Recommendation methods and systems, electronic devices and storage media

Abstract

This invention discloses a method, system, electronic device, and storage medium for recommending content. The method includes: acquiring a target user's first historical user behavior and generating first time-series behavior data based on the first historical user behavior; inputting the first time-series behavior data into a pre-trained user intention prediction model to obtain the target user's first user intention; determining multiple content items to be recommended and their corresponding interest indicators in a pre-built recommendation content library based on the first user intention, the interest indicators including the interest levels of multiple historical users for the content items to be recommended; determining the recommendation level of each content item to be recommended based on the interest indicators and the user similarity between the target user and historical users; further determining the target recommended content based on the recommendation level; and pushing the target recommended content to the target user. This invention improves the accuracy of media content delivery and the user's browsing experience, and can be widely applied in the field of computer technology.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for pushing recommended content, an electronic device, and a storage medium. Background Technology

[0002] With the rapid development of internet technology, various media content has emerged in people's lives, such as advertisements, news, and video ringtones. Current technologies recommend media content using random push or by pushing the same / related media content based on the user's browsing history. These methods, on the one hand, fail to accurately determine the target user's intentions, resulting in pushed media content that fails to arouse the target user's interest, or causing user annoyance by pushing the same media content repeatedly. On the other hand, they do not consider user feedback on media content, thus failing to select high-quality and suitable media content for the target user, affecting the accuracy of media content delivery and the user's browsing experience. Summary of the Invention

[0003] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0004] Therefore, one objective of this invention is to provide a method for recommending content, which improves the accuracy of media content delivery and the user's browsing experience.

[0005] Another objective of this invention is to provide a system for pushing recommended content.

[0006] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include:

[0007] In a first aspect, embodiments of the present invention provide a method for pushing recommended content, comprising the following steps:

[0008] Obtain the target user's first historical operation behavior, and generate first time-series behavior data based on the first historical operation behavior;

[0009] The first time-series behavioral data is input into a pre-trained user intention prediction model to obtain the first user intention of the target user.

[0010] Based on the first user's intention, multiple recommended content items and corresponding interest indicators are determined in a pre-built recommendation content library. The interest indicators include the interest of multiple historical users on the recommended content items.

[0011] The recommendation degree of each of the content to be recommended is determined based on the interest degree identifier and the user similarity between the target user and the historical users. Then, the target recommended content is determined based on the recommendation degree and the target recommended content is pushed to the target user.

[0012] The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network.

[0013] Furthermore, in one embodiment of the present invention, the step of obtaining the first historical operation behavior of the target user and generating first time-series behavior data based on the first historical operation behavior specifically includes:

[0014] Obtain the target user's historical operation records, and determine the first historical operation behavior and the corresponding operation time based on the historical operation records;

[0015] Based on the operation time, the first historical operation behavior is cleaned and sliced ​​to obtain user behavior data.

[0016] The user behavior data is sorted according to the operation time to obtain the first time-series behavior data.

[0017] Furthermore, in one embodiment of the present invention, the push method further includes the step of training the user intention prediction model, which specifically includes:

[0018] A user intention prediction model is constructed, which includes an input layer, a forward temporal convolutional network, a backward temporal convolutional network, and a fully connected layer.

[0019] Obtain the second historical operation behavior of the historical user, and determine the second time-series behavior data and the corresponding user intention tag based on the second historical operation behavior;

[0020] A training sample set is constructed based on the second time-series behavior data and the corresponding user intention tags;

[0021] The training sample set is input into the user intention prediction model for training, and the model parameters of the user intention prediction model are optimized to obtain the trained user intention prediction model.

[0022] Furthermore, in one embodiment of the present invention, the step of inputting the training sample set into the user intention prediction model for training, optimizing the model parameters of the user intention prediction model, and obtaining the trained user intention prediction model specifically includes:

[0023] The second temporal behavior data is input into the input layer, and the first temporal feature vector is obtained by convolution calculation through the forward temporal convolutional network;

[0024] The second temporal behavior data is input into the input layer in reverse order, and the second temporal feature vector is obtained by convolution calculation through the inverse temporal convolutional network.

[0025] The second time-series feature vector is reversed to obtain the third time-series feature vector, and the first time-series feature vector and the third time-series feature vector are concatenated to obtain the fourth time-series feature vector;

[0026] The fourth temporal feature vector is input into the fully connected layer, and the user intention prediction result is output.

[0027] The loss value of the user intention prediction model is determined based on the user intention prediction results and the user intention labels.

[0028] Update the model parameters of the user intention prediction model based on the loss value, and return to the step of inputting the second time-series behavioral data into the input layer;

[0029] When the loss value reaches a preset first threshold or the number of iterations reaches a preset second threshold, training stops, and the trained user intention prediction model is obtained.

[0030] Furthermore, in one embodiment of the present invention, the push method further includes the step of constructing the recommended content library, which specifically includes:

[0031] Obtain the historical browsing content and corresponding browsing duration of the historical users;

[0032] The historical browsing content with a browsing duration greater than or equal to a preset third threshold is identified as the first recommended content, and the content type and content richness of the first recommended content are determined.

