that the content interaction prediction method in this embodiment may be executed on the terminal, may also be executed on the server, or may be jointly executed by the terminal and the server. The above examples should not be construed as a limitation to the present application.

[0059] As Figure 1a shown, taking the terminal and the server jointly executing the content interaction prediction method as an example. The content interaction prediction system provided by the embodiments of the present application includes a terminal 10, a server 11, etc.; the terminal 10 and the server 11 are connected through a network, for example, through a wired or wireless network connection, etc. Among them, the content interaction prediction device may be integrated in the server.

[0060] Among them, the server 11 can be used to: obtain the target content; perform feature analysis processing on the target content in terms of interaction operations, to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature, where the target sharing operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends; determine the initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced operation feature for the second interaction operation; predict the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; and correct the initial predicted interaction result based on the negative noise information to obtain the target predicted interaction result of the target content in the first interaction operation. Among them, the server 11 can be a single server, or a server cluster or cloud server composed of multiple servers. For the content interaction prediction method or device disclosed in this application, multiple servers can form a blockchain, and the server is a node on the blockchain.

[0061] Among them, the terminal 10 can be used to: receive the target predicted interaction result of the target content in the first interaction operation sent by the server 11, and perform content recommendation based on the target predicted interaction result. Among them, the terminal 10 can include a mobile phone, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, a tablet computer, a laptop computer, or a personal computer (PC, Personal Computer), etc. A client can also be set on the terminal 10, and the client can be an application client or a browser client, etc.

[0062] The steps such as content interaction prediction in the above-mentioned server 11 can also be executed by the terminal 10.

[0063] The content interaction prediction method provided by the embodiments of this application relates to natural language processing and machine learning in the field of artificial intelligence.

[0064] Among them, artificial intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, and is a theory, method, technology, and application system that can perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making. Artificial intelligence technology is an interdisciplinary subject that involves a wide range of fields, including both hardware-level and software-level technologies. Among them, artificial intelligence software technology mainly includes several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, autonomous driving, and intelligent transportation.

[0065] Among them, natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between humans and computers in natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, the research in this field will involve natural language, that is, the language people use in daily life, so it has a close connection with the research of linguistics. Natural language processing technology usually includes technologies such as text processing, semantic understanding, machine translation, robot question answering, and knowledge graphs.

[0066] Among them, machine learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning from demonstration.

[0067] The following will be described in detail respectively. It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments.

[0068] This embodiment will be described from the perspective of the content interaction prediction device, which can be specifically integrated in an electronic device, and the electronic device can be a device such as a server or a terminal.

[0069] It is understandable that in the specific embodiments of the present application, user information, such as data related to reading duration, etc., is involved. When the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions.

[0070] The content interaction prediction method of the embodiments of the present application can be applied to scenarios such as content recommendation. This embodiment can be applied to various scenarios such as cloud technology, artificial intelligence, intelligent transportation, and assisted driving.

[0071] As Figure 1b shown, the specific process of this content interaction prediction method can be as follows:

[0072] 101. Obtain the target content; and perform feature analysis processing on the interaction operations of the target content to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature, where the target shared operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends.

[0073] Among them, the target content is the content for which the interaction result is to be predicted. The interaction result to be predicted may specifically include the interaction result on the first interaction operation or the interaction result on the second interaction operation. This embodiment does not limit this. It should be noted that the content form of the target content may include text, audio, image, video, etc.

[0074] Among them, the target shared operation feature is the feature shared by the first interaction operation and the second interaction operation; the target first operation feature can be regarded as the unique feature of the first interaction operation, and the target second operation feature can be regarded as the unique feature of the second interaction operation.

[0075] Among them, the second interaction operation is the pre-operation on which the first interaction operation depends, and there is a dependency relationship between them in terms of operation time. The operation time of the second interaction operation is before the operation time of the first interaction operation. For example, in a specific scenario, the target content is the content recommended to the target object, the second interaction operation is the click operation on the content, and the first interaction operation is the reading duration or sharing operation of the content, etc. Predicting the interaction result on the second interaction operation may specifically be to predict whether the target object clicks on the target content, and predicting the interaction result on the first interaction operation may specifically be to predict the reading duration of the target object for the target content or whether the target content will be shared. It is understandable that the target object needs to click on the target content first before reading or sharing the target content. Therefore, the click is the pre-operation corresponding to the reading duration or sharing.

[0076] Since there is a dependency relationship between the operation times of the first interaction operation and the second interaction operation, if the same content interaction prediction model is used to predict the interaction results of the first interaction operation and the second interaction operation, the sample numbers of the two will differ significantly during the model training process, resulting in fewer supervision signals for the first interaction operation and more supervision signals for the second interaction operation.

[0077] Specifically, when jointly training the first interaction operation (such as reading duration) and the second interaction operation (such as click operation), there are some task-specific problems:

[0078] One is the data sparsity problem of reading duration compared to click operation. For example, in some real recommendation systems, only a small portion of the exposed articles are clicked and read, which indicates that the data collected for training the prediction task of reading duration is generally much lower than that of the click prediction task. This will cause a problem - the shared parameter part of the multi-task learning model (such as shared feature embeddings and underlying shared expert layers) is mainly optimized under the supervision of click signals. Specifically, the shared parameters account for more than 99% of the total parameters of the multi-task learning model, which means that the optimization of the click task dominates in the multi-task model, while the optimization of the reading duration task is insufficient.

[0079] The second is the complex relationship and deep coupling between click operation and reading duration. The interaction operations of users with the target content follow the behavioral sequence pattern of "exposure - click - reading", so there is a high degree of dependence and serious coupling relationship between click operation and reading duration. Although the correlation between the two is very strong, there are still conflicts between the two goals, which will lead to a decline in the effect of a certain goal during joint optimization. For example, for an article with an eye-catching title, its content may not be of good quality, and users may close the article soon after clicking, which results in a high click rate but a short reading duration for this article.

[0080] In the current related technologies, the multi-task model for predicting click operation and reading duration often ignores the negative impact of clicks on reading duration. Due to the seesaw effect of the multi-task model - the model overly focuses on the task with rich supervision signals (such as click prediction), resulting in poor performance on the task with fewer supervision signals (such as reading duration prediction), that is, the prediction of the interaction result corresponding to the first interaction operation is inaccurate, and the existing model cannot achieve good results on both click prediction and reading duration prediction.

[0081] The content interaction prediction method of this application can provide a multi-task causal framework, which explicitly captures and removes this negative impact by introducing causal inference to improve the prediction effect of reading duration.

[0082] Among them, causal inference is a research field in statistics for analyzing the causal relationship between variables; in this embodiment, the first interaction operation depends on the second interaction operation, and the two can be regarded as having a causal relationship.

[0083] Optionally, in this embodiment, the step of "performing feature analysis processing on the interaction operation of the target content to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature" may include:

[0084] Performing feature analysis processing on the interaction operation of the target content through a content interaction prediction model to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature.

[0085] Among them, the content interaction prediction model is specifically a neural network model, and the neural network can be a residual network (ResNet, Residual Network), or a dense connection convolutional network (DenseNet, Dense Convolutional Network), etc. It should be understood that the neural network in this embodiment is not limited to the several types listed above.

[0086] Specifically, the content interaction prediction model is a multi-task model, which may include a feature extraction network with a multi-layer structure and tower networks corresponding to each task in the target multi-task; each layer of the feature extraction network includes multiple expert networks corresponding to each task and a gating network. For example, the content interaction prediction model may include a multi-layer feature extraction network and tower networks corresponding to the prediction tasks corresponding to the first interaction operation and the second interaction operation.

[0087] Among them, the feature extraction network can be used to extract the unique features and shared features of each task, and the tower network can be used to meet the specific application requirements of each task, such as classification tasks or prediction tasks. In each layer of the feature extraction network, multiple expert networks and gating networks corresponding to each task are provided, and the expert networks can include task-sharing experts and task-specific experts.

