Content recommendation method and information generation model training method

Abstract

The embodiment of the invention provides a content recommendation method and an information generation model training method. The content recommendation method comprises the steps of obtaining attribute information of a target object; inputting the attribute information into an information generation model to obtain initial recommendation information matched with the target recommendation scene; the initial recommendation information is utilized to adjust candidate recommendation content, target recommendation content is obtained, and the candidate recommendation content is the content which is obtained through screening based on the initial recommendation information and matched with the initial recommendation information; and in the target recommendation scene, displaying the target recommendation content to the target object. The candidate recommendation content is adjusted by using the initial recommendation information, so that the target recommendation content better fits the hobbies and interests of the target object, and the content recommendation accuracy is improved. The initial recommendation information is matched with the target recommendation scene, so that the target recommendation content achieves a very good effect on visual layout, and the satisfaction degree of the target object is enhanced.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of Internet technologies, and particularly to a content recommendation method and an information generation model training method. Background Art

[0002] With the booming development of the Internet and the wide application of intelligent devices, users' demand for personalized information has increased sharply. Against this background, a content recommendation system based on users' interests and hobbies has emerged, aiming to provide users with information services that better meet their individual needs.

[0003] Currently, a fixed content recommendation mechanism is usually adopted to recommend content to users. However, the fixed content may not match or have a low matching degree with users' interests and hobbies, lacking attraction to users, resulting in poor accuracy of content recommendation. Moreover, the fixed content recommendation mechanism focuses on the content itself and ignores the content layout scenario, resulting in that even if the content itself is of high quality, the recommendation effect may be affected due to inappropriate presentation methods. Summary of the Invention

[0004] In view of this, the embodiments of this specification provide a content recommendation method. One or more embodiments of this specification simultaneously relate to an information generation model training method, an information processing method based on an information generation model, a task platform, a content recommendation device, an information generation model training device, an information processing device based on an information generation model, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0005] According to the first aspect of the embodiments of this specification, a content recommendation method is provided, including:

[0006] Obtain the attribute information of the target object;

[0007] Input the attribute information into an information generation model to obtain initial recommendation information that matches the target recommendation scenario;

[0008] Use the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content, where the candidate recommendation content is the content that is screened based on the initial recommendation information and matches the initial recommendation information;

[0009] In the target recommendation scenario, display the target recommendation content to the target object.

[0010] According to the second aspect of the embodiments of this specification, an information generation model training method is provided, including:

[0011] Obtain object feedback information, where the object feedback information is information for which the target object provides feedback on the target recommended content. The target recommended content is obtained by adjusting candidate recommended content based on initial recommended information that matches the target recommendation scenario. The initial recommended information is obtained by an information generation model based on the attribute information of the target object, and the candidate recommended content is content that matches the initial recommended information and is filtered based on the initial recommended information.

[0012] Adjust the parameters of the information generation model according to the object feedback information to obtain an adjusted information generation model.

[0013] According to the third aspect of the embodiments of the present specification, an information processing method based on an information generation model is provided, which is applied to a task platform and includes:

[0014] Receive a model request sent by a terminal device;

[0015] Based on the model request, determine a target information generation model from multiple information generation models, where the target information generation model is used in the execution process of the content recommendation method.

[0016] According to the fourth aspect of the embodiments of the present specification, a task platform is provided, including a request interface and a response unit;

[0017] The request interface is used to receive a model request sent by a terminal device, where the model request includes at least one of a scenario identifier of the target recommendation scenario, scenario input data of the target recommendation scenario, and model specification parameters;

[0018] The response unit is used to determine a target information generation model from multiple information generation models based on the model request, where the target information generation model is used in the execution process of the content recommendation method.

[0019] According to the fifth aspect of the embodiments of the present specification, a content recommendation device is provided, including:

[0020] The first acquisition module is configured to acquire the attribute information of the target object;

[0021] The input module is configured to input the attribute information into the information generation model to obtain initial recommended information that matches the target recommendation scenario;

[0022] The first adjustment module is configured to use the initial recommended information to adjust the candidate recommended content to obtain the target recommended content, where the candidate recommended content is content that matches the initial recommended information and is filtered based on the initial recommended information;

[0023] The display module is configured to display the target recommended content to the target object in the target recommendation scenario.

[0024] According to a sixth aspect of the embodiments of the present specification, there is provided an information generation model training method and apparatus, including:

[0025] A second acquisition module, configured to acquire object feedback information, where the object feedback information is information fed back by a target object for target recommended content, and the target recommended content is obtained by adjusting candidate recommended content based on initial recommended information that matches a target recommendation scenario. The initial recommended information is obtained by an information generation model based on the attribute information of the target object, and the candidate recommended content is content that matches the initial recommended information and is screened based on the initial recommended information;

[0026] A second adjustment module, configured to adjust the parameters of the information generation model according to the object feedback information to obtain an adjusted information generation model.

[0027] According to a seventh aspect of the embodiments of the present specification, there is provided an information processing apparatus based on an information generation model, which is applied to a task platform and includes:

[0028] A receiving module, configured to receive a model request sent by a terminal device;

[0029] A determination module, configured to determine a target information generation model from multiple information generation models based on the model request, where the target information generation model is used in the execution process of a content recommendation method.

[0030] According to an eighth aspect of the embodiments of the present specification, there is provided a computing device, including:

[0031] A memory and a processor;

[0032] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the methods provided in the above first aspect or second aspect or third aspect are implemented.

[0033] According to a ninth aspect of the embodiments of the present specification, there is provided a computer-readable storage medium, which stores computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the methods provided in the above first aspect or second aspect or third aspect are implemented.

[0034] According to a tenth aspect of the embodiments of the present specification, there is provided a computer program product, including computer programs / instructions. When the computer programs / instructions are executed by a processor, the steps of the methods provided in the above first aspect or second aspect or third aspect are implemented.

[0035] The content recommendation method provided by an embodiment of this specification includes: obtaining the attribute information of the target object; inputting the attribute information into the information generation model to obtain the initial recommendation information that matches the target recommendation scenario; using the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content, where the candidate recommendation content is the content that is screened based on the initial recommendation information and matches the initial recommendation information; and presenting the target recommendation content to the target object in the target recommendation scenario. By using the information generation model to deeply analyze the attribute information of the target object, the user's interests and hobbies can be understood more accurately, so as to provide highly personalized initial recommendation information based on the characteristics of the target object. Using the initial recommendation information to adjust the candidate recommendation content makes the finally recommended target recommendation content more in line with the interests and hobbies of the target object, realizes the dynamic adaptation between the interests and hobbies and the content, significantly improves the attractiveness of the target recommendation content to the target object, and further improves the accuracy of content recommendation. Moreover, since the initial recommendation information matches the target recommendation scenario, it is ensured that the target recommendation content finally presented to the target object not only adapts to the interests and hobbies of the target object in terms of the content itself, but also achieves very good results in terms of visual layout and interaction design, enhancing the consistency of content recommendation and the satisfaction of the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flowchart of a content recommendation method provided by an embodiment of this specification;

[0037] Figure 2 is an architecture diagram of a content recommendation system provided by an embodiment of this specification;

[0038] Figure 3 is a process timing diagram of a content recommendation method provided by an embodiment of this specification;

[0039] Figure 4 is a flowchart of a note recommendation method provided by an embodiment of this specification;

[0040] Figure 5 is an interface schematic diagram of a note recommendation interface provided by an embodiment of this specification;

[0041] Figure 6 is a flowchart of an information generation model training method provided by an embodiment of this specification;

[0042] Figure 7 is a flowchart of an information processing method based on an information generation model provided by an embodiment of this specification;

[0043] Figure 8 is a structural schematic diagram of a task platform provided by an embodiment of this specification;

[0044] Figure 9 It is a schematic structural diagram of a content recommendation device provided by an embodiment of this specification;

[0045] Figure 10 It is a schematic structural diagram of an information generation model training device provided by an embodiment of this specification;

[0046] Figure 11 It is a schematic structural diagram of an information processing device based on an information generation model provided by an embodiment of this specification;

[0047] Figure 12 It is a structural block diagram of a computing device provided by an embodiment of this specification. Specific embodiments

[0048] Many specific details are set forth in the following description in order to provide a thorough understanding of this specification. However, this specification can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the connotation of this specification. Therefore, this specification is not limited by the specific embodiments disclosed below.

[0049] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a", "the" and "said" used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more of the associated listed items.

[0050] It should be understood that although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0051] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0052] In one or more embodiments of this specification, a large model refers to a deep learning model with a large number of model parameters, usually including hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than one quadrillion model parameters. A large model can also be called a foundation model. Through pre-training of the large model with a large amount of unlabeled corpus, a pre-trained model with more than one hundred million parameters is produced. This kind of model can adapt to a wide range of downstream tasks and has good generalization ability. For example, large language models (LLMs), multi-modal pre-training models, etc.

[0053] When a large model is actually applied, it only needs to fine-tune the pre-trained model with a small number of samples to be applied to different tasks. Large models can be widely applied in the fields of natural language processing (NLP), computer vision, etc. Specifically, they can be applied to tasks in the field of computer vision such as visual question answering (VQA), image captioning (IC), image generation, etc., and tasks in the field of natural language processing such as text-based sentiment classification, text summary generation, machine translation, etc. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0054] First, the noun terms involved in one or more embodiments of this specification are explained.

[0055] The deep self-attention (Transformer) model: is a network structure based on the multi-head self-attention mechanism module, mainly used to process sequence data. The Transformer model includes a stackable encoding unit (Encoder) and decoding unit (Decoder). This design allows the Transformer to efficiently learn long-term dependencies and is suitable for various natural language processing tasks including machine translation, text summarization, question answering systems, etc.

[0056] Bidirectional Encoder Representations from Transformers (BERT) model: It is a natural language processing (NLP) pre-trained model. By learning a large amount of unlabeled text data, this model can capture deep semantic information in the text and has achieved significant performance improvements in numerous NLP tasks.

