Content generation method and device

Abstract

The embodiment of the invention provides a content generation method and device, and the method comprises the steps: obtaining a target content generation option corresponding to a content generation task, and determining a corresponding scene parameter based on the target content generation option, the target content generation option is a currently selected content generation option in the content generation options of at least one dimension in the target industry; based on the target content generation option and the scene parameters, matching a corresponding target cue word template in a cue word database; and generating target prompt information based on the target content generation option, the scene parameter and the target prompt word template, and generating target content in the target industry through a content generation model based on the target prompt information. According to the method, the real demand of the user for the to-be-generated content is obtained by utilizing the content generation option, the generation of the content is more finely controlled by expanding parameters, and then the target content is automatically generated based on the deep learning model, so that the accuracy and efficiency of content generation are improved.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and more particularly to a content generation method and apparatus. Background Art

[0002] With the rapid development of computer and artificial intelligence technologies, deep learning models have shown great potential in content generation, enabling them to perform a variety of downstream content generation tasks. Currently, the operational barriers to content generation are high, and users often struggle to provide sufficiently specific and clear prompts. This makes it difficult for deep learning models to generate content that meets user needs, resulting in insufficient accuracy. To optimize content to better meet user needs, deep learning models typically require multiple iterations of generation and adjustment, resulting in low efficiency. Therefore, a content generation method that can improve both accuracy and efficiency is urgently needed. Summary of the Invention

[0003] In view of this, embodiments of this specification provide a content generation method. One or more embodiments of this specification also relate to a content generation apparatus, a computing device, a computer-readable storage medium, and a computer program product to address technical deficiencies in the prior art.

[0004] According to a first aspect of an embodiment of this specification, a content generation method is provided, including:

[0005] Obtaining a target content generation option corresponding to the content generation task, and determining corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry;

[0006] Based on the target content generation options and scenario parameters, the corresponding target prompt word template is matched in the prompt word database;

[0007] Based on the target content generation options, scenario parameters and target prompt word template, target prompt information is generated, and based on the target prompt information, the target content under the target industry is generated through the content generation model.

[0008] According to a second aspect of the embodiments of this specification, there is provided a content generation apparatus, including:

[0009] a determination module configured to obtain a target content generation option corresponding to the content generation task, and determine corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry;

[0010] a matching module configured to match corresponding target prompt word templates in a prompt word database based on target content generation options and scenario parameters;

[0011] The generation module is configured to generate target prompt information based on target content generation options, scenario parameters and target prompt word templates, and generate target content under the target industry through a content generation model based on the target prompt information.

[0012] According to a third aspect of an embodiment of this specification, a computing device is provided, including:

[0013] memory and processor;

[0014] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned content generation method are implemented.

[0015] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned content generation method are implemented.

[0016] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above-mentioned content generation method when executed by a processor.

[0017] One embodiment of this specification implements the following steps: obtaining a target content generation option corresponding to a content generation task; determining corresponding scenario parameters based on the target content generation option, wherein the target content generation option is the currently selected content generation option from at least one dimension of content generation options within a target industry; matching a corresponding target prompt word template in a prompt word database based on the target content generation option and the scenario parameters; generating target prompt information based on the target content generation option, the scenario parameters, and the target prompt word template; and generating target content for the target industry using a content generation model based on the target prompt information. In this way, during the content generation process, users only need to make selections guided by the content generation options to obtain the desired content, thereby lowering the barrier to entry for content generation. In addition, through the content generation options of at least one dimension under the target industry, the key information of the content to be generated is structured, and the target content generation options generated according to the user's selection in the content generation options can fully reflect the user's needs for the key information in the content to be generated, and expand the parameters - scene parameters on the basis of the target content generation options selected by the user, further refine the key information of the content to be generated, so that the target prompt word template obtained based on the target content generation options and scene parameters can more accurately meet the user's needs, and then, based on the target content generation options, scene parameters and target prompt word template, the target prompt information generated can more accurately and clearly reflect the user's needs, and ultimately make the generated target content more in line with the user's needs, thus improving the accuracy and efficiency of content generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is an application architecture diagram of content generation provided by an embodiment of this specification;

[0019] Figure 2 is a flow chart of a content generation method provided by one embodiment of this specification;

[0020] Figure 3 is a schematic diagram of a content generation option on a user interface provided by one embodiment of this specification;

[0021] Figure 4 This is a schematic structural diagram of a content generation device provided by one embodiment of this specification;

[0022] Figure 5 This is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0023] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0024] 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," "an," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include 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 associated listed items.