[0088] Among them, during the multi-task learning process, the output of the expert network can be weighted by using a gating network; the expert network in the multi-task model can be selectively controlled through the output weights. The gating network can be used to fuse the feature data extracted by the expert network. The weights of the expert networks output by the gating networks corresponding to different tasks are different, and thus the feature fusion data of each task at this level is different. Due to the weight assignment of the gating network, different expert networks can learn different signals from different perspectives; then, the feature fusion data corresponding to each task is used as the input of the feature extraction network at the next level to perform feature interaction until finally input into the tower network corresponding to each task to obtain the processing results of each task. In this content interaction prediction model, for each task, the corresponding gating network is also multi-layered, and the number of layers is the same as that of the feature extraction network. In this embodiment, in the scenario of processing multiple related tasks, both the specificity of the tasks and the correlation between the tasks are distinguished, which can effectively improve the generalization ability of the model and the accuracy of the processing results.

[0089] Among them, the expert network can adopt different network structures and parameters for different tasks based on the same representative input. For the tower networks of each task, the same network framework can be adopted, or different network frameworks can be adopted to make the multi-task model have flexible variability.

[0090] Among them, multi-task learning (MTL) is a field in machine learning, a machine learning method that puts multiple related tasks together based on shared representative data, and is also a type of transfer learning. This learning method enables different tasks to learn information in related fields and then share this part of the information in the model. Through mutual learning and sharing among multiple tasks, the generalization ability and effect of the model are improved.

[0091] In a specific embodiment, as Figure 1c shown, it is the model structure diagram of the content interaction prediction model. This content interaction prediction model can be composed of a multi-layer expert network (Experts) at the bottom and a task-specific tower (Tower) network at the top to learn the high-order interaction of each input embedded vector. Each expert module consists of multiple sub-networks, and each sub-network is called an expert. In the expert module, the task-sharing experts and task-specific experts are explicitly separated to avoid mutual interference between tasks; while the gating network is used to combine the knowledge of the experts at the lower level.

[0092] Among them, specifically, this content interaction prediction model can be used to predict the interaction result on the first interaction operation, denoted as task T; this content interaction prediction model can also be used to predict the interaction result on the second interaction operation, denoted as task C, and the gating network is denoted as G.

[0093] Among them, the gating network formula of task k in the j-th layer of the expert network is shown in Equation (1):

[0094] g k,j (x) = w k,j (g k,j-1 (x))S k,j (x) (1)

[0095] Among them, x is the input embedded vector, and w k,j is the weight function of task k, and its structure is based on a single-layer network with Softmax as the activation function, as shown in Equation (2):

[0096]

[0097] Among them, is the parameter matrix. Note that the gating network of the first layer is slightly different, and the corresponding formula of the gating network of the first layer is shown in Equation (3):

[0098] g k,1 (x) = w k,1 (x)S k,1 (x) (3)

[0099] Among them, S k,j is the selection matrix of task k in the j-th layer network, which is composed of selection vectors, including shared experts and exclusive experts of task k, as shown in Equation (4):

[0100]

[0101] Among them, etc. are respectively individual experts exclusive to task k in the j-th layer network, and there are m k task-exclusive experts, etc. are respectively individual experts shared by tasks in the j-th layer network, and there are m S task-shared experts. It should be noted that the selection matrix of the shared expert module is slightly different and is composed of all shared experts and task-exclusive experts.

[0102] Optionally, in this embodiment, the step of "performing feature analysis processing on the target content in terms of interaction operations to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target shared operation feature" may include:

[0103] Extracting the first operation feature of the target content in the first interaction operation, the second operation feature in the second interaction operation, and the shared operation feature;

[0104] Perform feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain the target first operation feature corresponding to the first interaction operation, the target second operation feature corresponding to the second interaction operation, and the target shared operation feature.

[0105] Specifically, the content information of the target content in each dimension can be input into the content interaction prediction model. The content information of the target content in each dimension can include the content title, content publisher information, content cover, content itself information, etc.

[0106] Among them, through the content interaction prediction model, feature analysis processing is performed on the interaction operation of the target content. As Figure 1c shown, specifically, the first operation feature can be extracted by the task-specific expert T in the first layer, the second operation feature can be extracted by the task-specific expert C in the first layer, and the shared operation feature can be extracted by the task-sharing expert in the first layer; then, through the feature extraction networks in the subsequent layers, feature interaction processing is performed on the first operation feature, the second operation feature, and the shared operation feature, so that the task-specific expert T in the last layer outputs the target first operation feature, the task-specific expert C in the last layer outputs the target second operation feature, and the task-sharing expert in the last layer outputs the target shared operation feature.

[0107] Optionally, in this embodiment, the step of "performing feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain the target first operation feature corresponding to the first interaction operation, the target second operation feature corresponding to the second interaction operation, and the target shared operation feature" may include:

[0108] Fuse the first operation feature and the shared operation feature, and update the first operation feature based on the fused feature;

[0109] Fuse the second operation feature and the shared operation feature, and update the second operation feature based on the fused feature;

[0110] Fuse the first operation feature, the second operation feature, and the shared operation feature, and update the shared operation feature based on the fused feature;

[0111] Return to execute the step of fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature until the target shared operation feature that meets the preset feature interaction condition, the target first operation feature corresponding to the first interaction operation, and the target second operation feature corresponding to the second interaction operation are obtained.

[0112] Among them, the preset feature interaction condition can be set according to the actual situation, and this embodiment does not limit this. For example, the preset feature interaction condition can specifically be that the number of updates reaches a preset number. In some embodiments, the preset feature interaction condition can be determined according to the number of layers of the feature extraction network in the content interaction prediction model.

[0113] Among them, the first operation feature and the shared operation feature can be fused through a gating network; there are various fusion methods, and this embodiment does not limit this. For example, the fusion method can be weighted operation or splicing, etc. And the fused feature is input into the next layer of task-specific expert T, and based on the output feature of the next layer of task-specific expert T, the first operation feature is updated. Specifically, the output feature processed by the next layer of task-specific expert T can be determined as the new first operation feature.

[0114] Among them, the second operation feature and the shared operation feature can be fused through a gating network; there are various fusion methods, and this embodiment does not limit this. For example, the fusion method can be weighted operation or splicing, etc. And the fused feature is input into the next layer of task-specific expert C, and based on the output feature of the next layer of task-specific expert C, the second operation feature is updated. Specifically, the output feature processed by the next layer of task-specific expert C can be determined as the new second operation feature.

[0115] Among them, the first operation feature, the second operation feature, and the shared operation feature can be fused through a gating network; there are various fusion methods, and this embodiment does not limit this. For example, the fusion method can be weighted operation or splicing, etc. And the fused feature is input into the next layer of task-sharing expert, and based on the output feature of the next layer of task-sharing expert, the shared operation feature is updated. Specifically, the output feature processed by the next layer of task-sharing expert can be determined as the new shared operation feature.

[0116] 102. Based on the target first operation feature, determine the initial predicted interaction result of the target content on the first interaction operation.

[0117] Optionally, in this embodiment, the step of "based on the target first operation feature, determine the initial predicted interaction result of the target content on the first interaction operation" may include:

[0118] Fuse the target shared operation feature and the target first operation feature to obtain a first fused operation feature;

[0119] Based on the first fused operation feature, determine the initial predicted interaction result of the target content on the first interaction operation.

[0120] Among them, the target shared operation feature and the target first operation feature can be fused through a gating network. This fusion method can be weighted operation, splicing processing, etc., and this embodiment does not limit this. And the first fused operation feature obtained by fusion is used as the input of the tower network T in the content interaction prediction model. After being processed by the tower network T, the initial predicted interaction result of the target content on the first interaction operation is obtained.

[0121] Among them, the tower network T can be a neural network structure, which can include a convolutional layer, a fully connected layer, etc., and this embodiment does not limit this. Specifically, the tower network T can include a multi-layer perceptron (MLP, Multilayer Perceptron). The first fused operation feature is fully connected through the multi-layer perceptron to predict the probability that the target content belongs to each preset interaction result on the first interaction operation, and according to the probability, the initial predicted interaction result of the target content on the first interaction operation is determined.

[0122] Among them, in some embodiments, the preset interaction result with the highest probability can be determined as the initial predicted interaction result of the target content on the first interaction operation; in some other embodiments, it can also be the preset interaction result with a probability greater than a preset value that is determined as the initial predicted interaction result of the target content on the first interaction operation.

[0123] For example, if the first interaction operation is the reading duration of the target content, the reading duration interval can be divided according to the actual situation. For example, the reading duration interval can be divided into three sub-intervals: the reading time is less than 3 minutes, the reading time is between 3 and 10 minutes, and the reading time is greater than 10 minutes. Then the preset interaction results corresponding to the first interaction operation can include the three situations corresponding to these three sub-intervals.