[0033] The interest level of the historical users in the first recommended content is determined based on the browsing duration and the richness of the content, and the interest level identifier of the first recommended content is determined based on the interest level of each historical user in the first recommended content.

[0034] The recommendation intention index value of the first recommended content is determined according to the content type, and a first mapping relationship between the recommendation intention index value and the first recommended content is established.

[0035] The recommended content library is generated based on the first recommended content, the interest index, the recommendation intention index value, and the first mapping relationship.

[0036] Furthermore, in one embodiment of the present invention, the step of determining multiple recommended content items and their corresponding interest indicators in a pre-built recommendation content library based on the first user intention specifically includes:

[0037] Based on the first user's intention, the corresponding recommendation intention index value is obtained by traversing and matching the recommended content library.

[0038] Based on the matched recommendation intention index value and the first mapping relationship, multiple first recommended contents are indexed in the recommendation content library as the content to be recommended, and the interest degree identifier of each content to be recommended is determined.

[0039] Furthermore, in one embodiment of the present invention, the step of determining the recommendation degree of each of the content to be recommended based on the interest identifier and the user similarity between the target user and the historical user, and then determining the target recommended content based on the recommendation degree, specifically includes:

[0040] Obtain a first user profile of the target user and a second user profile of the historical user, and determine a first user feature vector based on the first user profile and a second user feature vector based on the second user profile;

[0041] The user similarity between the target user and the historical user is determined based on the cosine similarity between the first user feature vector and the second user feature vector;

[0042] Based on the interest index, the interest of each historical user on the content to be recommended is determined, and the user similarity is used as a weight to perform a weighted sum of the interest of each historical user to obtain a first weighted sum.

[0043] The user similarity scores corresponding to each of the historical users are summed to obtain a first sum value, and the recommendation score of the content to be recommended is obtained by dividing the first weighted sum by the first sum value.

[0044] The content to be recommended that has a recommendation score greater than or equal to a preset fourth threshold is determined as the target recommended content; or, the content to be recommended that has a recommendation score less than or equal to a preset fifth threshold is determined as the target recommended content.

[0045] Secondly, embodiments of the present invention provide a recommendation content push system, comprising:

[0046] The time-series behavior data generation module is used to obtain the first historical operation behavior of the target user and generate first time-series behavior data based on the first historical operation behavior.

[0047] The user intention prediction module is used to input the first time-series behavioral data into a pre-trained user intention prediction model to obtain the first user intention of the target user.

[0048] The module for determining recommended content is used to determine multiple recommended contents and corresponding interest indicators in a pre-built recommended content library based on the first user's intention. The interest indicators include the interest of multiple historical users on the recommended contents.

[0049] The recommendation degree determination module is used to determine the recommendation degree of each of the content to be recommended based on the interest degree identifier and the user similarity between the target user and the historical users, and then determine the target recommended content based on the recommendation degree, and push the target recommended content to the target user;

[0050] The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network.

[0051] Thirdly, embodiments of the present invention provide an electronic device, the electronic device including a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for implementing communication between the processor and the memory, wherein when the program is executed by the processor, it implements the method for pushing recommended content as described in the first aspect above.

[0052] Fourthly, embodiments of the present invention also provide a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the recommended content push method as described in the first aspect above.

[0053] The advantages and beneficial effects of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention:

[0054] This invention acquires the first historical operation behavior of a target user and generates first time-series behavior data based on this behavior. The first time-series behavior data is then input into a pre-trained user intention prediction model to obtain the target user's first user intention. Based on this first user intention, multiple content items to be recommended and their corresponding interest indicators are determined from a pre-built recommendation content library. Furthermore, the recommendation degree of each content item is determined based on the interest indicators and the user similarity between the target user and historical users. Thus, target recommended content can be determined based on the recommendation degree and pushed to the target user. This invention generates time-series behavior data based on the target user's historical operation data and inputs it into the user intention prediction model, which can accurately predict the target user's user intention, avoiding the push of media content that fails to arouse the target user's interest or even causes annoyance. Based on this user intention, multiple content items to be recommended can be selected from the recommendation content library. The recommendation degree of the content items to be recommended is then determined based on their interest indicators and the user similarity between the target user and historical users. This allows for the selection of high-quality and suitable media content to be pushed to the target user, improving the accuracy of media content delivery and the user's browsing experience. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart illustrating the steps of a method for pushing recommended content, as provided in an embodiment of the present invention;

[0057] Figure 2 A flowchart of step S101 provided in an embodiment of the present invention;

[0058] Figure 3 A flowchart illustrating the steps of training the user intention prediction model according to an embodiment of the present invention;

[0059] Figure 4 Another flowchart for the steps of training the user intention prediction model provided in an embodiment of the present invention;

[0060] Figure 5 A flowchart illustrating the steps for constructing the recommended content library as provided in an embodiment of the present invention;

[0061] Figure 6 A flowchart of step S103 provided in an embodiment of the present invention;

[0062] Figure 7 A flowchart of step S104 provided in an embodiment of the present invention;

[0063] Figure 8 A schematic diagram of the structure of the recommended content push system provided in the embodiments of the present invention;

[0064] Figure 9 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0065] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. It should be noted that although functional modules are divided in the system schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system schematic diagram or the order in the flowchart. The step numbers in the following embodiments are only set for ease of explanation and do not limit the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0066] In the description of this invention, "multiple" means two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, the number of indicated technical features, or the order of the indicated technical features. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0067] The method for pushing recommended content provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the method for pushing recommended content, but is not limited to the above forms.