[0057] Generative Pre-trained Transformer (GPT): It is a deep learning model architecture mainly applied in the NLP field. Based on the Transformer architecture, GPT relies on the attention mechanism to capture long-range dependencies in the input data.

[0058] Diffusion Models: They are generative models that transform complex data distributions into simple noise distributions by gradually adding noise to the data and then learn an inverse process to reconstruct the original data from the noise. This inverse process can be regarded as a "denoising" process, gradually reducing the noise through a series of steps and finally generating samples similar to those in the training set. The training objective of diffusion models is to optimize this denoising process to enable the generation of high-quality samples.

[0059] Residual Networks (ResNet): It is a deep convolutional neural network architecture. ResNet introduces "skip connections" or "shortcut connections" that allow information to pass directly across one or more layers, thus alleviating the degradation problem during the training of deep networks. Specifically, these skip connections enable each layer to learn the residual mapping between the input and output rather than the direct mapping, which helps with optimization and accelerates convergence.

[0060] Recurrent Neural Network (RNN): It is a neural network architecture specifically designed for processing sequential data. RNNs have recurrent connections that allow information to be passed between time steps. This feature enables RNNs to capture temporal dependencies in the input data and is very suitable for processing types of sequential data such as natural language, speech signals, and time series.

[0061] Generative Adversarial Networks (GANs): It is a deep learning model architecture that realizes data generation through adversarial training between a generator and a discriminator.

[0062] Multimodal Large Model: It refers to a large-scale machine learning model that can process and integrate various modal data (such as text, images, audio, etc.). Through shared representation learning or cross-modal fusion techniques, these models achieve the comprehensive understanding and generation capabilities for different data types.

[0063] Text Information Generation Model: It is a model that can generate natural language text based on given inputs. Text information generation models include, but are not limited to, generative pre-trained transformers, recurrent neural networks, and deep self-attention models.

[0064] Image Information Generation Model: It is a model that can generate visual content based on given inputs. Image information generation models include, but are not limited to, generative adversarial networks and diffusion models.

[0065] Low-Rank Adaptation (LoRA): It is a technique for efficiently fine-tuning large pre-trained models. Traditionally, when fine-tuning a large pre-trained model, all parameters are updated, which requires a large amount of computational resources and time. LoRA proposes to update only part of the parameters, specifically those that can be represented by low-rank decomposition. Specifically, for each fully connected layer or convolutional layer, LoRA adds a small-scale low-rank matrix as learnable parameters while keeping the original large matrix unchanged. This method greatly reduces the number of parameters to be trained, lowers the cost of fine-tuning, and at the same time can maintain good performance.

[0066] Supervised Fine-Tuning (SFT): It is a further training method based on pre-trained models. In this method, the model is trained on a dataset containing input and expected output pairs so that the model can learn how to generate answers closer to the human level. Supervised fine-tuning is usually used to make the model adapt to data in specific tasks or domains, thereby improving the model's performance on these tasks.

[0067] Direct Preference Optimization (DPO): It is a model optimization method based on user preferences, used to improve the quality and compliance of the model's generated results.

[0068] Multimodal Deep Learning Model (CLIP, Contrastive Language–Image Pretraining): It is a deep learning model that can map images and texts into a common embedding space. The CLIP model consists of two main parts: an image encoder, which is responsible for converting the input image into a fixed-length vector representation; and a text encoder, which is responsible for converting the input text into a fixed-length vector representation. These two encoders encode images and texts respectively and map them into a shared multi-dimensional embedding space, where the correlation between texts and images can be calculated.

[0069] Offline Metrics: Refers to a series of quantitative criteria for evaluating a content recommendation system through historical data or a simulation environment without changing the existing system. Offline metrics include, but are not limited to, Precision, Recall, F1 Score, and Diversity.

[0070] A / B Testing: It is an online experimental method where users are randomly assigned to different experimental groups (such as Group A and Group B), different recommendation strategies are shown respectively, and the actual behavior data of users in each group are compared to evaluate which strategy is more effective. A / B testing can directly reflect the reactions of real users and provide real evaluation results.

[0071] Controls: refer to any elements in a computer user interface with which users can directly interact. These elements allow users to input data, select options, or trigger certain actions. Controls can be graphical (such as buttons, text boxes, checkboxes, etc.) or in more abstract forms (such as voice commands). Buttons are usually used to perform an operation or open a new interface; Text Boxes allow users to input or edit text information; Checkboxes are used to represent a boolean value (yes / no), and users can select or deselect them; Dropdown Lists provide a list of options for users to choose from, usually only showing one selected value; List Boxes display multiple options, and users can select one or more items; Labels display static text, usually used to explain the functions of other controls and can also be used as a button; Scroll Bars enable users to navigate through a large dataset, such as browsing a long document or list; Images display static pictures and can sometimes also be used as buttons; Comboboxes combine the functions of text boxes and dropdown lists, and users can either input text or select from the list. These controls usually have standard styles provided by the operating system or development toolkits, and developers can customize their appearance and behavior according to needs. Different operating systems and programming languages (such as Java, C#, Python, etc.) have their own sets of controls and provide corresponding APIs to create and manage these controls.

[0072] Traditional recommendation algorithms (such as collaborative filtering or deep learning models based on behavioral characteristics) rely on users' historical behaviors for preference modeling. However, new users have insufficient behavioral data, resulting in the inability to effectively train the recommendation model and the problem of difficulty in predicting users' interests, leading to low-quality recommendation results. Secondly, the content recommended in traditional recommendation schemes is usually fixed content set by the author, and this content may not match the interests of new users, lacking attraction to new users and resulting in poor accuracy of content recommendation. Moreover, the fixed-content recommendation mechanism focuses on the content itself and ignores the content layout scenario, resulting in the situation that even if the content itself is of high quality, the recommendation effect may be affected due to inappropriate presentation methods.

[0073] To solve the above problems, the embodiments of this specification propose a content recommendation solution that adapts to the recommendation scenario and new users based on a generative information generation model, obtains the attribute information of the target object; inputs the attribute information into the information generation model to obtain initial recommendation information that matches the target recommendation scenario; uses the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content, where the candidate recommendation content is the content that is filtered based on the initial recommendation information and matches the initial recommendation information; in the target recommendation scenario, display the target recommendation content to the target object. By introducing the attribute information of the target object, initial recommendation information that conforms to the interests and hobbies of the target object and the target recommendation scenario is dynamically generated, thereby solving the problem of poor recommendation effects caused by sparse behavior and insufficient information of new users, providing more personalized content recommendations for new users, and at the same time enhancing the visual effect and click-through rate of the content in the recommendation scenario. The initial recommendation information is used to replace the original recommendation information in the candidate recommendation content, so as to better match the interests of new users and improve the click-through rate.

[0074] In this specification, a content recommendation method is provided. This specification also relates to a note recommendation method, an information generation model training method, an information processing method based on an information generation model, a task platform, a content recommendation device, a note recommendation device, an information generation model training device, an information processing device based on an information generation model, a computing device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail one by one in the following embodiments.

[0075] See Figure 1 , Figure 1 FIG. shows a flowchart of a content recommendation method provided by an embodiment of this specification, which specifically includes the following steps:

[0076] Step 102: Obtain the attribute information of the target object.

[0077] It should be noted that the target object refers to the specific object (individual object or group object) targeted by the content recommendation service. The target object can be an individual user or a virtual character. The attribute information of the target object refers to various characteristics and data points related to the target object. The attribute information can be divided into static attribute information and dynamic attribute information. The static attribute information refers to the statistical characteristics of the target object, such as age, gender, occupation, etc. The dynamic attribute information refers to the behavior data of the target object, such as page visits, browsing history, purchase records, click operations, dwell time, historical preference recommendation information, preference recommendation content tags, etc. The portrait of the target object can be depicted through the attribute information of the target object, helping the content recommendation system understand the interests and hobbies and needs of the target object.

[0078] In practical applications, there are various ways to obtain the attribute information of the target object, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In one possible implementation manner of this specification, the attribute information of the target object can be read from other databases or data acquisition devices. For example, the static attribute information of the target object is read from the object information library, the dynamic attribute information of the target object sent by the client is received, and the static attribute information and the dynamic attribute information are fused to form the complete attribute information of the target object. In another possible implementation manner of this specification, the attribute information of the target object sent by the target object through the client can be received.

[0079] Step 104: Input the attribute information into the information generation model to obtain the initial recommendation information that matches the target recommendation scenario.

[0080] It should be noted that the information generation model refers to an algorithm or model that can receive the attribute information of the target object as input and output the initial recommendation information suitable for a specific recommendation scenario. The information generation model can be a machine learning-based model. For example, a deep learning model is used to predict the initial recommendation information that matches the target recommendation scenario. The information generation model can also be a rule-based algorithm. For example, the initial recommendation information that matches the target recommendation scenario is selected from the preset recommendation information through the preset logic. The initial recommendation information refers to the result processed by the information generation model. The initial recommendation information can be multimodal recommendation information, such as text recommendation information, image recommendation information, audio recommendation information, etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. The number of the initial recommendation information is one or more. For example, if the initial recommendation information includes 100 pieces of text recommendation information and 10 pieces of image recommendation information, then the initial recommendation information can be 100 text-image pairs. In the generation process of the initial recommendation information, the attribute information of the target object and the scenario requirements of the target recommendation scenario are considered. Therefore, the initial recommendation information is not only highly personalized information for the target object, but also information that can show very good effects in the target recommendation scenario. For example, the attribute information of the target object indicates that the target object lacks patience and is interested in the important data and conclusions in the report, and the target recommendation scenario includes a text box with a character capacity limit of 40, then the initial recommendation information can be text information with no more than 40 characters, such as "Climate change, key data at a glance".