[0124] For another example, if the first interaction operation is the sharing operation of the target content, the preset interaction results corresponding to the first interaction operation can include two situations: sharing and not sharing.

[0125] Optionally, in this embodiment, this content interaction prediction method may further include:

[0126] Based on the target second operation feature, determine the predicted interaction result of the target content on the second interaction operation.

[0127] Among them, in some embodiments, the step of "Based on the target second operation feature, determine the predicted interaction result of the target content on the second interaction operation" may include:

[0128] Fuse the target shared operation feature and the target second operation feature to obtain a second fused operation feature;

[0129] Determine the predicted interaction result of the target content on the second interaction operation based on the second fusion operation feature.

[0130] Among them, the target shared operation feature and the target second operation feature can be fused through a gating network. This fusion method can be weighted operation, splicing processing, etc., and this embodiment does not limit this. And the second fusion operation feature obtained by fusion is used as the input of the tower network C in the content interaction prediction model. After being processed by the tower network C, the predicted interaction result of the target content on the second interaction operation is obtained.

[0131] Among them, the tower network C can be a neural network structure, which can include a convolutional layer and a fully connected layer, etc., and this embodiment does not limit this. Specifically, the tower network C can include a multi-layer perceptron (MLP, Multilayer Perceptron). The second fusion operation feature is fully connected through the multi-layer perceptron to predict the probability that the target content belongs to each preset interaction result on the second interaction operation, and based on the probability, the predicted interaction result of the target content on the second interaction operation is determined.

[0132] Among them, in some embodiments, the preset interaction result with the highest probability can be determined as the predicted interaction result of the target content on the second interaction operation; in other embodiments, the preset interaction result with a probability greater than a preset value can also be determined as the predicted interaction result of the target content on the second interaction operation.

[0133] For example, if the second interaction operation is a click operation on the target content, the preset interaction results corresponding to the second interaction operation can include two situations: click and non-click.

[0134] Optionally, in this embodiment, the step of "determine the initial predicted interaction result of the target content on the first interaction operation based on the target first operation feature" may include:

[0135] Fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0136] Perform a fully connected process on the second fused feature information to obtain the target feature information corresponding to the second interaction operation;

[0137] Based on the target feature information, the target shared operation feature, and the target first operation feature, determine the initial predicted interaction result of the target content on the first interaction operation.

[0138] Among them, the target second operation feature and the target shared operation feature can be fused through a gating network. The fusion method can be weighted operation, splicing process, etc., and this embodiment does not limit this. And the second fused feature information obtained by fusion is used as the input of the tower network C in the content interaction prediction model. After the fully connected processing of the tower network C, the target feature information corresponding to the second interaction operation is obtained.

[0139] Optionally, in this embodiment, the step of "determining the initial predicted interaction result of the target content on the first interaction operation based on the target feature information, the target shared operation feature, and the target first operation feature" may include:

[0140] Fuse the target shared operation feature and the target first operation feature to obtain first fused feature information;

[0141] Perform feature selection processing on the second fused feature information to obtain second feature information;

[0142] Determine the initial predicted interaction result of the target content on the first interaction operation according to the target feature information, the first fused feature information, and the second feature information.

[0143] Among them, the target shared operation feature and the target first operation feature can be fused through a gating network. The fusion method can be weighted operation, splicing process, etc., and this embodiment does not limit this.

[0144] Among them, the second fused feature information can be used as the input of the gating network g r in the content interaction prediction model, and feature selection processing is performed through the gating network g r to obtain second feature information.

[0145] Among them, the step of "determining the initial predicted interaction result of the target content on the first interaction operation according to the target feature information, the first fused feature information, and the second feature information" may include:

[0146] Fuse the target feature information, the first fused feature information, and the second feature information to obtain a target interaction feature;

[0147] Based on the target interaction feature, determine the initial predicted interaction result of the target content on the first interaction operation.

[0148] Among them, there are various fusion methods, which can be weighted fusion or splicing processing, etc. This embodiment does not limit this. And the obtained target interaction feature after fusion is used as the input of the tower network T in the content interaction prediction model. After being processed by the tower network T, the initial predicted interaction result of the target content on the first interaction operation is obtained.

[0149] Specifically, the tower network T may include a multi-layer perceptron (MLP, Multilayer Perceptron). The target interaction feature is fully connected by the multi-layer perceptron to predict the probability that the target content belongs to each preset interaction result on the first interaction operation, and the initial predicted interaction result of the target content on the first interaction operation is determined according to the probability.

[0150] Among them, in some embodiments, the preset interaction result with the highest probability can be determined as the initial predicted interaction result of the target content on the first interaction operation; in other embodiments, it can also be that the preset interaction results with probabilities greater than the preset value are determined as the initial predicted interaction results of the target content on the first interaction operation.

[0151] In a specific embodiment, the target second operation feature is specifically the output of the last layer task-exclusive expert C in the content interaction prediction model, and the target shared operation feature is specifically the output of the last layer task-sharing expert in the content interaction prediction model. Through the gating network g C，L fuses the target second operation feature and the target shared operation feature, and the second fused feature information can be obtained. The second fused feature information can be denoted as g C，L (x), where x represents the input embedded vector; taking g C，L (x) as the input of the tower network C, after the full connection processing of the tower network C, the target feature information corresponding to the second interaction operation is obtained, and the target feature information can be denoted as t′ C (g C，L (x)).

[0152] Among them, t′ C can represent the first few layers of the tower network C, specifically, it can be only other layers except the full connection layer, because the output of the full connection layer is a scalar and contains too little information.

[0153] In addition, the gating network g r performs feature selection processing on the second fused feature information g C，L (x) to obtain the second feature information, and the second feature information can be denoted as g r (g C，L (x)). Then the fusion result of the target feature information and the second feature information can be as shown in formula (5):

[0154] rC f(x) = g r (g C，L (x))t′ C (g C，L (x)) (5)

[0155] Wherein, the fusion result r C (x) is fused with the first fused feature information to obtain the target interaction feature.

[0156] 103. Feature enhancement processing is performed on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0157] Optionally, in this embodiment, the step of "performing feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation" may include:

[0158] Fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0159] Perform a fully connected process on the second fused feature information to obtain the target feature information corresponding to the second interaction operation;

[0160] Fuse the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0161] Among them, the target second operation feature and the target shared operation feature can be fused through a gated network. This fusion method can be weighted operation, splicing process, etc. This embodiment does not limit this. And the second fused feature information obtained by fusion is used as the input of the tower network C in the content interaction prediction model. After the fully connected process of the tower network C, the target feature information corresponding to the second interaction operation is obtained.

[0162] Optionally, in this embodiment, the step of "fusing the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation" may include:

[0163] Feature selection processing is respectively performed on the target shared operation feature and the second fused feature information to obtain first feature information and second feature information;

[0164] Fuse the first feature information, the second feature information, and the target feature information to obtain an enhanced operation feature for the second interaction operation.

[0165] Among them, in some embodiments, the target shared operation features can be processed by a gating network to select features and obtain first feature information.

[0166] Among them, the second fused feature information can be used as the input of the gating network g in the content interaction prediction model. r Through the gating network g r perform feature selection processing to obtain second feature information.

[0167] Among them, there are various ways to fuse the first feature information, the second feature information, and the target feature information. This embodiment does not limit this. For example, the fusion method can be splicing processing or weighted operation, etc. Through the fusion processing, the enhanced operation features for the second interaction operation can be obtained.

[0168] Among them, the target feature information and the second feature information can be fused first, and then the fusion result r C (x) is fused with the first feature information to obtain the enhanced operation features. Among them, the target feature information t′ C (g C，L (x)) and the second feature information g r (g C，L (x)) can be shown as the formula (5) in step 102.

[0169] In a specific embodiment, the first interaction operation is the reading duration, and the second interaction operation is the click operation. Through feature enhancement processing, the click target can be enhanced to approximate the real negative impact, and finally the negative impact is subtracted from the original reading duration prediction to achieve the purpose of alleviating the negative impact of the click on the duration and strengthening the positive impact.

[0170] 104. Predict the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation features.

[0171] Among them, since the enhanced operation features fuse the feature output of the tower network C corresponding to the second interaction operation, the enhanced operation features are the features enhanced for the second interaction operation.