[0068] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0069] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0070] like Figure 1 The diagram shown is a flowchart of the steps of a method for pushing recommended content provided by an embodiment of the present invention. (Refer to...) Figure 1 This invention provides a method for pushing recommended content, specifically including the following steps:

[0071] S101. Obtain the first historical operation behavior of the target user, and generate first time-series behavior data based on the first historical operation behavior.

[0072] Specifically, when pushing media content to target users, this embodiment of the invention needs to predict and identify user intentions. Therefore, this embodiment of the invention obtains the target user's historical operation data and generates time-series behavior data, which is then used as input data into the user intention prediction model to predict user intentions.

[0073] like Figure 2 The diagram shown is a flowchart of step S101 provided in an embodiment of the present invention. (Refer to...) Figure 2As an optional implementation, the step of acquiring the target user's first historical operation behavior and generating first time-series behavior data based on the first historical operation behavior specifically includes:

[0074] S1011. Obtain the target user's historical operation records, and determine the first historical operation behavior and the corresponding operation time based on the historical operation records;

[0075] S1012. Perform data cleaning and data slicing on the first historical operation behavior according to the operation time to obtain user behavior data;

[0076] S1013. Sort the user behavior data according to the operation time to obtain the first time-series behavior data.

[0077] Specifically, the historical operation records of the target user are obtained. These records can be operation records of the target user on any one or more devices, including user terminals, flat-screen computers, PC hosts, in-vehicle computers, portable smart terminals, smart home devices, and wearable devices. These historical operation records include multiple historical operation behaviors, such as downloading, registering, uninstalling, searching, clicking, canceling, taking screenshots, recording screens, sharing, and making payments. When obtaining the target user's historical operation records, the operation time for each historical operation behavior also needs to be obtained. In some optional embodiments, a time period threshold for obtaining historical operation records can be preset; for example, the time period threshold can be set to one year, and only historical operation records from the past year can be obtained.

[0078] After acquiring historical operation records, data cleaning and data slicing of the user's historical operation behavior are required. Specifically, data cleaning is used to remove erroneous operations from historical operation behavior, thereby retaining historical operation behavior with the target user's subjective intent. In this embodiment of the invention, erroneous operations can be removed based on operation time to achieve data cleaning. Specifically, the existence of erroneous operations can be determined by whether the operation time and behavior logic of two adjacent historical operation behaviors are mutually exclusive. For example, if the historical operation behavior of "downloading software A" and the historical operation behavior of "uninstalling software A" are adjacent logically mutually exclusive operations, and the time interval between their operations is 10 seconds, then the historical operation behavior of "downloading software A" can be determined as an erroneous operation and can therefore be removed. The historical operation behavior of "uninstalling software A" can be removed together as a response to the erroneous operation behavior, or it can be retained separately to represent the user's intent. Similarly, logically mutually exclusive operations include screen recording and canceling screen recording, payment and canceling payment, etc., which will not be illustrated in detail in this embodiment of the invention.

[0079] Data slicing is used to sample historical operation behaviors at preset time intervals. For example, a user behavior data is obtained by sampling a preset time period (such as 10 minutes) from multiple historical operation behaviors, and the user behavior data is sorted according to the operation time to generate time-series behavior data.

[0080] It should be noted that user behavior data includes not only operational actions but also the objects of those actions, such as "downloading software A" or "clicking link C." Understandably, this embodiment of the invention ensures the accuracy and reliability of user behavior data through data cleaning and data slicing, thereby improving the accuracy of user intention prediction in subsequent steps.

[0081] S102. Input the first time-series behavioral data into the pre-trained user intention prediction model to obtain the first user intention of the target user.

[0082] The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network.

[0083] Specifically, the user intention prediction model in this embodiment of the invention is obtained by acquiring historical users' temporal behavior data and inputting it into a forward temporal convolutional network and an inverse temporal convolutional network for joint training. The training process is described below.

[0084] like Figure 3 The diagram shown is a flowchart of one step in training a user intention prediction model according to an embodiment of the present invention. (Refer to...) Figure 3 As an optional implementation, the push method further includes a step of training a user intent prediction model, which specifically includes:

[0085] A1. Construct a user intention prediction model, which includes an input layer, a forward temporal convolutional network, a backward temporal convolutional network, and a fully connected layer.