[0025] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, 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 "at the time of" or "when" or "in response to determining".

[0026] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, typically containing hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. A large model can also be called a cornerstone model / foundation model. It is pre-trained on a large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks and has good generalization capabilities, such as a large language model (LLM) and a multi-modal pre-training model.

[0027] When large models are used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. Large models can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, they can be applied to computer vision tasks such as visual question answering (VQA), image caption (IC), and image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of large models include digital assistants, intelligent robots, search, online education, office software, e-commerce, and intelligent design.

[0028] In this specification, a content generation method is provided. This specification also relates to a content generation apparatus, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0029] Considering the huge number of model parameters of large models and the limited computing resources of mobile terminals, the data processing method provided in the embodiment of the present application can be applied to Figure 1 The application scenarios shown are not limited to these. Figure 1 This is an application architecture diagram of content generation provided by an embodiment of this specification. Figure 1 In the illustrated application scenario, the large model is deployed on a server 10. Server 10 can be connected to one or more client devices 20 via a local area network (LAN), a wide area network (WAN), the Internet, or other types of data networks. Client devices 20 herein include, but are not limited to, smartphones, tablet computers, laptops, PDAs, personal computers, smart home devices, and in-vehicle devices. Client devices 20 can interact with users via a graphical user interface (GUI) to access the large model and implement the methods provided in the embodiments of this specification.

[0030] In an embodiment of the present disclosure, a system consisting of a client device and a server may perform the following steps: the client device receives a target content generation option and uploads it to the server, where the target content generation option is the currently selected content generation option from at least one dimension of content generation options within a target industry. The server determines corresponding scenario parameters based on the target content generation option; based on the target content generation option and the scenario parameters, matches a corresponding target prompt word template in a prompt word database; generates target prompt information based on the target content generation option, the scenario parameters, and the target prompt word template; and generates target content within the target industry using a content generation model based on the target prompt information.

[0031] It should be noted that, when the operating resources of the client device can meet the deployment and operating conditions of the large model, the embodiments of the present application can be carried out in the client device.

[0032] See also Figure 2 , Figure 2 A flow chart of a content generation method provided according to an embodiment of this specification is shown, which specifically includes the following steps 202 to 206.

[0033] Step 202: Obtain a target content generation option corresponding to the content generation task, and determine corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry.

[0034] If the embodiment of the present application is performed on the client side, the target content generation option is obtained by the client device after the user selects the content generation option for at least one dimension under the target industry displayed on the client's display interface. If the embodiment of the present application is performed on the server side, the target content generation option is sent by the client to the server side.

[0035] Among them, the target industry refers to the specific industry field where the content to be generated is expected to be applied.

[0036] Among them, content refers to various forms of information carriers generated through artificial intelligence technology, covering multiple modalities such as text, images, video, and audio.

[0037] Content generation options refer to user-configurable parameters when generating content. These parameters are used to capture the user's specific needs for content generation, thereby guiding the content generation model to create content that better meets the user's needs. Different target industries or different types of content correspond to different types of content generation options. The embodiments of this application do not impose specific restrictions on the specific content of content generation options.

[0038] For example, when the target industry is the fashion industry and the content to be generated is clothing images, the content generation options in at least one dimension may be:

[0039] Style and Construction: This section specifies the type of clothing (e.g., dress, shirt, jacket, pants, etc.), cut (e.g., slim, loose, A-line, H-line, etc.), and design details (neckline, sleeve, hem design) that the user desires. Style and Construction options help determine the basic outline and structural features of the garment.

[0040] Materials and fabrics: Identify the fabric type (e.g., cotton, silk, wool, denim, lace, etc.), material properties (e.g., gloss, transparency, elasticity, thickness, etc.), and pattern and texture (e.g., solid color, stripes, plaid, print, embroidery, etc.) of the garment you want. Material and fabric options help determine the visual effect and wearing experience of the garment.

[0041] Color and color matching: Clarify the main color (the primary color desired by the user), auxiliary color (secondary color used for matching), and color matching scheme (such as contrasting colors, adjacent colors, monochrome, etc.) of the clothing required by the user. Color and color matching options help generate designs that match the brand or personal style.

[0042] Style and Theme: This option allows you to specify the desired clothing style (e.g., minimalist, retro, street, business, sports), cultural elements (e.g., Chinese, Japanese, European, ethnic), and seasonal themes (e.g., spring / summer, autumn / winter, holiday specials). The Style and Theme options help generate clothing images with specific styles and themes.

[0043] Reference images: User-provided inspiration or style images.

[0044] Design description: The user’s description of the design concept, target audience, and usage scenarios.