[0172] Among them, in this embodiment, the negative noise modeling module in the content interaction prediction model can be used to predict the negative noise information generated by the second interaction operation on the first interaction operation based on the enhanced operation features. Specifically, the enhanced operation features can be input into the negative noise modeling module, and the negative noise modeling module processes it to predict the negative noise information generated by the second interaction operation on the first interaction operation.

[0173] Among them, the negative noise modeling module may include a multi-layer perceptron (MLP, Multilayer Perceptron). The enhanced operation features are fully connected through the multi-layer perceptron to predict the negative noise information generated by the second interaction operation on the first interaction operation.

[0174] 105. Based on the negative noise information, correct the initial predicted interaction result to obtain the target predicted interaction result of the target content on the first interaction operation.

[0175] Among them, the negative noise information includes the negative impact of the second interaction operation on the first interaction operation. The correction of the initial predicted interaction result may specifically be subtracting the negative noise information from the initial predicted interaction result, and the target predicted interaction result of the target content on the first interaction operation can be obtained. This target predicted interaction result is the predicted value after subtracting the negative impact.

[0176] Specifically, if the first interaction operation is the reading duration and the second interaction operation is the click operation, through correction, the negative impact can be subtracted from the original predicted value of the reading duration (i.e., the initial predicted interaction result) to alleviate the negative impact of the click operation on the reading duration and strengthen the positive impact.

[0177] In a specific embodiment, the content interaction prediction method provided in this application can be applied to a content recommendation scenario. After obtaining the target predicted interaction results of each content on the first interaction operation, the target recommended content can be selected from each content for recommendation according to the target predicted interaction result.

[0178] It should be noted that the content interaction prediction model can be specifically trained by other devices and then provided to the content interaction prediction device, or the content interaction prediction device can also train itself.

[0179] If the content interaction prediction device trains itself, before the step of "performing feature analysis processing on the target content for the interaction operation through the content interaction prediction model to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature", the content interaction prediction method may further include:

[0180] Obtain training data, where the training data includes sample content, the first expected interaction result of the sample content on the first interaction operation, and the second expected interaction result on the second interaction operation;

[0181] Through the content interaction prediction model, perform feature analysis processing on the sample content in terms of interaction operations, to obtain the target first operation feature of the sample content in the first interaction operation, the target second operation feature in the second interaction operation, and the target shared operation feature;

[0182] Based on the target first operation feature and the target second operation feature respectively, determine the initial first actual interaction result of the sample content in the first interaction operation and the second actual interaction result in the second interaction operation;

[0183] Perform feature enhancement processing on the target second operation feature and the target shared operation feature, to obtain the enhanced sample operation feature for the second interaction operation;

[0184] According to the enhanced sample operation feature, predict the actual negative noise information generated by the second interaction operation on the first interaction operation; and based on the actual negative noise information, correct the initial first actual interaction result to obtain the target first actual interaction result;

[0185] According to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result, adjust the parameters of the content interaction prediction model to obtain the trained content interaction prediction model.

[0186] Wherein, the first expected interaction result may be the expected probability of the sample content belonging to each preset interaction result in the first interaction operation; the second expected interaction result may be the expected probability of the sample content belonging to each preset interaction result in the second interaction operation.

[0187] Optionally, in this embodiment, the step "According to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result, adjust the parameters of the content interaction prediction model to obtain the trained content interaction prediction model" may include:

[0188] Calculate the first loss value between the initial first actual interaction result and the first expected interaction result;

[0189] Calculate the second loss value between the target first actual interaction result and the first expected interaction result;

[0190] Calculate the third loss value between the second actual interaction result and the second expected interaction result;

[0191] Adjust the parameters of the content interaction prediction model according to the first loss value, the second loss value, and the third loss value, to obtain a trained content interaction prediction model.

[0192] Among them, the training process specifically uses the backpropagation algorithm to adjust the parameters of the content interaction prediction model, and optimizes the parameters of the content interaction prediction model based on the first loss value, the second loss value, and the third loss value, so that the first loss value, the second loss value, and the third loss value meet the preset loss condition, to obtain a trained content interaction prediction model. Specifically, the preset loss condition may be that the sum of the first loss value, the second loss value, and the third loss value is less than a preset loss value, and the preset loss value can be set according to the actual situation.

[0193] Among them, there are various calculation methods for the loss value, and this embodiment does not limit this. For example, it can be a cross-entropy loss function, or a mean square error loss function.

[0194] Optionally, in this embodiment, the step of "calculating the second loss value between the target first actual interaction result and the first expected interaction result" may include:

[0195] Calculate the weight information corresponding to the sample content based on the first loss value and the third loss value;

[0196] Perform a loss calculation on the target first actual interaction result and the first expected interaction result to obtain an initial second loss value;

[0197] Fuse the weight information and the initial second loss value to obtain the second loss value.

[0198] Among them, the first loss value and the third loss value can be fused to obtain the weight information corresponding to the sample content, and this fusion method can be multiplication, addition, etc., and this embodiment does not limit this.

[0199] In a specific embodiment, the initial first actual interaction result is denoted as The first expected interaction result is denoted as The target first actual interaction result is denoted as The second actual interaction result is denoted as The second expected interaction result is denoted as Then the calculation processes of the first loss value, the second loss value, and the third loss value are respectively shown in formulas (6), (7), and (8):

[0200]

[0201]

[0202]

[0203] Among them, L T is the first loss value, L MTC is the second loss value, L C is the third loss value. S is the exposure sample data set, S + represents the data set of samples that have been clicked, i represents each sample content, w i represents the weight information corresponding to the sample content, L MTC,i represents the initial second loss value of the sample content i.

[0204] Among them, specifically, if the second interaction operation is a click operation, then represents the true click label, usually represented by 1 for click and 0 for non-click, L C represents the loss of the click task; if the first interaction operation is the reading time, the reading duration can be modeled as a multi-classification problem, and the continuous duration values are divided into multiple intervals, represents the true value vector corresponding to the interval j after discretizing the reading duration into M groups. Except for the value of the dimension where the true value belongs being 1, other values are all 0, L T represents the reading duration loss of the original multi-task, L MTC represents the corrected reading duration loss.

[0205] For example, the reading duration can be divided into three groups, namely the reading time less than 3 minutes, the reading time from 3 to 10 minutes, and the reading time greater than 10 minutes.

[0206] Among them, the specific calculation process of the weight information corresponding to the sample content can be as shown in formula (9):

[0207] w i =(L C，i ) α ×L T，i +β (9)

[0208] Among them, both α and β are hyperparameters, L C，i is the third loss value corresponding to the sample content i, L T，i is the first loss value corresponding to the sample content i.

[0209] Specifically, since the negative noise modeling module only enhances the ability to capture the impact of clicks, but whether this impact is positive or negative is not yet clear. Therefore, the capture of knowledge with negative click impacts can be enhanced in the optimization function. A reasonable assumption is that samples with accurate click task predictions but inaccurate reading duration predictions are more likely to be adversely affected by clicks than other samples. Therefore, the weights of samples with small click loss function values but large reading duration loss function values can be increased, and vice versa. So when calculating the second loss value L MTC the weights of samples can be increased by adding the product of the click loss function value L C，i predicted by the original multi-task framework and the reading duration loss function value L T，i .

[0210] In this embodiment, the step of "adjusting the parameters of the content interaction prediction model according to the first loss value, the second loss value, and the third loss value to obtain the trained content interaction prediction model" may include:

[0211] Fusing the first loss value, the second loss value, and the third loss value to obtain a total loss value;

[0212] Based on the total loss value, adjusting the parameters of the content interaction prediction model to obtain the trained content interaction prediction model.

[0213] Among them, there are various fusion methods, such as weighted fusion, etc. Specifically, the calculation process of the total loss value L can be as shown in formula (10):

[0214] L = L C + L T + L MTC (10)

[0215] The content interaction prediction method provided by this application can be used in scenarios such as content recommendation. It can be used to help alleviate the interference of clicks on reading duration or viewing duration in multi-task recommendation, and can also be used to alleviate the interference between other multi-behaviors with dependency relationships, such as the interference of clicks on sharing, etc. This application can capture the negative impacts between targets through the negative noise modeling module, subtract this negative impact from the original predicted value to obtain a predicted value with alleviated negative impacts, and finally use the predicted value after subtracting the negative impact.