[0086] A2. Obtain the second historical operation behavior of historical users, and determine the second time sequence behavior data and the corresponding user intention tags based on the second historical operation behavior;

[0087] A3. Construct a training sample set based on the second time-series behavioral data and the corresponding user intention tags;

[0088] A4. Input the training sample set into the user intention prediction model for training, optimize the model parameters of the user intention prediction model, and obtain the trained user intention prediction model.

[0089] Specifically, Temporal Convolutional Network (TCN) is a novel type of convolutional neural network proposed by Lea et al. in 2016, which can be used to solve time series prediction problems. In this embodiment of the invention, the user intention prediction model includes an input layer, a forward temporal convolutional network, a backward temporal convolutional network, and a fully connected layer. Each temporal behavior data is used to calculate the temporal feature vector using both the forward and backward temporal convolutional networks, and the calculation results are integrated and output through the fully connected layer.

[0090] The process of obtaining the second historical user behavior is the same as that of obtaining the first historical user behavior. The second historical user behavior also requires data cleaning and slicing to obtain the corresponding second time-series behavior data. The difference lies in that the last few historical user behaviors (i.e., those occurring latest) are used to generate user intent tags and are not included in the second time-series behavior data. For example, if a set of second historical user behaviors includes "downloading software A", "registering software A", "searching for product B", "clicking a link C containing product B", "sharing link C", "creating an order D containing product B", and "paying for order D", then the generated second time-series behavior data could be {"downloading software A", "registering software A", "searching for product B", "clicking a link C containing product B", "sharing link C", "creating an order D containing product B"}, and the corresponding user intent tag could be "purchase product B".

[0091] A training sample set can be constructed by using the second-series behavioral data of multiple different historical users and their corresponding user intention tags. This training sample set is then input into the constructed user intention prediction model, and the model parameters are optimized through iterative training to obtain the trained user intention prediction model.

[0092] like Figure 4 The diagram shown is another flowchart illustrating the steps of training a user intention prediction model according to an embodiment of the present invention. (Refer to...) Figure 4 As an optional implementation, the step of inputting the training sample set into the user intention prediction model for training, optimizing the model parameters of the user intention prediction model, and obtaining a trained user intention prediction model specifically includes:

[0093] A41. Input the second temporal behavioral data into the input layer, and perform convolution calculation through a forward temporal convolutional network to obtain the first temporal feature vector;

[0094] A42. Input the second temporal behavioral data into the input layer in reverse order, and perform convolution calculation through the inverse temporal convolutional network to obtain the second temporal feature vector;

[0095] A43. The second time series feature vector is reversed to obtain the third time series feature vector, and the first time series feature vector and the third time series feature vector are concatenated to obtain the fourth time series feature vector;

[0096] A44. Input the fourth temporal feature vector into the fully connected layer and output the user intention prediction result.

[0097] A45. Determine the loss value of the user intention prediction model based on the user intention prediction results and user intention labels;

[0098] A46. Update the model parameters of the user intention prediction model based on the loss value and return to the step of inputting the second time series behavioral data into the input layer;

[0099] A47. When the loss value reaches the preset first threshold or the number of iterations reaches the preset second threshold, training is stopped, and the trained user intention prediction model is obtained.

[0100] Specifically, both the forward temporal convolutional network and the inverse temporal convolutional network in this embodiment of the invention include multiple convolutional layers arranged sequentially. The input layer is used to input the second temporal behavioral data and temporally segment it to obtain multiple individual behavioral data. These individual behavioral data are then input sequentially into each convolutional layer of the forward temporal convolutional network and in reverse order into each convolutional layer of the inverse temporal convolutional network. The forward and inverse temporal convolutional networks perform convolution calculations and feature extraction respectively, and output corresponding first and second temporal feature vectors. Since the second temporal feature vector is input in reverse order into the inverse temporal convolutional network... The first time-series feature vector is extracted from the convolutional network, so it needs to be reversed to obtain the third time-series feature vector. Both the third and first time-series feature vectors can be used to represent the user intentions corresponding to the input sample data. The first and third time-series feature vectors are concatenated to obtain the fourth time-series feature vector. The fourth time-series feature vector combines the feature extraction results of the forward and inverse time-series convolutional networks. Then, the fourth time-series feature vector is input into the fully connected layer, and the probability values ​​of different user intentions can be calculated through the activation function, thereby outputting the corresponding user intention prediction results.

[0101] During model iteration, the accuracy of model predictions can be evaluated based on the user intention prediction results and the aforementioned user intention labels, thereby updating the model parameters. For user intention prediction models, the accuracy of the model's prediction results can be measured by a loss function. The loss function is defined on a single training data point and measures the prediction error of that training data. Specifically, the loss value of that training data point is determined by the label of that individual training data point and the model's prediction result for that training data point. In actual training, a training sample set contains many training samples. Therefore, a cost function is generally used to measure the overall error of the training sample set. The cost function is defined on the entire training sample set and is used to calculate the average prediction error of all training samples, providing a better measure of the model's prediction performance. For general machine learning models, the aforementioned cost function, plus a regularization term to measure model complexity, can serve as the training objective function. Based on this objective function, the loss value for the entire training sample set can be calculated. There are many commonly used loss functions, such as 0-1 loss, squared loss, absolute loss, logarithmic loss, and cross-entropy loss, all of which can be used as loss functions for machine learning models. These will not be elaborated upon here. In this embodiment of the invention, any one of these loss functions can be selected to determine the training loss value. Based on the training loss value, the backpropagation algorithm is used to update the model parameters. After several iterations, a well-trained user intent recognition model can be obtained. The specific number of iterations can be preset, or training can be considered complete when the loss value reaches a preset threshold.