[0081] In practical applications, in order to ensure that the initial recommendation information generated by the information generation model is information that matches the target recommendation scenario, in a possible implementation manner of this specification, the attribute information can be input into a pre-trained information generation model to obtain the initial recommendation information that matches the target recommendation scenario, where the pre-trained information generation model is trained based on multiple training recommendation information that matches the target recommendation scenario and the training attribute information respectively corresponding to the multiple training recommendation information. By training the information generation model, when the information generation model faces new attribute information, it also has the ability to generate recommendation information that matches the target recommendation scenario. In another possible implementation manner of this specification, instead of training the information generation model to have the ability to generate recommendation information that matches the target recommendation scenario, the scenario constraint information of the target recommendation scenario can be added to the model input of the information generation model, so that the information generation model can understand the information generation background, and thus generate the initial recommendation information that matches the target recommendation scenario.

[0082] In an optional embodiment of this specification, the initial recommendation information includes at least one of text recommendation information and image recommendation information; the above-mentioned inputting the attribute information into the information generation model to obtain the initial recommendation information that matches the target recommendation scenario may include the following steps:

[0083] Input the attribute information and the scenario constraint information of the target recommendation scenario into the information generation model to obtain text recommendation information; and / or,

[0084] Input the attribute information and the scenario constraint information into the information generation model to obtain image recommendation information.

[0085] It should be noted that the scenario constraint information refers to the information that describes the requirements or restrictive conditions of the target recommendation scenario. The scenario constraint information can be divided into text constraint information and image constraint information. The text constraint information includes, for example, word count constraints (such as 20 - 40 words), font constraints (such as Song typeface), font size constraints (such as small font size 4), etc. The image constraint information includes, for example, aspect ratio constraints (such as a landscape image with an aspect ratio of 16:9 or a portrait image with an aspect ratio of 9:16), size constraints (such as not exceeding 5MB), color tone constraints (such as warm color tone), style constraints (such as retro style), etc. Through the scenario constraint information, it can be ensured that the initial recommendation information not only conforms to the interests and hobbies of the target object, but also adapts to the current target recommendation scenario. The text recommendation information refers to the personalized text content obtained by processing through the information generation model, such as product descriptions, social posts, comment summaries, advertising copywriting, etc. The image recommendation information refers to the personalized image content obtained after processing by the information generation model, such as product pictures, social media photos, artworks, etc.

[0086] In practical applications, the information generation models for generating text recommendation information and image recommendation information can be the same or different. For example, the same multi-modal large model can be used to generate text recommendation information and image recommendation information, or a text information generation model can be used to generate text recommendation information, and an image generation information model can be used to generate image recommendation information.

[0087] Applying the solution of the embodiments of this specification, scene constraint information of the target recommendation scene is added to the model input of the information generation model, so that there is no need to train the information generation model to generate the ability to generate recommendation information matching the target recommendation scene, and the generation efficiency of the initial recommendation information is improved.

[0088] Step 106: Use the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content, where the candidate recommendation content is the content that is screened based on the initial recommendation information and matches the initial recommendation information.

[0089] It should be noted that the candidate recommendation content refers to a series of recommendable content in the content recommendation process. The candidate recommendation content is screened from a wide range of data sources based on the initial recommendation information, but has not been personalized adjusted according to a specific user or scene. The candidate recommendation content includes notes, articles, blogs, product usage instructions, etc. Taking the candidate recommendation content as a note as an example, the candidate recommendation content can be the notes in the note pool that highly match the initial recommendation information. Adjustment refers to the process of further processing the candidate recommendation content based on the initial recommendation information. Through adjustment, it can be ensured that the target recommendation content finally presented to the target object not only meets the interests and hobbies of the target object, but also achieves very good effects in terms of visual layout, interaction design, etc. The adjustment process may involve steps such as reordering, adding new content items, removing irrelevant items, or modifying the display form of the content. The target recommendation content refers to the recommendation content that is finally to be shown to the target object, such as a note with a title of 20 characters and a cover image style of a retro style.

[0090] In practical applications, there are various ways to adjust candidate recommended content using initial recommended information to obtain target recommended content, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In one possible implementation manner of this specification, the initial recommended information can be directly used to replace the original recommended information in the candidate recommended content to obtain the target recommended content. For example, if the initial recommended information is the title "Quick Overview of Key Data on Climate Change", and the original recommended information in the candidate recommended content is the title "Research Report on Global Climate Change", directly using the initial recommended information to replace the original recommended information in the candidate recommended content, the obtained target recommended content is the candidate recommended content with the title modified to "Quick Overview of Key Data on Climate Change". In another possible implementation manner of this specification, the initial recommended information and the original recommended information in the candidate recommended content can be fused to obtain fused recommended information, and the fused recommended information is used to replace the original recommended information in the candidate recommended content to obtain the target recommended content. For example, if the initial recommended information is the title "Quick Overview of Key Data on Climate Change", and the original recommended information in the candidate recommended content is the title "Research Report on Global Climate Change", the fused recommended information obtained by fusing the initial recommended information and the original recommended information is the title "Report on Quick Overview of Key Data on Global Climate Change", and using the fused recommended information to replace the original recommended information in the candidate recommended content, the obtained target recommended content is the candidate recommended content with the title modified to "Report on Quick Overview of Key Data on Global Climate Change".

[0091] In an optional embodiment of this specification, before adjusting the candidate recommended content using the initial recommended information to obtain the target recommended content, the following steps may further be included:

[0092] From multiple original contents, screen out candidate recommended contents that match the initial recommended information.

[0093] It should be noted that the original content refers to the data included in a wide range of data sources. The original content may or may not match the initial recommended information. Therefore, candidate recommended contents that match the initial recommended information can be screened out from multiple original contents. When screening candidate recommended contents based on the initial recommended information from multiple original contents, it can be screened from all the original contents, or from some of the original contents that meet the recommended conditions among all the original contents. Among them, the recommended conditions are such as within a preset time (such as within one year), the popularity exceeds a preset popularity threshold (such as the number of likes exceeds 1000), etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard.

[0094] Exemplarily, assume that the initial recommended information is the title "Climate Change, Key Data at a Glance". The first original content is "Title: Analysis of the Global Temperature Change Trend in 2024. Body: This report provides the change data of the global temperature in the past year, including the temperature anomalies of each continent and ocean. Through detailed charts and statistical data, it reveals the impact of climate change on the Earth's environment." The second original content is "Title: The Latest Fashion Trends, A Review of the 2023 Autumn / Winter Fashion Shows. Body: This article reviews the highlights of the 2023 autumn / winter fashion weeks, introducing the new designs and trends of major brands. From haute couture to street style, it takes you to explore the latest fashion vane." It can be determined that the first original content is the content that matches the initial recommended information, that is, the first original content is the candidate recommended content. The second original content is the content that does not match the initial recommended information.

[0095] In practical applications, there are multiple ways to screen out the candidate recommended content that matches the initial recommended information from multiple original contents, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In one possible implementation manner of this specification, a deep learning model can be used to calculate the similarity between the initial recommended information and the original content, and the content that matches the initial recommended information can be screened out according to the similarity. For example, if the initial recommended information is text recommended information, a semantic embedding model (such as BERT) can be used to calculate the text similarity between the text recommended information and the original content; if the initial recommended information is image recommended information, an image feature extraction model (such as ResNet) can be used to calculate the image similarity between the image recommended information and the original content. In another possible implementation manner of this specification, similarity algorithms such as cosine similarity and Euclidean distance can be used to calculate the similarity between the initial recommended information and the original content, and the content that matches the initial recommended information can be screened out according to the similarity.

[0096] Furthermore, there are multiple ways to screen out the content that matches the initial recommended information according to the similarity, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In one possible implementation manner of this specification, the original content with a similarity greater than a preset similarity threshold to the initial recommended information can be determined as the candidate recommended content. If the initial recommended information includes text recommended information and image recommended information, the original content with both text similarity and image similarity greater than the preset similarity threshold to the initial recommended information can be determined as the candidate recommended content. Among them, the preset similarity threshold is specifically set according to the actual situation. In another possible implementation manner of this specification, the first M (M is a positive integer, such as 100) original contents with a relatively high similarity to the initial recommended information can be determined as the candidate recommended content.

[0097] It should be noted that if there is no content in the original content that matches the current initial recommendation information, the current initial recommendation information is skipped, and the content that matches the next initial recommendation information is selected from the original content as the candidate recommendation content.

[0098] In an optional embodiment of this specification, the candidate recommendation content includes the original recommendation information; the above-mentioned use of the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content may include the following steps:

[0099] Use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content.

[0100] It should be noted that the original recommendation information refers to the information of the same type as the initial recommendation information in the candidate recommendation content. For example, if the initial recommendation information is the title, the original recommendation information is the original title of the candidate recommendation content. Another example is that if the initial recommendation information is the cover image, the original recommendation information is the original cover image of the candidate recommendation content. The candidate recommendation content includes not only the original recommendation information but also other content information. Taking the candidate recommendation content as the candidate recommendation note as an example, the candidate recommendation note includes not only the note title and note cover, but also the note text, background music, and so on.

[0101] In practical applications, there are various ways to use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content, which is specifically selected according to the actual situation. The embodiments of this specification do not make any limitations on this. In a possible implementation manner of this specification, the initial recommendation information can be used to replace the original recommendation information in the candidate recommendation content, and the replaced recommendation content obtained after replacement is directly determined as the target recommendation content. In another possible implementation manner of this specification, the replaced recommendation content can be obtained by replacement, and the target replacement recommendation content that highly conforms to the attribute information of the target object is screened out from the replaced recommendation content, and these target replacement contents are determined as the target recommendation content.

[0102] Applying the solution of the embodiments of this specification, by replacing the original recommendation information in the candidate recommendation content with the initial recommendation information, the target recommendation content is made more in line with the interests and hobbies of the target object, realizing the dynamic adaptation between the interests and hobbies and the content, significantly enhancing the attraction of the target recommendation content to the target object, and further improving the accuracy of content recommendation.

[0103] In an optional embodiment of this specification, the above-mentioned use of the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content may include the following steps:

[0104] Use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the replaced recommendation content;

[0105] According to the initial recommendation information, sort the replacement recommendation content, and based on the sorting result, screen out the target recommendation content from the replacement recommendation content.