[0216] In a specific embodiment, as Figure 1c shown, this embodiment can predict the interaction results of the target content on the first interaction operation (such as reading duration) and on the second interaction operation (such as click) through this content interaction prediction model. Specifically, the negative noise modeling module is used to learn the negative impact on the reading duration task due to the click task.

[0217] Among them, the negative noise modeling module can also be called the Negative Impact Modeling (NIM) module, which can specifically be a multi-layer feedforward network with width and depth as hyperparameters. The input of the negative noise modeling module can include the output g S，L (x) and r C (x), where the calculation process of r C (x) can refer to Equation (5) in the above embodiment, and the calculation process of g S，L (x) can be as shown in Equation (11):

[0218] g S，L (x) = w S，L (g S，L-1 (x))S S，L (x) (11)

[0219] Among them, S S，L (x) only selects the task-sharing experts of this layer, and w S，L is the weight function. Finally, the predicted value of the negative impact of clicks on the reading duration can be obtained, that is, the negative noise information Specifically, it is as shown in Equation (12):

[0220]

[0221] Among them, t NIM represents the tower network corresponding to the negative noise modeling module.

[0222] Among them, in this embodiment, integrating r C (x) into the feature calculation of the tower network T and the negative noise modeling module can make the reading-duration-specific tower network T more inclined to learn the information of the reading duration during optimization, reducing confusion; and making the input of the negative impact modeling module contain more click information, so as to more effectively capture the negative impact of clicks on the reading duration.

[0223] Specifically,[[]] represents the reading duration prediction mainly affected by the click signal. If the original reading duration output in the multi-task learning framework is expressed as then the predicted value of the reading duration after removing the click influence can be obtained, as shown in Equation (13):

[0224]

[0225] Among them,[[]] is the initial predicted interaction result of the reading duration The target predicted interaction result obtained after correction. Specifically, it is the output of the tower network T, that is, the original estimated reading duration, which is also the initial predicted interaction result on the first interaction operation described in the above embodiment; Specifically, it is the output value of the negative noise modeling module, representing the estimated reading duration affected by the negative impact of the click signal.

[0226] The reading duration intervals misclassified due to the negative impact of the click signal will have higher values, and then subtracting this value from the original estimated value will make the correct reading duration interval have a higher predicted probability.

[0227] In a specific embodiment, the content interaction prediction method provided in this application is significantly superior to the existing related models in predicting the reading duration, and the prediction effect of the click task is also good. The test effect on the log data of a certain recommendation system is as Figure 1d shown. Among them, Model A is the model provided in this application after including the negative impact modeling module; NFM, DeepFM, AutoInt, and AFN are single-task models, and MMOE, AITM, and PLE are multi-task models. Compared with other models, Model A provided in this application has achieved the best results in all reading duration-related metrics.

[0228] Among them, the related metrics include Mean Absolute Error - Classification (MAE_class), Root Mean Square Error - Classification (RMSE_class), Recall, F1, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). F1 is a comprehensive evaluation metric, and the higher the F1 value, the better the prediction effect. AUC is a model evaluation metric.

[0229] Among them, the estimation of the reading duration is very important for the recommendation system because a longer reading duration usually represents that the target object has a greater interest in the recommended content, thus effectively making up for the shortcoming that clicks may not be able to reflect the true preferences of the target object, because clicks only reflect the target object's interest in the content title. Accurate estimation of the reading duration helps to recommend content that truly meets the interests of the target object, thereby improving the user experience.

[0230] As can be seen from the above, this embodiment can obtain the target content; and perform feature analysis processing on the target content in terms of interaction operations, to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature. The target sharing operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; based on the target first operation feature, determine the initial predicted interaction result of the target content in the first interaction operation; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced operation feature for the second interaction operation; according to the enhanced operation feature, predict the negative noise information generated by the second interaction operation on the first interaction operation; based on the negative noise information, correct the initial predicted interaction result to obtain the target predicted interaction result of the target content in the first interaction operation. This application can correct the predicted interaction result of the target content in the first interaction operation by capturing the negative noise information generated by the second interaction operation on the first interaction operation, improving the prediction accuracy of the interaction result in the first interaction operation.

[0231] According to the method described in the previous embodiment, the following will take the case where the content interaction prediction device is specifically integrated in the server as an example for further detailed description.

[0232] An embodiment of this application provides a content interaction prediction method, as Figure 2 shown, the specific process of this content interaction prediction method can be as follows:

[0233] 201. The server obtains the target content; and performs feature analysis processing on the target content in terms of interaction operations, to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature. The target sharing operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends.

[0234] Among them, the target sharing operation feature is a feature shared by the first interaction operation and the second interaction operation; the target first operation feature can be regarded as the unique feature of the first interaction operation, and the target second operation feature can be regarded as the unique feature of the second interaction operation.

[0235] Among them, the second interaction operation is a pre-operation on which the first interaction operation depends, and there is a dependency relationship between the two in terms of operation time. The operation time of the second interaction operation is before the operation time of the first interaction operation. For example, in a specific scenario, the target content is the content recommended to the target object. The second interaction operation is a click operation on the content, and the first interaction operation is a reading duration or sharing operation on the content, etc. Predicting the interaction result of the second interaction operation can specifically be predicting whether the target object clicks on the target content. Predicting the interaction result of the first interaction operation can specifically be predicting the reading duration of the target object on the target content, or whether the target object will share the target content. It can be understood that the target object needs to click on the target content first before reading or sharing the target content. Therefore, the click is the pre-operation corresponding to the reading duration or sharing.

[0236] Optionally, in this embodiment, the step of "performing feature analysis processing on the interaction operation of the target content to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature of the target content on the second interaction operation, and the target sharing operation feature" may include:

[0237] Performing feature analysis processing on the interaction operation of the target content through a content interaction prediction model to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature of the target content on the second interaction operation, and the target sharing operation feature.

[0238] Among them, the content interaction prediction model is specifically a neural network model. The neural network can be a Residual Network (ResNet), or a Dense Convolutional Network (DenseNet), etc. It should be understood that the neural network in this embodiment is not limited to the several types listed above.

[0239] Optionally, in this embodiment, the step of "performing feature analysis processing on the interaction operation of the target content to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature of the target content on the second interaction operation, and the target sharing operation feature" may include:

[0240] Extracting the first operation feature of the target content on the first interaction operation, the second operation feature of the target content on the second interaction operation, and the sharing operation feature;

[0241] Performing feature interaction processing on the first operation feature, the second operation feature, and the sharing operation feature to obtain the target first operation feature corresponding to the first interaction operation, the target second operation feature corresponding to the second interaction operation, and the target sharing operation feature.

[0242] Specifically, the content information of the target content in each dimension can be input into the content interaction prediction model. The content information of the target content in each dimension may include the content title, content publisher information, content cover, content itself information, etc.

[0243] Optionally, in this embodiment, the step of "performing feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain the target first operation feature corresponding to the first interaction operation, the target second operation feature corresponding to the second interaction operation, and the target shared operation feature" may include:

[0244] Fuse the first operation feature and the shared operation feature, and update the first operation feature based on the fused feature;

[0245] Fuse the second operation feature and the shared operation feature, and update the second operation feature based on the fused feature;

[0246] Fuse the first operation feature, the second operation feature, and the shared operation feature, and update the shared operation feature based on the fused feature;

[0247] Return to execute the step of fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature until the target shared operation feature that meets the preset feature interaction condition, the target first operation feature corresponding to the first interaction operation, and the target second operation feature corresponding to the second interaction operation are obtained.

[0248] Among them, the preset feature interaction condition can be set according to the actual situation, and this embodiment does not limit it. For example, the preset feature interaction condition can specifically be that the number of update times reaches the preset number of times. In some embodiments, the preset feature interaction condition can be determined according to the number of layers of the feature extraction network in the content interaction prediction model.

[0249] Among them, the first operation feature and the shared operation feature can be fused through a gating network; there are various fusion methods, and this embodiment does not limit it. For example, the fusion method can be weighted operation or splicing, etc. And input the fused feature into the next layer of task-specific expert T, and update the first operation feature based on the output feature of the next layer of task-specific expert T. Specifically, the output feature processed by the next layer of task-specific expert T can be determined as the new first operation feature.

[0250] Among them, the second operation feature and the shared operation feature can be fused through a gating network; there are various fusion methods, which are not limited in this embodiment. For example, the fusion method can be weighted operation or splicing, etc. And the fused feature is input into the next-layer task-specific expert C, and based on the output feature of the next-layer task-specific expert C, the second operation feature is updated. Specifically, the output feature processed by the next-layer task-specific expert C can be determined as the new second operation feature.