[0102] After obtaining the trained user intention prediction model, the first time-series behavioral data is input into the user intention prediction model to obtain the target user's first user intention.

[0103] S103. Based on the first user's intention, determine multiple recommended content items and their corresponding interest indicators in the pre-built recommendation content library. The interest indicators include the interest of multiple historical users in the recommended content items.

[0104] Specifically, the recommended content library in this embodiment of the invention includes multiple first recommended contents, each of which has a corresponding interest index, which is composed of the interest of multiple historical users in that first recommended content. Furthermore, each first recommended content has a corresponding recommendation intention index value. The correspondence can be one-to-one, one-to-many, many-to-one, or many-to-many. This embodiment of the invention does not limit the correspondence between the first recommended content and the recommendation intention index value.

[0105] like Figure 5 The diagram shown is a flowchart of one step in constructing a recommended content library according to an embodiment of the present invention. (Refer to...) Figure 5As an optional implementation, the push method further includes the step of building a recommended content library, which specifically includes:

[0106] B1. Obtain the browsing history and corresponding browsing duration of historical users;

[0107] B2. Determine the historical browsing content with a browsing duration greater than or equal to the preset third threshold as the first recommended content, and determine the content type and content richness of the first recommended content;

[0108] B3. Determine the level of interest of historical users in the first recommended content based on browsing time and content richness, and determine the interest index of the first recommended content based on the interest of each historical user in the first recommended content.

[0109] B4. Determine the recommendation intention index value of the first recommended content based on the content type, and establish the first mapping relationship between the recommendation intention index value and the first recommended content;

[0110] B5. Generate a recommended content library based on the first recommended content, interest level identifier, recommendation intention index value, and the first mapping relationship.

[0111] Specifically, the historical browsing content and corresponding browsing duration of historical users are obtained. Considering that historical browsing content with too short a browsing duration may be due to accidental operation, a third threshold (such as 5 seconds) can be preset to filter out historical browsing content with a browsing duration lower than the third threshold, and the remaining content is added to the recommendation content library as the first recommended content.

[0112] Determine the content type of the primary recommended content, such as marketing advertisements, emotional problem counseling, emotion management courses, immersive scenario experiences, and trending public opinion. Determine the content richness of the primary recommended content. This richness can be calculated using different methods for different presentation formats. For example, for text-based recommended content, richness can be determined by the number of words, with a positive correlation between the two; for video-based recommended content, richness can be determined by the video length, with a positive correlation between the two.

[0113] The interest level of historical users towards the first recommended content is determined based on browsing duration and content richness. It can be understood that a longer browsing duration indicates a higher level of interest in the first recommended content. Furthermore, browsing duration is closely related to content richness; for two recommended content items with the same level of interest, the recommended content with richer content will inevitably require a longer browsing time. In this embodiment of the invention, interest level is calculated using the following formula:

[0114]

[0115] Where r represents interest level, T represents browsing time, P represents content richness, and μ represents a preset coefficient, which can be calibrated through multiple experiments.

[0116] Based on the content type, the recommendation intention index value of the first recommended content can be determined, and a mapping relationship between the two can be established. As mentioned before, the mapping relationship can be a one-to-one correspondence, or a one-to-many, many-to-one, or many-to-many cross-mapping. In this embodiment of the invention, each recommendation intention index value can correspond to multiple first recommended contents simultaneously, and each first recommended content can correspond to multiple recommendation intention index values ​​simultaneously.

[0117] A recommended content library can be generated based on the first recommended content, interest level identifier, recommendation intention index value, and the first mapping relationship.

[0118] In some optional embodiments, different recommendation content libraries can be established for different user groups. For example, for teenagers, to help them establish correct worldviews, outlooks on life, and values, the recommendation content library may include social sciences, humanities and history, literary works, philosophy of life, etc.; for university students, the recommendation content library may include movies, sports news, science and technology journals, immersive scenario game experiences, plot creative designs, etc.; for middle-aged people, the recommendation content library may include investment and financial management, home products, emotional problem counseling, emotion management courses, children's education, etc. It should be noted that the above content is only an example illustrating a specific implementation scenario of the recommendation content library of the present invention and should not be considered as a limitation of the present invention. In specific implementations, the construction of the recommendation content library can be carried out according to the actual situation.