[0106] It should be noted that the replacement recommendation content refers to the new recommendation content obtained by replacing the original recommendation information with the initial recommendation information. Compared with the candidate recommendation content, the replacement recommendation content is more in line with the interests and hobbies of the target object and is more adaptable to the target recommendation scenario. Sorting refers to the process of sorting the replacement recommendation content according to the initial recommendation information to determine the priority of the replacement recommendation content. The sorting result is used to reflect the matching degree between the replacement recommendation content and the initial recommendation information.

[0107] In practical applications, there are various ways to sort the replacement recommendation content according to the initial recommendation information, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In a possible implementation manner of this specification, the replacement recommendation content can be sorted only according to the initial recommendation information to obtain a sorting result, and this process can be regarded as a rough sorting process. For example, the initial recommendation information can be used as a feature to calculate the matching index between each replacement recommendation content and the initial recommendation information, and the sorting result of the replacement recommendation content can be determined according to the matching index. Among them, the matching index is used to describe the matching degree between the replacement recommendation content and the initial recommendation information. The matching index can be a specific matching value, such as 0.8, or a matching level, such as very matching, relatively matching, and the level of very matching is higher than that of relatively matching. When sorting the replacement recommendation content, the replacement content can be sorted from large to small according to the matching value or from high to low according to the matching level. Of course, it can also be sorted from small to large according to the matching value or from low to high according to the matching level to obtain a sorting result. The sorting method is specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In another possible implementation manner of this specification, on the basis of the above-mentioned rough sorting process, other evaluation information can be introduced for fine sorting. Other evaluation information such as content popularity, content timeliness, etc. For example, assume that there are ten replacement recommendation contents. The ten replacement recommendation contents are roughly sorted according to the initial recommendation information, and the first eight replacement recommendation contents with higher matching degrees are retained. Then, the first eight replacement recommendation contents are finely sorted according to the content popularity of the first eight replacement recommendation contents to obtain the sorting result of the first eight replacement recommendation contents.

[0108] Further, when screening out target recommended content from the replacement recommended content according to the sorting result, the replacement recommended content with a high degree of matching with the initial recommended information is screened out as the target recommended content for screening. For example, if the replacement content is sorted from largest to smallest according to the matching value or from highest to lowest according to the matching level, the top N replacement recommended content can be selected as the target recommended content; if the replacement content is sorted from smallest to largest according to the matching value or from lowest to highest according to the matching level, the last N replacement recommended content can be selected as the target recommended content, where N is a positive integer, such as 10, and is specifically configured according to the actual situation.

[0109] Applying the solution of the embodiment of this specification, since the sorting result of the replacement recommended content is obtained based on the initial recommended information, and the initial recommended information is generated based on the attribute information of the target object, the sorting result can reflect the degree of fit between the replacement recommended content and the interests and hobbies of the target object, avoiding the neglect of the preferences of new users by the traditional recommendation logic and enhancing the personalization of the recommendation. Further screening out the target recommended content according to the sorting result of the replacement recommended content makes the target recommended content more in line with the interests and hobbies of the target object and improves the accuracy of the target recommended content.

[0110] Step 108: In the target recommendation scenario, display the target recommended content to the target object.

[0111] It should be noted that the target recommendation scenario refers to the specific environment or context for content recommendation. Such as news applications for recommending text, e-commerce platforms for recommending products, social platforms for recommending friends, social platforms for recommending notes, and so on. The recommendation strategies or display methods corresponding to different target recommendation scenarios may be different. Displaying refers to the process of displaying the target recommended content to the target object in a predetermined manner. In the target recommendation scenario, there are various ways to display the target recommended content to the target object, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In the first possible implementation manner of this specification, all the target recommended content can be directly displayed in the target recommendation scenario. In the second possible implementation manner of this specification, the target recommended content can be displayed to the target object in a carousel manner, such as automatically or manually switched by the target object to display different target recommended content in the target recommendation scenario. In the third possible implementation manner of this specification, only part of the target recommended content, such as the title and cover, can be displayed to the target object, so that the target object can preview the target recommended content.

[0112] By applying the solution of the embodiments of this specification and deeply analyzing the attribute information of the target object using the information generation model, the user's interests and hobbies can be understood more accurately, and thus highly personalized initial recommendation information can be provided based on the characteristics of the target object. Adjusting the candidate recommended content using the initial recommendation information makes the finally recommended target recommended content more in line with the interests and hobbies of the target object, can bridge the difference between the distribution of new users and core users, realizes the dynamic adaptation between interests and hobbies and content, significantly improves the attractiveness of the target recommended content to the target object, and further improves the accuracy of content recommendation. Moreover, since the initial recommendation information matches the target recommendation scenario, it ensures that the target recommended content finally presented to the target object not only adapts to the interests and hobbies of the target object in terms of the content itself, but also achieves very good results in terms of visual layout and interaction design, enhancing the consistency of content recommendation and the satisfaction of the target object.

[0113] In an optional embodiment of this specification, in the above target recommendation scenario, presenting the target recommended content to the target object may include the following steps:

[0114] In the target recommendation scenario, typesetting the target recommended content based on the scenario typesetting information of the target recommendation scenario, and presenting the typeset target recommended content to the target object.

[0115] It should be noted that the scene layout information is used to describe the visual and layout requirements of the target recommended content in the target recommended scene. The scene layout information includes, but is not limited to, page structure (such as single-column display structure, double-column display structure), content display mode (such as bubble box display, tiled display, card-style display), screen size, resolution, etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. The single-column display structure means arranging the target recommended content in a vertical or horizontal single column in sequence. The single-column display structure is suitable for occasions such as mobile devices and readers that require a simple browsing experience. It has the advantages of being simple and intuitive, easy to scroll through, and suitable for small-screen devices. The double-column display structure means dividing the target recommended content into two parallel columns, and each column can be in card style or list style. The double-column display structure can provide a richer visual hierarchy and is suitable for larger-screen devices. Of course, the page structure can also be a multi-column display structure, which will not be elaborated in the embodiments of this specification. The bubble box display means displaying the target recommended content in the form of a small circular or oval window. The tiled display means arranging the target recommended content in a rectangular block form tightly, and each block can contain the cover image and title of the target recommended content. The tiled display is suitable for scenarios such as e-commerce websites and media platforms that need to display rich information, and has the advantages of high information density and being suitable for detailed content display. The card-style display means designing each target recommended content into an independent card, usually including the cover image and title of the target recommended content. The card-style display is suitable for mixed display scenarios of various content types (such as videos, articles, products, etc.) and is more visually appealing.

[0116] By applying the solution of the embodiments of this specification, the target recommended content is typeset and displayed by using the scene layout information, improving the display effect of the target recommended content in the target recommended scene and enhancing the satisfaction of the target object.

[0117] In an optional embodiment of this specification, due to the limitations caused by problems such as insufficient quality and diversity of training data in the information generation model itself, the quality of the initial recommended information generated by the information generation model may also be uneven. Therefore, the quality of the initial recommended information can be detected, and the initial recommended information with poor quality can be eliminated. That is, before using the initial recommended information to adjust the candidate recommended content to obtain the target recommended content, the following steps may also be included:

[0118] Detect the quality of the initial recommended information to obtain a quality detection result;

[0119] According to the quality detection result, screen out the target initial recommended information from the initial recommended information;

[0120] Using the initial recommended information to adjust the candidate recommended content to obtain the target recommended content may include the following steps:

[0121] Adjust the candidate recommended content by using the target initial recommended information to obtain the target recommended content.

[0122] It should be noted that quality detection refers to the process of conducting multi-dimensional quality assessment and detection on the initial recommended information. Through quality detection, potential problems in the initial recommended information can be identified and the overall quality of the initial recommended information can be measured. The quality detection result refers to a series of evaluation indicators and conclusions obtained after conducting quality detection on the initial recommended information. The quality detection result can be divided into two types: passing the detection and failing the detection. The quality detection result includes, but is not limited to, accuracy detection results, rationality detection results, credibility detection results, security detection results, etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. The accuracy detection result is used to reflect whether the initial recommended information conforms to the interests and hobbies of the target object, the rationality detection result is used to reflect whether the initial recommended information is reasonable, the credibility detection result is used to reflect the authority of the source of the initial recommended information and the authenticity of the information, and the security detection result is used to reflect whether the initial recommended information conforms to the requirements of laws, regulations and social responsibilities.

[0123] In practical applications, there are various ways to conduct quality detection on the initial recommended information to obtain the quality detection result, which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations in this regard. In one possible implementation manner of this specification, technologies such as NLP and image recognition can be used to conduct quality detection on the initial recommended information to obtain the quality detection result. In another possible implementation manner of this specification, pre-constructed low-quality recommended information can be obtained, and the initial recommended information can be matched with the low-quality recommended information to generate the quality detection result of the initial recommended information. Further, when screening out the target initial recommended information from the initial recommended information according to the quality detection result, all the initial recommended information indicating passing the detection in the quality detection result can be determined as the target initial recommended information, or a preset number of initial recommended information can be randomly selected from the initial recommended information indicating passing the detection in the quality detection result as the target initial recommended information. The implementation manner of "adjust the candidate recommended content by using the target initial recommended information to obtain the target recommended content" can refer to the implementation manner of "adjust the candidate recommended content by using the initial recommended information to obtain the target recommended content" above, and the embodiments of this specification will not elaborate further.

[0124] Exemplarily, assume there are two initial recommendation messages. The first initial recommendation message is "Quick Overview of Key Data on Climate Change", and the second initial recommendation message is "Quick Overview of Key Data on Climate Shallowing". Quality inspection is performed on the first initial recommendation message, and it is determined that the quality inspection result passes. Since the second initial recommendation message contains the typo "shallowing", quality inspection is performed on the second initial recommendation message, and it is determined that the quality inspection result fails. Thus, the first initial recommendation message is determined as the target initial recommendation message.