[0251] Among them, the first operation feature, the second operation feature and the shared operation feature can be fused through a gating network; there are various fusion methods, which are not limited in this embodiment. For example, the fusion method can be weighted operation or splicing, etc. And the fused feature is input into the next-layer task-sharing expert, and based on the output feature of the next-layer task-sharing expert, the shared operation feature is updated. Specifically, the output feature processed by the next-layer task-sharing expert can be determined as the new shared operation feature.

[0252] 202. The server determines an initial predicted interaction result of the target content on the first interaction operation based on the target first operation feature.

[0253] Optionally, in this embodiment, the step of "determining an initial predicted interaction result of the target content on the first interaction operation based on the target first operation feature" may include:

[0254] Fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0255] Perform a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation;

[0256] Based on the target feature information, the target shared operation feature and the target first operation feature, determine an initial predicted interaction result of the target content on the first interaction operation.

[0257] Among them, the target second operation feature and the target shared operation feature can be fused through a gating network. The fusion method can be weighted operation or splicing process, etc., which are not limited in this embodiment. And the obtained second fused feature information is used as the input of the tower network C in the content interaction prediction model. After the fully connected process of the tower network C, the target feature information corresponding to the second interaction operation is obtained.

[0258] Optionally, in this embodiment, the step of "determining an initial predicted interaction result of the target content on the first interaction operation based on the target feature information, the target shared operation feature and the target first operation feature" may include:

[0259] Fuse the target shared operation feature and the target first operation feature to obtain first fused feature information;

[0260] Perform feature selection processing on the second fused feature information to obtain second feature information;

[0261] Based on the target feature information, the first fused feature information, and the second feature information, determine an initial predicted interaction result of the target content in the first interaction operation.

[0262] Among them, the target shared operation feature and the target first operation feature can be fused through a gated network. This fusion method can be weighted operation, splicing processing, etc., and this embodiment does not limit this.

[0263] Among them, the second fused feature information can be used as the input of the gated network g in the content interaction prediction model. r through the gated network g r Perform feature selection processing to obtain second feature information.

[0264] Among them, the step of "Based on the target feature information, the first fused feature information, and the second feature information, determine an initial predicted interaction result of the target content in the first interaction operation" may include:

[0265] Fuse the target feature information, the first fused feature information, and the second feature information to obtain a target interaction feature;

[0266] Based on the target interaction feature, determine an initial predicted interaction result of the target content in the first interaction operation.

[0267] Among them, there are various fusion methods, which can be weighted fusion, splicing processing, etc., and this embodiment does not limit this. And use the fused target interaction feature as the input of the tower network T in the content interaction prediction model. After being processed by the tower network T, obtain the initial predicted interaction result of the target content in the first interaction operation.

[0268] Among them, in some embodiments, the preset interaction result with the highest probability can be determined as the initial predicted interaction result of the target content in the first interaction operation; in other embodiments, it can also be the preset interaction result with a probability greater than the preset value is determined as the initial predicted interaction result of the target content in the first interaction operation.

[0269] 203. The server performs feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0270] Optionally, in this embodiment, the step of "performing feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation" may include:

[0271] Fusing the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0272] Performing a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation;

[0273] Fusing the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0274] Among them, the target second operation feature and the target shared operation feature can be fused through a gating network. This fusion method can be weighted operation, splicing process, etc., and this embodiment does not limit this. And the second fused feature information obtained by fusion is used as the input of the tower network C in the content interaction prediction model. After the fully connected process of the tower network C, the target feature information corresponding to the second interaction operation is obtained.

[0275] Optionally, in this embodiment, the step of "fusing the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation" may include:

[0276] Performing feature selection processing on the target shared operation feature and the second fused feature information respectively to obtain first feature information and second feature information;

[0277] Fusing the first feature information, the second feature information, and the target feature information to obtain an enhanced operation feature for the second interaction operation.

[0278] Among them, in some embodiments, the target shared operation feature can be subjected to feature selection processing through a gating network to obtain first feature information.

[0279] Among them, the second fused feature information can be used as the input of the gating network g r in the content interaction prediction model, and feature selection processing is performed through the gating network g r to obtain second feature information.

[0280] Among them, there are various ways to fuse the first feature information, the second feature information, and the target feature information, and this embodiment does not limit this. For example, the fusion method can be splicing processing or weighted operation, etc. Through the fusion processing, the enhanced operation feature for the second interaction operation can be obtained.

[0281] 204. The server predicts the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature.

[0282] Among them, since the enhanced operation feature fuses the feature output of the tower network C corresponding to the second interaction operation, the enhanced operation feature is the feature enhanced for the second interaction operation.

[0283] Among them, in this embodiment, the negative noise information generated by the second interaction operation on the first interaction operation can be predicted through the negative noise modeling module in the content interaction prediction model based on the enhanced operation feature. Specifically, the enhanced operation feature can be input into the negative noise modeling module, and the negative noise modeling module processes it to predict the negative noise information generated by the second interaction operation on the first interaction operation.

[0284] Among them, the negative noise modeling module can include a multi-layer perceptron (MLP, Multilayer Perceptron). The enhanced operation feature is fully connected through the multi-layer perceptron to predict the negative noise information generated by the second interaction operation on the first interaction operation.

[0285] 205. The server corrects the initial predicted interaction result based on the negative noise information to obtain the target predicted interaction result of the target content on the first interaction operation.

[0286] Among them, the negative noise information includes the negative impact of the second interaction operation on the first interaction operation. The correction of the initial predicted interaction result can specifically be subtracting the negative noise information from the initial predicted interaction result, and then the target predicted interaction result of the target content on the first interaction operation can be obtained. This target predicted interaction result is the predicted value after subtracting the negative impact.

[0287] Specifically, if the first interaction operation is the reading duration and the second interaction operation is the click operation, through correction, the negative impact can be subtracted from the original predicted value of the reading duration (i.e., the initial predicted interaction result) to achieve the purpose of alleviating the negative impact of the click operation on the reading duration and strengthening the positive impact.

[0288] In a specific embodiment, the content interaction prediction method provided by the present application can be applied to a content recommendation scenario. After obtaining the target prediction interaction results of each content on the first interaction operation, the target recommended content can be selected from each content according to the target prediction interaction results for recommendation.

[0289] As can be seen from the above, in this embodiment, the server can obtain the target content; and perform feature analysis processing on the target content in terms of interaction operations to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature. The target shared operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends. Based on the target first operation feature, determine the initial prediction interaction result of the target content on the first interaction operation; perform feature enhancement processing on the target second operation feature and the target shared operation feature to obtain the enhanced operation feature for the second interaction operation; predict the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; and correct the initial prediction interaction result based on the negative noise information to obtain the target prediction interaction result of the target content on the first interaction operation. The present application can capture the negative noise information generated by the second interaction operation on the first interaction operation to correct the prediction interaction result of the target content on the first interaction operation, thereby improving the prediction accuracy of the interaction result on the first interaction operation.

[0290] To better implement the above method, an embodiment of the present application further provides a content interaction prediction device, as Figure 3 shown. The content interaction prediction device may include an acquisition unit 301, a determination unit 302, an enhancement unit 303, a prediction unit 304, and a correction unit 305, as follows:

[0291] (1) Acquisition unit 301;

[0292] The acquisition unit is configured to acquire the target content; and perform feature analysis processing on the target content in terms of interaction operations to obtain the target first operation feature of the target content on the first interaction operation, the target second operation feature on the second interaction operation, and the target shared operation feature. The target shared operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends.

[0293] Optionally, in some embodiments of the present application, the acquisition unit may include a feature extraction subunit and a feature interaction subunit, as follows:

[0294] The feature extraction subunit is used to extract the first operation feature of the target content in the first interaction operation, the second operation feature in the second interaction operation, and the shared operation feature;

[0295] The feature interaction subunit is used to perform feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain the target first operation feature corresponding to the first interaction operation, the target second operation feature corresponding to the second interaction operation, and the target shared operation feature.