[0119] like Figure 6 The diagram shown is a flowchart of step S103 provided in an embodiment of the present invention. (Refer to...) Figure 6 As an optional implementation, the step of determining multiple recommended content items and their corresponding interest indicators from a pre-built recommendation content library based on the first user's intention specifically includes:

[0120] S1031. Based on the first user's intention, perform a traversal and matching in the recommended content library to obtain the corresponding recommendation intention index value;

[0121] S1032. Based on the matched recommendation intention index value and the first mapping relationship, index multiple first recommended contents in the recommendation content library as recommended contents, and determine the interest degree identifier of each recommended content.

[0122] Specifically, the first user intent identified in the aforementioned steps is traversed and matched in the recommended content library to obtain the corresponding recommended intent index value. For example, if the first user intent is "to purchase product B", the matched recommended intent index value can be "introduction of product B's functions", "how to use product B", "purchase link of product B", "products that go with product B", etc., or it can simply match the recommended intent index value of "purchase link of product B". The specific matching rules are set according to the actual situation, and this embodiment of the invention does not limit this.

[0123] After matching the corresponding recommendation intention index value, multiple first recommended contents can be quickly indexed in the recommendation content library based on the first mapping relationship according to the recommendation intention index value as to be recommended. At the same time, it is also necessary to obtain the interest degree identifier corresponding to the content to be recommended for subsequent recommendation degree calculation.

[0124] S104. Determine the recommendation level of each content to be recommended based on the interest level identifier and the user similarity between the target user and historical users. Then, determine the target recommended content based on the recommendation level and push the target recommended content to the target user.

[0125] Specifically, in this embodiment of the invention, the similarity between the target user and historical users is calculated, and the recommendation degree of each content to be recommended is calculated based on the similarity and the interest identifier determined in the aforementioned steps, thereby determining the target recommended content and pushing it.

[0126] like Figure 7 The diagram shown is a flowchart of step S104 provided in an embodiment of the present invention. (Refer to...) Figure 7 As an optional implementation, the step of determining the recommendation level of each piece of content to be recommended based on interest identifiers and user similarity between the target user and historical users, and then determining the target recommended content based on the recommendation level, specifically includes:

[0127] S1041. Obtain the first user profile of the target user and the second user profile of the historical users, and determine the first user feature vector based on the first user profile and the second user feature vector based on the second user profile.

[0128] S1042. Determine the user similarity between the target user and historical users based on the cosine similarity between the first user feature vector and the second user feature vector;

[0129] S1043. Determine the interest level of each historical user in the recommended content based on the interest level identifier, and use user similarity as a weight to perform a weighted sum of the interest levels of each historical user to obtain the first weighted sum.

[0130] S1044. Sum the user similarity corresponding to each historical user to obtain the first sum value, and divide the first weighted sum by the first sum value to obtain the recommendation degree of the content to be recommended.

[0131] S1045. Determine the content to be recommended as the target recommended content if the recommendation score is greater than or equal to the preset fourth threshold, or determine the content to be recommended as the target recommended content if the recommendation score is less than or equal to the preset fifth threshold.

[0132] Specifically, a user profile is a labeled user model abstracted from information such as user demographics, online browsing content, online social activities, and consumption behavior. It reflects the characteristics of different users. The content of a user profile is not entirely fixed; the features emphasized vary depending on the industry and product. For most internet companies, user profiles include demographic attributes and behavioral characteristics. Demographic attributes mainly refer to the user's age, gender, province and city, education level, marital status, fertility status, industry and occupation, etc.; behavioral characteristics mainly include indicators such as activity level and loyalty. In addition to these more general features, different types of websites extract user profiles with different focuses. Content-driven media or reading websites, as well as search engines or general navigation websites, often extract user interest characteristics for the content they browse, such as sports, entertainment, food, finance, travel, real estate, automobiles, etc. Social networking sites also extract user social networks, which can identify closely connected user groups and star nodes that play the role of opinion leaders in the community. E-commerce shopping websites generally extract indicators such as users' online shopping interests and spending power. Online shopping interests mainly refer to users' category preferences when shopping online, such as clothing, bags, home goods, maternity and baby products, personal care products, food, etc.

[0133] In this embodiment of the invention, a first user profile of the target user and a second user profile of the historical user are constructed based on existing user profiling technology. Both the first user profile and the second user profile have multiple labels with the same dimensions, such as age, gender, occupation, education level, etc. Vectorization processing of these labels can yield the multi-dimensional first user feature vector corresponding to the first user profile and the multi-dimensional second user feature vector corresponding to the second user profile.

[0134] The user similarity between the target user and historical users is determined based on the cosine similarity between the first user feature vector and the second user feature vector.

[0135] Cosine similarity, also known as cosine similarity, measures the similarity between two vectors by measuring the cosine of the angle between them. It can be calculated using the following formula:

[0136]

[0137] Here, similarity represents cosine similarity, and θ represents the angle between vectors A and B. When two vectors point in the same direction, the cosine similarity is 1; when the angle between the two vectors is 90°, the cosine similarity is 0; and when the two vectors point in completely opposite directions, the cosine similarity is -1. Cosine similarity is typically used in high-dimensional positive spaces, therefore its value ranges from -1 to 1. For high-dimensional feature vectors, their Euclidean distance in the feature space is usually large, while using cosine similarity might result in a small angle between them, thus leading to high similarity. Furthermore, in fields such as text, images, and videos, the feature dimensions of the objects studied are often very high. Cosine similarity maintains the property of "1 when they are the same, 0 when they are orthogonal, and -1 when they are opposite" even in high-dimensional cases, while the value of Euclidean distance is affected by the dimensionality, its range is not fixed, and its meaning is relatively ambiguous.