[0125] It should be noted that if the initial recommendation message is a text-image pair that includes both a text recommendation message and an image recommendation message, when performing quality inspection on the initial recommendation message, the relevance between the text recommendation message and the image recommendation message in a text-image pair can also be detected (such as using a CLIP model for detection) to obtain the quality inspection result.

[0126] Applying the solution of the embodiments of this specification, using the quality inspection result to represent the target initial recommendation message that passes the inspection to adjust the candidate recommendation content ensures the quality of the target recommendation content and improves the accuracy of content recommendation.

[0127] In an optional embodiment of this specification, after presenting the target recommendation content to the target object in the target recommendation scenario, the following steps may further be included:

[0128] Obtain object feedback information, where the object feedback information is the information fed back by the target object regarding the target recommendation content;

[0129] Based on the object feedback information, adjust the parameters of the information generation model to obtain an adjusted information generation model.

[0130] It should be noted that the object feedback information refers to the reactions and evaluations made by the target object regarding the target recommendation content. The feedback of the target object on the target recommendation content can be explicit interaction information (such as scoring, commenting, liking, etc.) or implicit information (such as click-through rate, dwell time, conversion rate, bounce rate, etc.). Through the object feedback information, the recommendation quality and effect of the target recommendation content can be determined. The adjusted information generation model refers to an information generation model with better performance obtained after parameter adjustment. This model can better understand the interests and actual needs of the target object, thereby providing more accurate and personalized content recommendations.

[0131] In practical applications, the methods for adjusting the parameters of the information generation model based on object feedback information include, but are not limited to, methods such as SFT, DPO, LoRA, etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations on this. There are various ways to obtain object feedback information. For example, the object feedback information displayed by the target object can be directly determined through methods such as questionnaires and reviewing comments. It is also possible to determine the implicit object feedback information of the target object by analyzing the behavioral data of the target object for the target recommended content (such as the number of clicks, browsing duration, purchase behavior, etc.).

[0132] Exemplarily, taking the method of adjusting the parameters of the information generation model based on object feedback information as SFT as an example, the process of adjusting the parameters of the information generation model will be described. After obtaining the object feedback information, the specified recommended content selected by the target object can be screened out from the target recommended content according to the object feedback information, and the initial recommended information corresponding to the specified recommended content is determined as the sample recommended information. Then, the attribute information of the target object is determined as the sample attribute information. Next, the sample attribute information is input into the information generation model to obtain the predicted recommended information output by the information generation model. Finally, the model parameters of the information generation model are adjusted according to the predicted recommended information and the sample recommended information to obtain the adjusted information generation model. The sample recommended information is the recommended information that matches the target recommendation scenario. The sample recommended information is the generation target of the information generation model and is used to guide the training process of the information generation model. When adjusting the model parameters of the information generation model according to the predicted recommended information and the sample recommended information, the predicted loss value can be calculated based on the predicted recommended information and the sample recommended information, and the model parameters of the information generation model are adjusted according to the predicted loss value until the training process meets the preset stop condition to obtain the adjusted information generation model. Among them, there are many functions for calculating the predicted loss value, such as the cross-entropy loss function, L1 norm loss function, maximum loss function, mean square error loss function, logarithmic loss function, etc., which are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations on this. The preset stop condition includes, but is not limited to, the predicted loss value being less than or equal to the preset loss threshold, and the number of iterations reaching the preset number of iterations. Among them, the preset loss threshold and the preset number of iterations are specifically selected according to the actual situation, and the embodiments of this specification do not make any limitations on this.

[0133] In a possible implementation of this specification, after calculating the predicted loss value, the predicted loss value is compared with a preset loss threshold. If the predicted loss value is greater than the preset loss threshold, it indicates that the difference between the predicted recommendation information and the sample recommendation information is large, and the ability of the information generation model to generate recommendation information is poor. At this time, the model parameters of the information generation model can be adjusted, and the information generation model can be continuously trained until the predicted loss value is less than or equal to the preset loss threshold, indicating that the difference between the predicted recommendation information and the sample recommendation information is small, reaching the preset stop condition, and obtaining the adjusted information generation model.

[0134] In another possible implementation of this specification, in addition to comparing the magnitude relationship between the predicted loss value and the preset loss threshold, the iteration count can also be combined to determine whether the current information generation model has been adjusted. Specifically, if the predicted loss value is greater than the preset loss threshold, the model parameters of the information generation model are adjusted, and the information generation model is continuously trained until the preset iteration count is reached, at which point the iteration stops, and the adjusted information generation model is obtained.

[0135] Applying the solution of the embodiment of this specification, by continuously adjusting the parameters of the information generation model according to the object feedback information truly provided by the target object, the information generation model can generate initial recommendation information that better suits the interests and hobbies of the target object, improving the accuracy of the information generation model.

[0136] In an optional embodiment of this specification, taking DPO as an example of the method for adjusting the parameters of the information generation model based on the object feedback information, after presenting the target recommendation content to the target object in the target recommendation scenario, the following steps may further be included:

[0137] Obtain object feedback information, where the object feedback information is the information fed back by the target object for the target recommendation content;

[0138] Based on the object feedback information, screen out the preferred recommendation information and non-preferred recommendation information from the initial recommendation information;

[0139] According to the preferred recommendation information and the non-preferred recommendation information, adjust the parameters of the information generation model to obtain the adjusted information generation model.

[0140] It should be noted that preference recommendation information refers to the information filtered from the initial recommendation information and explicitly preferred by the target object. The target recommended content corresponding to the preference recommendation information usually receives a high evaluation, a long stay time, or other positive feedback from the target object, indicating that the target object highly approves of it. Non-preference recommendation information refers to the information filtered from the initial recommendation information that has not been selected or explicitly preferred by the target object. The target recommended content corresponding to the non-preference recommendation information usually receives a low evaluation, a short stay time, or other negative feedback from the target object, indicating that the target object has little interest in it or is dissatisfied. Optionally, the non-preference recommendation information may also include replacement recommended content that has not been exposed (not shown to the target object).

[0141] Exemplarily, on a news recommendation platform, if an article receives a rating of 4 stars or above and the average user stay time exceeds 5 minutes, the corresponding initial recommendation information is determined as preference recommendation information; conversely, if the article rating is below 2 stars and the stay time is less than 1 minute, the corresponding initial recommendation information is determined as non-preference recommendation information.

[0142] In practical applications, the process of adjusting the parameters of the information generation model based on the preference recommendation information and the non-preference recommendation information can be regarded as a process of performing DPO on the information generation model. When adjusting the parameters of the information generation model according to the preference recommendation information and the non-preference recommendation information, the preference loss value can be calculated based on the preference recommendation information and the non-preference information, and the model parameters of the information generation model can be adjusted according to the preference loss value to obtain the adjusted information generation model.

[0143] It is worth noting that when adjusting the parameters of the information generation model according to the preference recommendation information and the non-preference recommendation information, all the model parameters of the information generation model can be adjusted, or the LoRA method can be used to only update the parameters that can be represented by low-rank decomposition in the information generation model, thereby reducing the amount of model parameter adjustment and improving the parameter adjustment efficiency of the information generation model.

[0144] Applying the solution of the embodiments of this specification, by adjusting the parameters of the information generation model according to the preference recommendation information and the non-preference recommendation information, the adjusted information generation model can generate recommended content that better conforms to the preferences of the target object.

[0145] Considering that the number of model parameters of the information generation model is relatively large and the computing resources of the client are limited, the content recommendation method proposed in the embodiments of this specification can be applied to a content recommendation system as shown in Figure 2 but is not limited thereto. See Figure 2 , Figure 2The architecture diagram of a content recommendation system provided by an embodiment of this specification is shown. The content recommendation system may include a client 202 and a server 204;

[0146] The client 202 is used to send the attribute information of the target object to the server 204;

[0147] The server 204 is used to input the attribute information into an information generation model to obtain initial recommendation information that matches the target recommendation scenario; use the initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content; and display the target recommendation content to the target object in the target recommendation scenario.

[0148] It should be noted that the candidate recommendation content is the content that is screened based on the initial recommendation information and matches the initial recommendation information.

[0149] In practical applications, the information generation model may be deployed in the server 204. The server 204 can connect to one or more clients 202 through a local area network connection, a wide area network connection, an Internet connection, or other types of data networks. The data transmitted by the client 202 may need to be processed such as encoding, transcoding, and compression before being sent to the server 204. The client 202 can be a browser, an application (APP), or a web application such as a HyperText Markup Language 5 (H5) application, or a light application (also known as a mini-program, a lightweight application program), or a cloud application, etc. The client 202 can be developed based on the software development kit (SDK) provided by the server 204 for the corresponding service, such as developed based on the Real Time Communication (RTC) SDK. The client 202 can be deployed in an electronic device and needs to rely on the device or certain APPs in the device to run. The electronic device can, for example, have a display screen and support information browsing, etc., such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can usually be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc. The client 202 can also interact with the user through a user graphical interface to call the information generation model, thereby implementing the content recommendation method provided by the embodiment of this specification.

[0150] The server 204 may include servers that provide various services. For example, a server that provides communication services for multiple clients, or a server for background training that supports models used on the client, or a server that processes data sent by the client, etc. It should be noted that the server 204 may be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server may also be a server of a distributed system, or a server combined with a blockchain. The server may also be a cloud server of 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, content delivery network (CDN, Content Delivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0151] It is worth noting that the content recommendation method provided in the embodiments of this specification is generally executed by the server. However, in other embodiments of this specification, when the operating resources of the client can meet the deployment and running conditions of the information generation model, the client may also have a similar function to the server, so as to execute the content recommendation method provided in the embodiments of this specification. In other embodiments, the content recommendation method provided in the embodiments of this specification may also be jointly executed by the client and the server.

[0152] See Figure 3 , Figure 3 shows the process timing diagram of a content recommendation method provided in an embodiment of this specification. During the content recommendation process, the server and the client perform data interaction;

[0153] The client is used to send the attribute information of the target object to the server.