[0296] Optionally, in some embodiments of the present application, the feature interaction subunit may specifically be used to fuse the first operation feature and the shared operation feature, and update the first operation feature based on the fused feature; fuse the second operation feature and the shared operation feature, and update the second operation feature based on the fused feature; fuse the first operation feature, the second operation feature, and the shared operation feature, and update the shared operation feature based on the fused feature; return to execute the step of fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature, until the target shared operation feature that meets the preset feature interaction condition, the target first operation feature corresponding to the first interaction operation, and the target second operation feature corresponding to the second interaction operation are obtained.

[0297] (2) Determination unit 302;

[0298] The determination unit is used to determine the initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature.

[0299] Optionally, in some embodiments of the present application, the determination unit may include a third fusion subunit, a second fully connected subunit, and a first determination subunit, as follows:

[0300] The third fusion subunit is used to fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0301] The second fully connected subunit is used to perform a fully connected process on the second fused feature information to obtain the target feature information corresponding to the second interaction operation;

[0302] The first determination subunit is used to determine the initial predicted interaction result of the target content in the first interaction operation based on the target feature information, the target shared operation feature, and the target first operation feature.

[0303] Optionally, in some embodiments of the present application, the first determination subunit may specifically be configured to fuse the target shared operation feature and the target first operation feature to obtain first fused feature information; perform feature selection processing on the second fused feature information to obtain second feature information; and determine an initial predicted interaction result of the target content on the first interaction operation according to the target feature information, the first fused feature information, and the second feature information.

[0304] (3) Enhancement unit 303;

[0305] The enhancement unit is configured to perform feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0306] Optionally, in some embodiments of the present application, the enhancement unit may include a first fusion subunit, a first fully connected subunit, and a second fusion subunit, as follows:

[0307] The first fusion subunit is configured to fuse the target second operation feature and the target shared operation feature to obtain second fused feature information;

[0308] The first fully connected subunit is configured to perform a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation;

[0309] The second fusion subunit is configured to fuse the target feature information with the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

[0310] Optionally, in some embodiments of the present application, the second fusion subunit may specifically be configured to perform feature selection processing on the target shared operation feature and the second fused feature information respectively to obtain first feature information and second feature information; and fuse the first feature information, the second feature information, and the target feature information to obtain an enhanced operation feature for the second interaction operation.

[0311] (4) Prediction unit 304;

[0312] The prediction unit is configured to predict negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature.

[0313] (5) Correction unit 305;

[0314] The correction unit is configured to correct the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content on the first interaction operation.

[0315] Optionally, in some embodiments of the present application, the obtaining unit may specifically be configured to perform feature analysis processing on the target content in terms of interaction operations through a content interaction prediction model, so as to obtain a target first operation feature of the target content in a first interaction operation, a target second operation feature in a second interaction operation, and a target sharing operation feature.

[0316] Optionally, in some embodiments of the present application, the content interaction prediction device may further include a training unit, which is configured to train the content interaction prediction model; specifically, the training unit may include a data acquisition subunit, a feature analysis subunit, a second determination subunit, a reinforcement subunit, a prediction subunit, and an adjustment subunit, as follows:

[0317] The data acquisition subunit is configured to acquire training data, where the training data includes sample content, a first expected interaction result of the sample content in the first interaction operation, and a second expected interaction result in the second interaction operation;

[0318] The feature analysis subunit is configured to perform feature analysis processing on the sample content in terms of interaction operations through a content interaction prediction model, so as to obtain a target first operation feature of the sample content in a first interaction operation, a target second operation feature in a second interaction operation, and a target sharing operation feature;

[0319] The second determination subunit is configured to respectively determine an initial first actual interaction result of the sample content in the first interaction operation and a second actual interaction result in the second interaction operation based on the target first operation feature and the target second operation feature;

[0320] The reinforcement subunit is configured to perform feature reinforcement processing on the target second operation feature and the target sharing operation feature to obtain a reinforced sample operation feature for the second interaction operation;

[0321] The prediction subunit is configured to predict actual negative noise information generated by the second interaction operation on the first interaction operation according to the reinforced sample operation feature; and based on the actual negative noise information, correct the initial first actual interaction result to obtain a target first actual interaction result;

[0322] The adjustment subunit is configured to adjust parameters of the content interaction prediction model according to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result, so as to obtain a trained content interaction prediction model.

[0323] Optionally, in some embodiments of the present application, the adjustment subunit may specifically be configured to calculate a first loss value between the initial first actual interaction result and the first expected interaction result; calculate a second loss value between the target first actual interaction result and the first expected interaction result; calculate a third loss value between the second actual interaction result and the second expected interaction result; and adjust the parameters of the content interaction prediction model according to the first loss value, the second loss value, and the third loss value to obtain a trained content interaction prediction model.

[0324] Optionally, in some embodiments of the present application, the step of "calculating a second loss value between the target first actual interaction result and the first expected interaction result" may include:

[0325] Calculate weight information corresponding to the sample content based on the first loss value and the third loss value;

[0326] Perform loss calculation on the target first actual interaction result and the first expected interaction result to obtain an initial second loss value;

[0327] Fuse the weight information and the initial second loss value to obtain a second loss value.

[0328] As can be seen from the above, in this embodiment, the acquisition unit 301 can be used to acquire target content; and perform feature analysis processing on the interaction operation of the target content to obtain a target first operation feature of the target content in the first interaction operation, a target second operation feature of the target content in the second interaction operation, and a target shared operation feature, where the target shared operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; the determination unit 302 is used to determine an initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature; the reinforcement unit 303 is used to perform feature reinforcement processing on the target second operation feature and the target shared operation feature to obtain a reinforced operation feature for the second interaction operation; the prediction unit 304 is used to predict negative noise information generated by the second interaction operation on the first interaction operation according to the reinforced operation feature; and the correction unit 305 is used to correct the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content in the first interaction operation. The present application can correct the predicted interaction result of the target content in the first interaction operation by capturing negative noise information generated by the second interaction operation on the first interaction operation, thereby improving the prediction accuracy of the interaction result in the first interaction operation.

[0329] The embodiments of the present application further provide an electronic device, such as Figure 4As shown, it shows a schematic structural diagram of an electronic device involved in an embodiment of the present application. The electronic device can be a terminal or a server, etc. Specifically:

[0330] The electronic device may include a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, an input unit 404, and other components. Those skilled in the art can understand that Figure 4 the structural diagram of the electronic device shown in does not constitute a limitation on the electronic device. It may include more or fewer components than shown, or combine certain components, or have different component arrangements. Among them:

[0331] The processor 401 is the control center of the electronic device, connecting various parts of the entire electronic device through various interfaces and circuits. By running or executing software programs and / or modules stored in the memory 402, and by calling the data stored in the memory 402, it executes various functions of the electronic device and processes data. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor may not be integrated into the processor 401 either.

[0332] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. Among them, the program storage area may store the operating system, application programs required for at least one function (such as the sound playback function, image playback function, etc.); the data storage area may store data created according to the use of the electronic device. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0333] The electronic device also includes a power supply 403 that powers each component. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 403 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0334] The electronic device may further include an input unit 404, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0335] Although not shown, the electronic device may further include a display unit and the like, which will not be elaborated here. Specifically, in this embodiment, the processor 401 in the electronic device will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402 to implement various functions as follows:

[0336] Obtain the target content; perform feature analysis processing on the interaction features of the target content to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature, where the target sharing operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends; based on the target first operation feature, determine the initial predicted interaction result of the target content in the first interaction operation; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced operation feature for the second interaction operation; predict the negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; and correct the initial predicted interaction result based on the negative noise information to obtain the target predicted interaction result of the target content in the first interaction operation.

[0337] For the specific implementation of each of the above operations, reference may be made to the previous embodiments, which will not be elaborated here.

[0338] As can be seen from the above, this embodiment can obtain the target content; and perform feature analysis processing on the target content in terms of interaction operations to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature, where the target sharing operation feature is the feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is the pre-operation on which the first interaction operation depends; based on the target first operation feature, determine the initial predicted interaction result of the target content in the first interaction operation; perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced operation feature for the second interaction operation; according to the enhanced operation feature, predict the negative noise information generated by the second interaction operation on the first interaction operation; based on the negative noise information, correct the initial predicted interaction result to obtain the target predicted interaction result of the target content in the first interaction operation. This application can correct the predicted interaction result of the target content in the first interaction operation by capturing the negative noise information generated by the second interaction operation on the first interaction operation, thereby improving the prediction accuracy of the interaction result in the first interaction operation.