[0138] In this embodiment of the invention, the cosine similarity between the calculated first user feature vector and the second user feature vector is directly used as the user similarity between the target user and the historical user.

[0139] The interest level of each historical user in the recommended content is determined based on interest indicators. Then, the recommendation score of the recommended content is calculated by combining the user similarity between the target user and each historical user. The specific calculation formula is as follows:

[0140]

[0141] Among them, S i,u This represents the recommendation degree of content i to be recommended for target user u, where U represents the set of historical users ν, sim(u,v) represents the user similarity between target user u and historical user ν, and r i,v This indicates the user's interest level in the content to be recommended, i.

[0142] After calculating the recommendation score of each piece of content to be recommended, several pieces of content with a recommendation score higher than the preset fourth threshold can be selected as target recommended content, or content with a recommendation score lower than the preset fifth threshold can be selected as target recommended content, and then the target recommended content is pushed to the target user.

[0143] The method steps of the embodiments of the present invention have been described above. It is understood that the embodiments of the present invention generate time-series behavioral data based on the historical operation data of the target user and input it into the user intention prediction model. This can accurately predict the user intention of the target user, avoiding the situation where the pushed media content fails to arouse the target user's interest or even causes the target user to become annoyed. Based on this user intention, multiple recommended content items can be selected from the recommended content library. Then, the recommendation degree of the recommended content is determined based on the interest level identifier of the recommended content and the user similarity between the target user and historical users. This allows for the selection of high-quality and suitable media content to be pushed to the target user, improving the accuracy of media content push and the user's browsing experience.

[0144] like Figure 8 The diagram shown is a structural schematic of the recommended content push system provided in an embodiment of the present invention. (Refer to...) Figure 8 This invention provides a content recommendation system, comprising:

[0145] The time-series behavior data generation module is used to obtain the first historical operation behavior of the target user and generate the first time-series behavior data based on the first historical operation behavior.

[0146] The user intention prediction module is used to input the first time-series behavioral data into the pre-trained user intention prediction model to obtain the first user intention of the target user.

[0147] The module for determining recommended content is used to determine multiple recommended content items and their corresponding interest indicators from a pre-built recommended content library based on the first user's intention. The interest indicators include the interest of multiple historical users in the recommended content items.

[0148] The recommendation degree determination module is used to determine the recommendation degree of each piece of content to be recommended based on the interest degree identifier and the user similarity between the target user and historical users. Then, the target recommended content is determined based on the recommendation degree and pushed to the target user.

[0149] The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network.

[0150] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0151] This invention also provides an electronic device, comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned method for pushing recommended content. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0152] like Figure 9 The diagram shown is a hardware structure schematic of an electronic device provided in an embodiment of the present invention. (Refer to...) Figure 9 This invention provides an electronic device, comprising:

[0153] The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0154] The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is invoked by the processor 901 to execute the method for pushing recommended content according to the embodiments of this invention.

[0155] The input / output interface 903 is used to implement information input and output;

[0156] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0157] Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904);

[0158] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0159] This invention also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the above-mentioned method for pushing recommended content.

[0160] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0161] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0162] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0163] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0164] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0165] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0166] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or, if necessary, processing in other suitable ways, and then stored in computer memory.