[0154] The server is used to input the attribute information and the scenario constraint information of the target recommendation scenario into the information generation model to obtain initial recommendation information that matches the target recommendation scenario. The initial recommendation information includes at least one of text recommendation information and image recommendation information. Screen out candidate recommendation content that matches the initial recommendation information from the original content. Perform quality detection on the initial recommendation information to obtain a quality detection result. According to the quality detection result, screen out the target initial recommendation information from the initial recommendation information. Use the target initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain replacement recommendation content. Sort the replacement recommendation content according to the target initial recommendation information, and according to the sorting result, screen out the target recommendation content from the replacement recommendation content. Typeset the target recommendation content based on the scenario layout information of the target recommendation scenario to obtain the typeset target recommendation content. Send the typeset target recommendation content to the client.

[0155] The client is also used to display the formatted target recommended content in the target recommendation scenario.

[0156] The server is also used to obtain object feedback information, and based on the object feedback information, adjust the parameters of the information generation model to obtain an adjusted information generation model.

[0157] Applying the solution of the embodiments of this specification, the client is responsible for collecting the attribute information of the target object and displaying the target recommended content. The attribute information of the target object is collected by the client and transmitted to the server. After several stages of processing including initial recommended information generation, candidate recommended content matching and screening, initial recommended information detection and screening, and replacement recommended content sorting, the server transmits the target recommended content to the client. Moreover, the object feedback information of the target object can be directly uploaded to the server, enabling the server to adjust the parameters of the information generation model based on the object feedback information, continuously optimizing the initial recommended information generation strategy of the information generation model, so that the information generation model can generate initial recommended information that better conforms to the system preferences and the interests and hobbies of the target object.

[0158] It should be noted that after generating the target recommended content, the generated target recommended content is comprehensively evaluated through offline metrics and A / B testing, and it is determined that the content recommendation solution proposed in the embodiments of this specification significantly improves the recommendation effect.

[0159] The following combination of attached Figure 4 , taking the application of the content recommendation method provided in this specification in the note recommendation scenario as an example, further illustrates the content recommendation method. Among them, Figure 4 shows a flowchart of a note recommendation method provided by an embodiment of this specification, specifically including the following steps:

[0160] Step 402: Obtain the attribute information of the target object.

[0161] Step 404: Input the attribute information into the information generation model to obtain note recommendation information that matches the note recommendation scenario.

[0162] Step 406: Use the note recommendation information to adjust the candidate recommended notes to obtain target recommended notes, where the candidate recommended notes are notes that are screened based on the note recommendation information and match the note recommendation information.

[0163] Step 408: In the note recommendation scenario, display the target recommended notes to the target object.

[0164] It should be noted that a note refers to written words, images, or other forms of content created by an individual or a team for the purpose of recording information, organizing thoughts, reminding, etc. during learning, working, or daily life. A note can be a few simple lines of text or a complex electronic document containing various elements such as charts, links, and multimedia files. Notes are widely used in various life and work scenarios, such as knowledge point notes and teaching plan notes in learning and education scenarios, diet record notes and exercise record notes in health and life scenarios, and so on. The note recommendation information includes at least one of the title recommendation information and the cover recommendation information. The title is a highly generalized and brief description of the note content and can serve as an entry point for the target object to quickly understand the note content. The cover refers to an image or graphic attached to the front of the note, which can enhance the visual effect and recognition of the note. The cover can be a beautiful picture, chart, illustration, or even a simple icon, aiming to convey the content or emotion of the note through visual elements.

[0165] In actual application, the implementation manners of steps 402 to 408 are the same as those of steps 102 to 108 above, and the embodiments of this specification will not be elaborated herein.

[0166] Applying the solution of the embodiments of this specification, by using the information generation model to deeply analyze the attribute information of the target object, the user's interests and hobbies can be understood more accurately, and thus highly personalized note recommendation information can be provided based on the characteristics of the target object. Using the note recommendation information to adjust the candidate recommended notes makes the finally recommended target recommended notes more in line with the interests and hobbies of the target object, realizes the dynamic adaptation between the interests and hobbies and the notes, significantly improves the attractiveness of the target recommended notes to the target object, and further improves the accuracy of note recommendation. Moreover, since the note recommendation information matches the target recommendation scenario, it is ensured that the target recommended notes finally presented to the target object not only adapt to the interests and hobbies of the target object in terms of content itself, but also achieve very good effects in terms of visual layout and interaction design, enhancing the consistency of note recommendation and the satisfaction of the target object.

[0167] See Figure 5 , Figure 5The figure shows a schematic diagram of a note recommendation interface provided by an embodiment of this specification. The note recommendation interface includes a "Follow" control, a "Discover" control, a "Location Exploration" control, a "Home" control, a "Popular" control, a "+", a "Messages" control, and a "My" control. Among them, after the user triggers the "Follow" control, the note recommendation interface can display target recommended notes of the followed authors to the user. After the user triggers the "Discover" control, the note recommendation interface can display personalized target recommended notes to the user. After the user triggers the "Location Exploration" control, the note recommendation interface can display target recommended notes near the user's geographical location or in a specific area to the user. The "Home" control is used to display personalized target recommended content to the user after being triggered. The "Home" is the default page for the user to enter the note recommendation interface. After the user triggers the "Popular" control, the note recommendation interface can display target recommended notes that are currently popular or highly discussed to the user. After the user triggers the "+", the user can perform operations such as creating a new note, uploading a picture, and recording a video. After the user triggers the "Messages" control, the user can receive and send messages, notifications, etc., to maintain communication and interaction among users. After the user triggers the "My" control, the note recommendation interface can display functions such as personal information, favorites, and history to the user, facilitating the user to manage and view their own notes and related information.

[0168] As Figure 5 shown, the note recommendation interface also includes four target recommended notes presented in a two-column layout. Among them, the first target recommended note is a video note published by author 1, with the title "Holiday Travel Outfit" and a cover image of a beautiful woman's outfit. The number of likes for the first target recommended note is 22. The second target recommended note is a note published by author 2, with the title "Good Fashion Items" and a cover image of a long-sleeved coat. The number of likes for the second target recommended note is 12. The third target recommended note is a note published by author 3, with the title "Fashion Accessories" and a cover image of a fashionable shawl. The number of likes for the third target recommended note is 11. The fourth target recommended note is a note published by author 4, with the title "XX Store is Really Fun to Shop" and a cover image of a water cup. The number of likes for the fourth target recommended note is 23. Taking the first target recommended note as an example, when the user triggers the first target recommended note, they can be redirected to the details page corresponding to the first target recommended note. The details page includes basic attribute information such as an outfit image, the content "The long-haired girl is carrying a crossbody bag, wearing a top of XX brand, paired with pants of XX brand, relaxed and comfortable, meeting both comfort and style", "XX brand", and the corresponding brand logo.

[0169] In traditional recommendation solutions, the methods and logics for generating recommended content are relatively fixed, resulting in insufficient generalization ability for recommending to new users, making it difficult to adapt to the rapidly changing interests and scenario requirements of users, leading to low flexibility and difficulty in being flexibly applied in multiple scenarios. To address this problem, the embodiments of this specification propose a training method for an information generation model, which continuously trains the information generation model based on object feedback information, enabling the information generation model to adapt to different application scenarios and enhancing the flexibility and applicability of the generation model. Refer to Figure 6 , Figure 6 shows a flowchart of a training method for an information generation model provided by an embodiment of this specification, which specifically includes the following steps:

[0170] Step 602: Obtain object feedback information, where the object feedback information is the information feedback by the target object for the target recommended content, and the target recommended content is obtained by adjusting the candidate recommended content based on the initial recommended information that matches the target recommendation scenario. The initial recommended information is obtained by the information generation model based on the attribute information of the target object, and the candidate recommended content is the content that matches the initial recommended information screened based on the initial recommended information.

[0171] Step 604: Adjust the parameters of the information generation model according to the object feedback information to obtain an adjusted information generation model.

[0172] It should be noted that the implementation manners of steps 602 to 604 can refer to the implementation manners of the above "Obtain object feedback information, where the object feedback information is the information feedback by the target object for the target recommended content; based on the object feedback information, adjust the parameters of the information generation model to obtain an adjusted information generation model", and the embodiments of this specification will not elaborate further.

[0173] Applying the solution of the embodiments of this specification, by continuously adjusting the parameters of the information generation model according to the object feedback information truly feedback by the target object, the information generation model can adapt to the rapidly changing interests and scenario requirements of users, generate initial recommended information that better fits the interests of the target object, and improve the flexibility, accuracy of the information generation model and the generalization ability for recommending to new users.

[0174] Refer to Figure 7 , Figure 7 shows a flowchart of an information processing method based on an information generation model provided by an embodiment of this specification. The information processing method based on the information generation model is applied to a task platform and specifically includes the following steps:

[0175] Step 702: Receive a model request sent by a terminal device.

[0176] Step 704: Based on the model request, determine a target information generation model from multiple information generation models, where the target information generation model is used in the execution process of the content recommendation method.

[0177] It should be noted that the target information generation model is an information generation model applicable to the target recommendation scenario. The model request includes at least one of the scenario identifier of the target recommendation scenario, the scenario input data of the target recommendation scenario, and the model specification parameters. There are multiple ways to determine the target information generation model from multiple information generation models based on the model request. In the first possible implementation manner of this specification, the corresponding target information generation model can be searched from at least one information generation model included in the model library based on the model request; in the second possible implementation manner of this specification, the target information generation model can be obtained through training based on the model request; in the third possible implementation manner of this specification, the target information generation model can be constructed based on the model request.

[0178] Exemplarily, first, based on the scenario identifier of the target recommendation scenario, search for at least one pre-trained information generation model from the model library, then based on the model specification parameters, screen and obtain an initial information generation model from at least one information generation model, and then based on the scenario input data of the target recommendation scenario, train the screened initial information generation model to obtain a target information generation model suitable for the user's needs.

[0179] Applying the solution of the embodiments of this specification, obtaining the target information generation model according to the user's needs realizes personalized model services, provides an efficient, flexible and easy-to-use model service method for users, and improves the user experience.