[0339] Those of ordinary skill in the art can understand that all or part of the steps in the above various methods of the embodiments can be completed by instructions or by controlling relevant hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0340] Therefore, an embodiment of this application provides a computer-readable storage medium, which stores multiple instructions that can be loaded by a processor to execute the steps in any content interaction prediction method provided by the embodiment of this application. For example, the instructions can execute the following steps:

[0341] Obtain target content; and perform feature analysis processing on the target content in terms of interaction operations, to obtain a target first operation feature of the target content in a first interaction operation, a target second operation feature of the target content in a second interaction operation, and a target shared operation feature, where the target shared operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; determine an initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature; perform feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation; predict, according to the enhanced operation feature, negative noise information generated by the second interaction operation on the first interaction operation; and correct the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content in the first interaction operation.

[0342] For the specific implementation of each of the above operations, reference may be made to the previous embodiments and will not be elaborated here.

[0343] Wherein, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc.

[0344] Since the instructions stored in the computer-readable storage medium can execute the steps in any of the content interaction prediction methods provided in the embodiments of the present application, the beneficial effects achievable by any of the content interaction prediction methods provided in the embodiments of the present application can be realized. For details, reference may be made to the previous embodiments and will not be elaborated here.

[0345] According to one aspect of the present application, there is provided a computer program product or a computer program, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various optional implementation manners of the above content interaction prediction aspect.

[0346] The above has introduced in detail a content interaction prediction method and related devices provided in the embodiments of the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A content interaction prediction method, characterized in that, it includes: obtaining target content; and performing feature analysis processing on the interaction operations of the target content to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature of the target content on a second interaction operation, and a target shared operation feature, where the target shared operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; determining an initial predicted interaction result of the target content on the first interaction operation based on the target first operation feature; performing feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation; predicting negative noise information generated by the second interaction operation on the first interaction operation according to the enhanced operation feature; correcting the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content on the first interaction operation.

2. The method according to claim 1, characterized in that, the performing feature analysis processing on the interaction operations of the target content to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature of the target content on a second interaction operation, and a target shared operation feature includes: extracting a first operation feature of the target content on a first interaction operation, a second operation feature of the target content on a second interaction operation, and a shared operation feature; performing feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain a target first operation feature corresponding to the first interaction operation, a target second operation feature corresponding to the second interaction operation, and a target shared operation feature.

3. The method according to claim 2, characterized in that, the performing feature interaction processing on the first operation feature, the second operation feature, and the shared operation feature to obtain a target first operation feature corresponding to the first interaction operation, a target second operation feature corresponding to the second interaction operation, and a target shared operation feature includes: fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature; fusing the second operation feature and the shared operation feature, and updating the second operation feature based on the fused feature; fusing the first operation feature, the second operation feature, and the shared operation feature, and updating the shared operation feature based on the fused feature; returning to execute the step of fusing the first operation feature and the shared operation feature, and updating the first operation feature based on the fused feature until a target shared operation feature, a target first operation feature corresponding to the first interaction operation, and a target second operation feature corresponding to the second interaction operation that meet the preset feature interaction conditions are obtained.

4. The method according to claim 1, characterized in that, Performing feature enhancement processing on the target second operation feature and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation includes: Fusing the target second operation feature and the target shared operation feature to obtain second fused feature information; Performing a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation; Fusing the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation.

5. The method according to claim 4, wherein, The fusing the target feature information and the target shared operation feature to obtain an enhanced operation feature for the second interaction operation includes: Performing feature selection processing on the target shared operation feature and the second fused feature information respectively to obtain first feature information and second feature information; Fusing the first feature information, the second feature information and the target feature information to obtain an enhanced operation feature for the second interaction operation.

6. The method according to claim 1, wherein, The determining an initial predicted interaction result of the target content on the first interaction operation based on the target first operation feature includes: Fusing the target second operation feature and the target shared operation feature to obtain second fused feature information; Performing a fully connected process on the second fused feature information to obtain target feature information corresponding to the second interaction operation; Determining an initial predicted interaction result of the target content on the first interaction operation based on the target feature information, the target shared operation feature and the target first operation feature.

7. The method according to claim 6, wherein, The determining an initial predicted interaction result of the target content on the first interaction operation based on the target feature information, the target shared operation feature and the target first operation feature includes: Fusing the target shared operation feature and the target first operation feature to obtain first fused feature information; Performing feature selection processing on the second fused feature information to obtain second feature information; Determining an initial predicted interaction result of the target content on the first interaction operation according to the target feature information, the first fused feature information and the second feature information.

8. The method according to claim 1, wherein, The performing feature analysis processing on the target content for an interaction operation to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature on a second interaction operation, and a target shared operation feature includes: Performing feature analysis processing on the target content for an interaction operation through a content interaction prediction model to obtain a target first operation feature of the target content on a first interaction operation, a target second operation feature on a second interaction operation, and a target shared operation feature.

9. The method according to claim 8, wherein, Before performing feature analysis processing on the target content through the content interaction prediction model to obtain the target first operation feature of the target content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature, the following steps are further included: Obtain training data, where the training data includes sample content, the first expected interaction result of the sample content in the first interaction operation, and the second expected interaction result in the second interaction operation; Through the content interaction prediction model, perform feature analysis processing on the sample content to obtain the target first operation feature of the sample content in the first interaction operation, the target second operation feature in the second interaction operation, and the target sharing operation feature; Based on the target first operation feature and the target second operation feature respectively, determine the initial first actual interaction result of the sample content in the first interaction operation and the second actual interaction result in the second interaction operation; Perform feature enhancement processing on the target second operation feature and the target sharing operation feature to obtain the enhanced sample operation feature for the second interaction operation; According to the enhanced sample operation feature, predict the actual negative noise information generated by the second interaction operation on the first interaction operation; and based on the actual negative noise information, correct the initial first actual interaction result to obtain the target first actual interaction result; According to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result, adjust the parameters of the content interaction prediction model to obtain the trained content interaction prediction model.

10. The method according to claim 9, wherein, The step of adjusting the parameters of the content interaction prediction model according to the initial first actual interaction result, the target first actual interaction result, the first expected interaction result, the second actual interaction result, and the second expected interaction result to obtain the trained content interaction prediction model includes: Calculate the first loss value between the initial first actual interaction result and the first expected interaction result; Calculate the second loss value between the target first actual interaction result and the first expected interaction result; Calculate the third loss value between the second actual interaction result and the second expected interaction result; According to the first loss value, the second loss value, and the third loss value, adjust the parameters of the content interaction prediction model to obtain the trained content interaction prediction model.

11. The method according to claim 10, wherein, The step of calculating the second loss value between the target first actual interaction result and the first expected interaction result includes: Based on the first loss value and the third loss value, calculate the weight information corresponding to the sample content; Perform loss calculation on the target first actual interaction result and the first expected interaction result to obtain the initial second loss value; Fuse the weight information and the initial second loss value to obtain the second loss value.

12. A content interaction prediction device characterized in that it includes: An acquisition unit, configured to acquire target content; and perform feature analysis processing on the target content in terms of interaction operations, to obtain a target first operation feature of the target content in a first interaction operation, a target second operation feature of the target content in a second interaction operation, and a target shared operation feature, where the target shared operation feature is a feature shared by the first interaction operation and the second interaction operation, and the second interaction operation is a pre-operation on which the first interaction operation depends; A determination unit, configured to determine an initial predicted interaction result of the target content in the first interaction operation based on the target first operation feature; A reinforcement unit, configured to perform feature reinforcement processing on the target second operation feature and the target shared operation feature to obtain a reinforced operation feature for the second interaction operation; A prediction unit, configured to predict negative noise information generated by the second interaction operation on the first interaction operation according to the reinforced operation feature; A correction unit, configured to correct the initial predicted interaction result based on the negative noise information to obtain a target predicted interaction result of the target content in the first interaction operation.

13. An electronic device characterized in that it includes a memory and a processor; the memory stores an application program, and the processor is configured to run the application program in the memory to execute the operations in the content interaction prediction method according to any one of claims 1 to 11.

14. A computer-readable storage medium characterized in that the computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the content interaction prediction method according to any one of claims 1 to 11.

15. A computer program product, including a computer program or instruction characterized in that when the computer program or instruction is executed by a processor, it implements the steps in the content interaction prediction method according to any one of claims 1 to 11.

Images