[0167] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0168] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0169] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0170] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for pushing recommended content, characterized in that, Includes the following steps: Obtain the target user's first historical operation behavior, and generate first time-series behavior data based on the first historical operation behavior; The first time-series behavioral data is input into a pre-trained user intention prediction model to obtain the first user intention of the target user. Based on the first user's intention, multiple recommended content items and corresponding interest indicators are determined in a pre-built recommendation content library. The interest indicators include the interest of multiple historical users on the recommended content items. The recommendation degree of each of the content to be recommended is determined based on the interest degree identifier and the user similarity between the target user and the historical users. Then, the target recommended content is determined based on the recommendation degree and the target recommended content is pushed to the target user. The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network. The push method further includes the step of training the user intention prediction model, which specifically includes: A user intention prediction model is constructed, which includes an input layer, a forward temporal convolutional network, a backward temporal convolutional network, and a fully connected layer. Obtain the second historical operation behavior of the historical user, and determine the second time-series behavior data and the corresponding user intention tag based on the second historical operation behavior; A training sample set is constructed based on the second time-series behavior data and the corresponding user intention tags; The training sample set is input into the user intention prediction model for training, and the model parameters of the user intention prediction model are optimized to obtain the trained user intention prediction model. The step of inputting the training sample set into the user intention prediction model for training, optimizing the model parameters of the user intention prediction model, and obtaining the trained user intention prediction model specifically includes: The second temporal behavior data is input into the input layer, and the first temporal feature vector is obtained by convolution calculation through the forward temporal convolutional network; The second temporal behavior data is input into the input layer in reverse order, and the second temporal feature vector is obtained by convolution calculation through the inverse temporal convolutional network. The second time-series feature vector is reversed to obtain the third time-series feature vector, and the first time-series feature vector and the third time-series feature vector are concatenated to obtain the fourth time-series feature vector; The fourth temporal feature vector is input into the fully connected layer, and the user intention prediction result is output. The loss value of the user intention prediction model is determined based on the user intention prediction results and the user intention labels. Update the model parameters of the user intention prediction model based on the loss value, and return to the step of inputting the second time-series behavioral data into the input layer; When the loss value reaches a preset first threshold or the number of iterations reaches a preset second threshold, training stops, and the trained user intention prediction model is obtained. The push method further includes the step of constructing the recommended content library, which specifically includes: Obtain the historical browsing content and corresponding browsing duration of the historical users; The historical browsing content with a browsing duration greater than or equal to a preset third threshold is identified as the first recommended content, and the content type and content richness of the first recommended content are determined. The interest level of the historical users in the first recommended content is determined based on the browsing duration and the richness of the content, and the interest level identifier of the first recommended content is determined based on the interest level of each historical user in the first recommended content. The recommendation intention index value of the first recommended content is determined according to the content type, and a first mapping relationship between the recommendation intention index value and the first recommended content is established. The recommended content library is generated based on the first recommended content, the interest index, the recommendation intention index value, and the first mapping relationship. The degree of interest is calculated using the following formula: Where r represents interest level, T represents browsing time, P represents content richness, and μ represents a preset coefficient obtained through experimental calibration.

2. The method for pushing recommended content according to claim 1, characterized in that, The step of obtaining the target user's first historical operation behavior and generating first time-series behavior data based on the first historical operation behavior specifically includes: Obtain the target user's historical operation records, and determine the first historical operation behavior and the corresponding operation time based on the historical operation records; Based on the operation time, the first historical operation behavior is cleaned and sliced ​​to obtain user behavior data. The user behavior data is sorted according to the operation time to obtain the first time-series behavior data.

3. The method for pushing recommended content according to claim 1, characterized in that, The step of determining multiple recommended content items and their corresponding interest indicators from a pre-built recommendation content library based on the first user's intention specifically includes: Based on the first user's intention, the corresponding recommendation intention index value is obtained by traversing and matching the recommended content library. Based on the matched recommendation intention index value and the first mapping relationship, multiple first recommended contents are indexed in the recommendation content library as the content to be recommended, and the interest degree identifier of each content to be recommended is determined.

4. A method for pushing recommended content according to any one of claims 1 to 3, characterized in that, The step of determining the recommendation level of each of the content to be recommended based on the interest identifier and the user similarity between the target user and the historical users, and then determining the target recommended content based on the recommendation level, specifically includes: Obtain a first user profile of the target user and a second user profile of the historical user, and determine a first user feature vector based on the first user profile and a second user feature vector based on the second user profile; The user similarity between the target user and the historical user is determined based on the cosine similarity between the first user feature vector and the second user feature vector; Based on the interest index, the interest of each historical user on the content to be recommended is determined, and the user similarity is used as a weight to perform a weighted sum of the interest of each historical user to obtain a first weighted sum. The user similarity scores corresponding to each of the historical users are summed to obtain a first sum value, and the recommendation score of the content to be recommended is obtained by dividing the first weighted sum by the first sum value. The content to be recommended that has a recommendation score greater than or equal to a preset fourth threshold is determined as the target recommended content; or, the content to be recommended that has a recommendation score less than or equal to a preset fifth threshold is determined as the target recommended content.

5. A content recommendation system, characterized in that, A method for pushing recommended content as described in any one of claims 1 to 4, comprising: The time-series behavior data generation module is used to obtain the first historical operation behavior of the target user and generate first time-series behavior data based on the first historical operation behavior. The user intention prediction module is used to input the first time-series behavioral data into a pre-trained user intention prediction model to obtain the first user intention of the target user. The module for determining recommended content is used to determine multiple recommended contents and corresponding interest indicators in a pre-built recommended content library based on the first user's intention. The interest indicators include the interest of multiple historical users on the recommended contents. The recommendation degree determination module is used to determine the recommendation degree of each of the content to be recommended based on the interest degree identifier and the user similarity between the target user and the historical users, and then determine the target recommended content based on the recommendation degree, and push the target recommended content to the target user; The user intention prediction model is obtained by jointly training a forward temporal convolutional network and a backward temporal convolutional network.

6. An electronic device, characterized in that: The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for enabling communication between the processor and the memory. When the program is executed by the processor, it implements the steps of the method for pushing recommended content as described in any one of claims 1 to 4.

7. A storage medium, said storage medium being a computer-readable storage medium for computer-readable storage, characterized in that: The storage medium stores one or more programs, which can be executed by one or more processors to implement the steps of the method for pushing recommended content as described in any one of claims 1 to 4.

Images