[0180] In an optional embodiment of this specification, the model request includes the scenario identifier of the target recommendation scenario; the above-mentioned determining the target information generation model from multiple information generation models based on the model request may include the following steps:

[0181] Based on the scenario identifier of the target recommendation scenario, search for a target information generation model adapted to the target recommendation scenario from the model library, where the model library stores multiple information generation models adapted to different recommendation scenarios.

[0182] It should be noted that the scenario identifier refers to a unique or specific label used to distinguish different recommendation scenarios. The model library is a database for storing and managing various pre-trained deep learning models. Multiple information generation models adapted to different recommendation scenarios cover different content recommendation scenarios and requirements. The model library allows users to select appropriate models according to their own needs, or directly call the models through application programming interfaces for content recommendation.

[0183] Multiple models stored in the model library and applicable to different recommendation scenarios are optimized for specific application environments. For example, based on the scenario identifier "note recommendation" of the target recommendation scenario, a target information generation model suitable for the note recommendation scenario can be found from the model library.

[0184] Applying the solution of the embodiments of this specification, based on the scenario requirements, the target information generation model suitable for the scenario can be accurately found through the scenario identifier, making the content recommendation more accurate and more in line with the scenario, thereby improving the user experience and the content recommendation quality.

[0185] In an optional embodiment of this specification, the model request includes the scenario input data of the target recommendation scenario; based on the model request, determining the target information generation model from multiple information generation models may include the following steps:

[0186] Determine an initial information generation model suitable for the target recommendation scenario from multiple information generation models;

[0187] Based on the scenario input data of the target recommendation scenario, adjust the parameters of the initial information generation model to obtain the target information generation model.

[0188] It should be noted that the initial information generation model refers to the model in multiple information generation models that is suitable for the target recommendation scenario. The initial information generation model may not only be applicable to the target recommendation scenario but also to other recommendation scenarios, and it is a general information generation model that can be applicable to different recommendation scenarios. The initial information generation model can be used to process the tasks of the target recommendation scenario, but the effect may not be very good. At this time, the parameters of the initial information generation model can be adjusted based on the scenario input data of the target recommendation scenario. For example, optimizing the initial information generation model based on the scenario input data of the note recommendation scenario can obtain the target information generation model suitable for the note recommendation scenario. The scenario input data of the target recommendation scenario can be understood as the sample set of the sample recommendation task (including preference recommendation information and non-preference recommendation information).

[0189] Applying the solution of the embodiments of this specification, based on the scenario requirements, further training the general initial information generation model through the scenario input data to obtain the target information generation model suitable for the scenario, making the target information generation model more in line with the scenario, thereby improving the user experience and the content recommendation quality.

[0190] In an optional embodiment of this specification, the model request includes model specification parameters; based on the model request, determining the target information generation model from multiple information generation models may include the following steps:

[0191] Based on the model specification parameters, find the corresponding target information generation model from the model library, where the model library stores information generation models with multiple different model specification parameters.

[0192] It should be noted that the model specification parameters refer to various parameters that define the model structure and behavior. These parameters can be roughly divided into two categories: model parameters (learnable parameters) and hyperparameters. Model parameters refer to the parameters that are automatically adjusted through the backpropagation algorithm during model training, including but not limited to the weight matrix (weights) and bias terms (biases). For example, in a simple fully connected layer, the weight matrix is a two-dimensional tensor that connects the neurons of the input layer and the output layer; the bias term is a one-dimensional vector that provides an additional offset value for each output neuron. Hyperparameters refer to the parameters set before starting model training to control the learning process and architecture of the model. Hyperparameters include but not limited to the learning rate (Learning Rate) and the number of neurons per layer (Number of Neurons per Layer), and specific selections are made according to the actual situation.

[0193] Applying the solution of the embodiment of this specification, based on the model specification parameters, the corresponding target information generation model can be accurately found, ensuring the efficient and stable operation of the target information generation model and improving the user experience.

[0194] In an optional embodiment of this specification, after determining the target information generation model from multiple information generation models based on the model request, the following steps may further be included:

[0195] Deploy the target information generation model, and based on the target information generation model, construct a recommendation task processing interface so that the terminal device can schedule the target information generation model to execute the target recommendation task.

[0196] It should be noted that the recommendation task processing interface is an interactive programming interface for the terminal device to schedule the target information generation model to perform target recommendation task processing, and the recommendation task processing interface is usually provided in the form of an application programming interface.

[0197] In actual applications, there are multiple ways to deploy the target information generation model. In one possible implementation, the target information generation model can be deployed on the cloud-side device using the infrastructure provided by the cloud service provider. In another possible implementation, the target information generation model can be deployed on the edge device using a lightweight framework. For example, the target information generation model can be deployed on a distributed system, and based on the target information generation model, a recommendation task processing interface can be constructed and provided to the terminal device so that the terminal device can schedule the target information generation model to execute the target recommendation task.

[0198] Applying the solution provided in the embodiments of this specification to deploy a target information generation model and construct a recommendation task processing interface based on the target information generation model can enable the terminal device to efficiently call the target information generation model and improve the processing quality and response speed of the target recommendation task.

[0199] See Figure 8 , Figure 8 shows a schematic structural diagram of a task platform provided in an embodiment of this specification. The task platform 800 includes a request interface 802 and a response unit 804;

[0200] The request interface 802 is used to receive a model request sent by the terminal device, where the model request includes at least one of a scenario identifier of the target recommendation scenario, scenario input data of the target recommendation scenario, and model specification parameters;

[0201] The response unit 804 is used to determine a target information generation model from multiple information generation models based on the model request, where the target information generation model is used in the execution process of the content recommendation method.

[0202] In an optional embodiment of this specification, the task platform further includes a recommendation task processing interface, which is constructed based on the target information generation model;

[0203] The recommendation task processing interface is used for the terminal device to schedule and execute the target recommendation task.

[0204] Applying the solution of the embodiments of this specification, the task platform adapts to user needs to obtain the target information generation model, realizes personalized model services, provides a highly efficient, flexible and easy-to-use model service platform for users, and improves the user experience.

[0205] The above is a schematic solution of a task platform in this embodiment. It should be noted that the technical solution of this task platform and the technical solution of the above information processing method based on the information generation model belong to the same concept. For the details not described in the technical solution of the task platform, reference can be made to the description of the technical solution of the above information processing method based on the information generation model.

[0206] Corresponding to the above embodiment of the content recommendation method, this specification also provides an embodiment of a content recommendation device, Figure 9 shows a schematic structural diagram of a content recommendation device provided in an embodiment of this specification. As Figure 9 shown, the device includes:

[0207] The first acquisition module 902 is configured to acquire attribute information of the target object;

[0208] An input module 904, configured to input attribute information into an information generation model to obtain initial recommendation information that matches a target recommendation scenario;

[0209] A first adjustment module 906, configured to use the initial recommendation information to adjust candidate recommendation content to obtain target recommendation content, where the candidate recommendation content is content that is filtered based on the initial recommendation information and matches the initial recommendation information;

[0210] A display module 908, configured to display the target recommendation content to a target object in the target recommendation scenario.

[0211] Optionally, the initial recommendation information includes at least one of text recommendation information and image recommendation information; the input module 904 is further configured to input the attribute information and the scenario constraint information of the target recommendation scenario into the information generation model to obtain text recommendation information; and / or input the attribute information and the scenario constraint information into the information generation model to obtain image recommendation information.

[0212] Optionally, the candidate recommendation content includes original recommendation information; the first adjustment module 906 is further configured to use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content.

[0213] Optionally, the first adjustment module 906 is further configured to use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain replacement recommendation content; sort the replacement recommendation content according to the initial recommendation information, and screen out the target recommendation content from the replacement recommendation content according to the sorting result.

[0214] Optionally, the device further includes: a first screening module, configured to perform quality detection on the initial recommendation information to obtain a quality detection result; screen out target initial recommendation information from the initial recommendation information according to the quality detection result; the first adjustment module 906 is further configured to use the target initial recommendation information to adjust the candidate recommendation content to obtain the target recommendation content.

[0215] Optionally, the device further includes: a second screening module, configured to obtain object feedback information, where the object feedback information is information fed back by the target object for the target recommendation content; based on the object feedback information, screen out preferred recommendation information and non-preferred recommendation information from the initial recommendation information; adjust the parameters of the information generation model according to the preferred recommendation information and the non-preferred recommendation information to obtain an adjusted information generation model.

[0216] Applying the solution of the embodiments of this specification, by deeply analyzing the attribute information of the target object using the information generation model, the user's interests and hobbies can be understood more accurately, so as to provide highly personalized initial recommendation information based on the characteristics of the target object. Using the initial recommendation information to adjust the candidate recommendation content makes the finally recommended target recommendation content more in line with the interests and hobbies of the target object, realizes the dynamic adaptation between the interests and hobbies and the content, significantly improves the attractiveness of the target recommendation content to the target object, and further improves the accuracy of content recommendation. Moreover, since the initial recommendation information matches the target recommendation scenario, it is ensured that the target recommendation content finally presented to the target object not only adapts to the interests and hobbies of the target object in terms of the content itself, but also achieves very good results in terms of visual layout and interaction design, enhancing the consistency of content recommendation and the satisfaction of the target object.

[0217] The above is a schematic solution of a content recommendation device according to this embodiment. It should be noted that the technical solution of this content recommendation device and the technical solution of the above content recommendation method belong to the same concept. For the details not described in detail in the technical solution of the content recommendation device, reference can be made to the description of the technical solution of the above content recommendation method.

[0218] Corresponding to the above information generation model training method embodiment, this specification also provides an information generation model training device embodiment. Figure 10 Fig. shows a schematic structural diagram of an information generation model training device provided by an embodiment of this specification. As Figure 10 shown, the device includes:

[0219] A second acquisition module 1002, configured to acquire object feedback information, where the object feedback information is information feedback by the target object for the target recommendation content, and the target recommendation content is obtained by adjusting the candidate recommendation content based on the initial recommendation information that matches the target recommendation scenario. The initial recommendation information is obtained by the information generation model based on the attribute information of the target object, and the candidate recommendation content is content that matches the initial recommendation information screened based on the initial recommendation information;

[0220] A second adjustment module 1004, configured to adjust the parameters of the information generation model according to the object feedback information to obtain an adjusted information generation model.

[0221] Applying the solution of the embodiments of this specification, by continuously adjusting the parameters of the information generation model according to the object feedback information truly fed back by the target object, the information generation model can adapt to the rapidly changing interests and hobbies and scenario requirements of the user, generate initial recommendation information that more conforms to the interests and hobbies of the target object, and improves the flexibility, accuracy of the information generation model and the generalization ability for new user recommendations.

[0222] The above is a schematic solution of an information generation model training device according to this embodiment. It should be noted that the technical solution of this information generation model training device and the technical solution of the above information generation model training method belong to the same concept. For the details not described in detail in the technical solution of the information generation model training device, reference can be made to the description of the technical solution of the above information generation model training method.

[0223] Corresponding to the above embodiment of the information processing method based on the information generation model, this specification also provides an embodiment of an information processing device based on the information generation model. Figure 11 The following shows a schematic structural diagram of an information processing device based on the information generation model provided by an embodiment of this specification. As Figure 11 shown, this device is applied to a task platform and includes:

[0224] A receiving module 1102, configured to receive a model request sent by a terminal device;

[0225] A determining module 1104, configured to determine a target information generation model from multiple information generation models based on the model request, where the target information generation model is used in the execution process of the content recommendation method.

[0226] Optionally, the model request includes a scenario identifier of a target recommendation scenario; the determining module 1104 is further configured to search for a target information generation model adapted to the target recommendation scenario from a model library based on the scenario identifier of the target recommendation scenario, where the model library stores multiple information generation models adapted to different recommendation scenarios.

[0227] Optionally, the model request includes scenario input data of a target recommendation scenario; the determining module 1104 is further configured to determine an initial information generation model adapted to the target recommendation scenario from multiple information generation models; and perform parameter adjustment on the initial information generation model based on the scenario input data of the target recommendation scenario to obtain the target information generation model.

[0228] Applying the solution of the embodiment of this specification, the target information generation model is obtained according to the user's needs, realizing personalized model services, providing the user with an efficient, flexible and easy-to-use model service method, and improving the user experience.

[0229] The above is a schematic solution of an information processing device based on the information generation model according to this embodiment. It should be noted that the technical solution of this information processing device based on the information generation model and the technical solution of the above information processing method based on the information generation model belong to the same concept. For the details not described in detail in the technical solution of the information processing device based on the information generation model, reference can be made to the description of the technical solution of the above information processing method based on the information generation model.

[0230] Figure 12 The block diagram of a computing device provided by an embodiment of this specification is shown. The components of the computing device 1200 include but are not limited to a memory 1210 and a processor 1220. The processor 1220 is connected to the memory 1210 through a bus 1230, and a database 1250 is used to store data.

[0231] The computing device 1200 further includes an access device 1240, which enables the computing device 1200 to communicate via one or more networks 1260. Examples of these networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1240 may include one or more of any type of wired or wireless network interfaces (e.g., Network Interface Card (NIC)), such as IEEE802.11 Wireless Local Area Networks (WLAN) wireless interface, Worldwide Interoperability for Microwave Access (Wi-MAX) interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth interface, Near Field Communication (NFC) interface, and so on.

[0232] In an embodiment of this specification, the above components of the computing device 1200 and Figure 12 other components not shown may also be connected to each other, for example, through a bus. It should be understood that Figure 12 the shown block diagram of the computing device is only for example purposes and is not a limitation on the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0233] The computing device 1200 can be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, personal digital assistant, laptop computer, notebook computer, netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.) or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer. The computing device 1200 can also be a mobile or stationary server.

[0234] Wherein, the processor 1220 is configured to execute a computer program / instructions, and when the computer program / instructions are executed by the processor, the steps of the above content recommendation method, or information generation model training method, or information processing method based on the information generation model are implemented.

[0235] The above is a schematic solution of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solutions of the above content recommendation method, information generation model training method, and information processing method based on the information generation model belong to the same concept. For the details not described in the technical solution of the computing device, reference can be made to the descriptions of the technical solutions of the above content recommendation method, information generation model training method, or information processing method based on the information generation model.

[0236] An embodiment of this specification also provides a computer-readable storage medium, which stores a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above content recommendation method, or information generation model training method, or information processing method based on the information generation model are implemented.

[0237] The above is a schematic solution of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solutions of the above content recommendation method, information generation model training method, and information processing method based on the information generation model belong to the same concept. For the details not described in the technical solution of the storage medium, reference can be made to the descriptions of the technical solutions of the above content recommendation method, information generation model training method, or information processing method based on the information generation model.

[0238] An embodiment of this specification also provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above content recommendation method, or information generation model training method, or information processing method based on the information generation model are implemented.

[0239] The above is a schematic solution of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solutions of the above content recommendation method, information generation model training method, and information processing method based on the information generation model belong to the same concept. For the details not described in detail in the technical solution of the computer program product, reference can be made to the descriptions of the above content recommendation method, information generation model training method, or information processing method based on the information generation model.

[0240] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0241] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0242] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the embodiments of this specification are not limited by the described order of actions, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0243] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0244] The preferred embodiments of the present specification disclosed above are only used to help explain the present specification. The alternative embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, according to the content of the embodiments of the present specification, many modifications and variations can be made. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present specification, so that those skilled in the art can well understand and utilize the present specification. The present specification is only limited by the claims and their full scope and equivalents.

Claims

1. A content recommendation method, characterized in that Including: Obtain the attribute information of the target object; Input the attribute information into an information generation model to obtain initial recommendation information that matches the target recommendation scenario; Use the initial recommendation information to adjust candidate recommendation content to obtain target recommendation content, where the candidate recommendation content is content that is filtered based on the initial recommendation information and matches the initial recommendation information; In the target recommendation scenario, display the target recommendation content to the target object.

2. The method according to claim 1, wherein The initial recommendation information includes at least one of text recommendation information and image recommendation information; The step of inputting the attribute information into an information generation model to obtain initial recommendation information that matches the target recommendation scenario includes: Input the attribute information and the scenario constraint information of the target recommendation scenario into the information generation model to obtain the text recommendation information; and / or, Input the attribute information and the scenario constraint information into the information generation model to obtain the image recommendation information.

3. The method according to claim 1, wherein The candidate recommendation content includes original recommendation information; The step of using the initial recommendation information to adjust the candidate recommendation content to obtain target recommendation content includes: Use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content.

4. The method according to claim 3, wherein The step of using the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain the target recommendation content includes: Use the initial recommendation information to replace the original recommendation information in the candidate recommendation content to obtain replacement recommendation content; Sort the replacement recommendation content according to the initial recommendation information, and filter out the target recommendation content from the replacement recommendation content according to the sorting result.

5. The method according to any one of claims 1 to 4, characterized in that Before using the initial recommendation information to adjust the candidate recommendation content to obtain target recommendation content, it further includes: Perform quality detection on the initial recommendation information to obtain a quality detection result; Filter out target initial recommendation information from the initial recommendation information according to the quality detection result; The step of using the initial recommendation information to adjust the candidate recommendation content to obtain target recommendation content includes: Use the target initial recommendation information to adjust the candidate recommendation content to obtain target recommendation content.

6. The method according to any one of claims 1 to 4, characterized in that, After displaying the target recommendation content to the target object in the target recommendation scenario, it further includes: Obtain object feedback information, where the object feedback information is information that the target object gives feedback on the target recommendation content; Based on the object feedback information, filter out preferred recommendation information and non-preferred recommendation information from the initial recommendation information; Adjust the parameters of the information generation model according to the preferred recommendation information and the non-preferred recommendation information to obtain an adjusted information generation model.

7. A method for training an information generation model, characterized in that, Including: Obtain object feedback information, where the object feedback information is information fed back by a target object for target recommended content, and the target recommended content is obtained by adjusting candidate recommended content based on initial recommended information matching a target recommendation scenario. The initial recommended information is obtained by an information generation model based on the attribute information of the target object, and the candidate recommended content is content matching the initial recommended information screened based on the initial recommended information. According to the object feedback information, adjust the parameters of the information generation model to obtain an adjusted information generation model.

8. An information processing method based on an information generation model, characterized in that, Applied to a task platform, including: Receive a model request sent by a terminal device. Based on the model request, determine a target information generation model from multiple information generation models, where the target information generation model is used in the execution process of the method according to any one of claims 1 to 6.

9. The method according to claim 8, wherein The model request includes a scenario identifier of a target recommendation scenario. The determining, based on the model request, a target information generation model from multiple information generation models includes: Based on the scenario identifier of the target recommendation scenario, search in a model library for a target information generation model adapted to the target recommendation scenario, where the model library stores multiple information generation models adapted to different recommendation scenarios.

10. The method according to claim 8, wherein The model request includes scenario input data of a target recommendation scenario. The determining, based on the model request, a target information generation model from multiple information generation models includes: Determine an initial information generation model adapted to the target recommendation scenario from the multiple information generation models. Based on the scenario input data of the target recommendation scenario, adjust the parameters of the initial information generation model to obtain a target information generation model.

11. A task platform, characterized in that, Includes a request interface and a response unit; The request interface is used to receive a model request sent by a terminal device, where the model request includes at least one of a scenario identifier of a target recommendation scenario, scenario input data of a target recommendation scenario, and model specification parameters. The response unit is used to determine a target information generation model from multiple information generation models based on the model request, where the target information generation model is used in the execution process of the method according to any one of claims 1 to 6.

12. A computing device, characterized in that, Includes: A memory and a processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium, characterized in that, It stores computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 10 are implemented.

14. A computer program product, characterized in that, Includes computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images