[0045] Taking the insurance industry as an example, refer to Figure 3 , Figure 3 A schematic diagram showing content generation options on a user interface provided according to an embodiment of this specification is shown.

[0046] like Figure 3 As shown, when the target industry is the insurance industry and the content to be generated is copywriting, the content generation options for at least one dimension can be:

[0047] Topic selection: Users can choose from preset topics (such as product promotion, science education, brand stories, etc.), or customize by entering keywords for specific fields or products to ensure clear content direction.

[0048] Copy type: Specify the required copy format, such as social media posts, blog articles, speeches, etc., to facilitate the model to generate targeted content.

[0049] Style and Tone: Users can specify the style (e.g., humorous, serious, motivational) and tone (e.g., colloquial, formal) of the copy to ensure it fits the brand or personal style.

[0050] Constraints: Set specific requirements for the copy, such as opening design (such as "golden 3 seconds"), word limit, frequency of use of specific vocabulary, etc., to enhance the practicality and compliance of the copy.

[0051] User perspective: Allows users to input core perspectives or key points to ensure that generated content is consistent with user opinions. Figure 1 To.

[0052] Reference Cases: Users can upload or link to excellent cases as reference templates when generating new copy, thereby improving content quality and creativity.

[0053] In some embodiments of the present application, content generation options include structural content generation options and personalized content generation options, wherein the structural content generation options reflect the overall framework and direction of copy content generation and determine the basic structure and style of the copy content; the personalized content generation options reflect the specific needs and preferences of users and emphasize personalization and differentiation.

[0054] Continuing with the previous example, in the insurance scenario, the content generation options are topic selection, copy type, style and tone, constraints, user perspective, and reference cases. Of these, topic selection, copy type, style and tone, and constraints reflect the basic structure and style of the copy content. Therefore, topic selection, copy type, style and tone, and constraints are structural content generation options. User perspective and reference cases reflect the specific needs and preferences of users. Therefore, they are personalized content generation options.

[0055] In an optional implementation of this embodiment, the scenario parameter is a scenario tag. The scenario tag is used to define the application scenario of the content to be generated. It provides the model with the overall direction and style of the generated content. Specifically, the scenario tag of a content can be a recommendation, product review, knowledge sharing, etc.

[0056] In actual implementation, a variety of methods can be used to determine the scene labels corresponding to the target content generation options, and this application does not impose any restrictions on this. For example, a rule-based keyword matching method, specifically, through the mapping relationship between predefined keywords and scene labels, quickly matches the target content generation options with the corresponding scene labels. For another example, a text classification method based on machine learning, specifically, through a pre-trained language model, semantically understands the target content generation options, thereby automatically assigning scene labels.

[0057] The following uses the rule-based keyword matching method as an example to illustrate the specific process of determining scenario parameters based on target content generation options, when the content to be generated is copywriting and the scenario parameters are scenario tags:

[0058] Build a mapping table between keywords and scene tags: collect commonly used keywords for each scene tag.

[0059] For example:

[0060] When the scene label is a recommendation, commonly used keywords include: "recommended", "easy to use", "must buy", etc.

[0061] When the scenario label is product review, commonly used keywords include: "review", "evaluation", "trial", etc.

[0062] When the scenario label is knowledge sharing, commonly used keywords include: "tutorial", "popular science", "guide", etc.

[0063] Extract keywords from user input: Perform keyword extraction on target content generation options.

[0064] Matching scene tags: Match the extracted keywords with the keywords in the mapping table to find the most relevant scene tags.

[0065] Multi-tag processing: If multiple scene tags are matched, you can sort them according to the frequency of keyword occurrence or other methods to select the most appropriate tag.

[0066] In an optional implementation of this embodiment, the scenario parameters are scenario tags and scenario options under the scenario tags. Scenario options are scenario-related constraints refined in a specific application scenario, reflecting the user's specific preferences in that scenario. For example, in a "grass recommendation" scenario, a user may prefer to start with "real user experience" or want to emphasize "the product's unique selling point." Scenario options help the model more accurately meet the user's personalized needs.

[0067] In some embodiments of the present application, scenario parameters include structural scenario parameters and personalized scenario parameters. Structural scenario parameters reflect the overall framework and direction of content generation, such as scenario tags. Personalized scenario parameters reflect user preferences, such as the scenario options selected by a user under a specific scenario tag based on historical user behavior data.

[0068] In the case where the scene parameter is a scene tag and a scene option under the scene tag, the corresponding scene parameter is determined based on the target content generation option, including:

[0069] Through the scene recognition model, the scene labels corresponding to the target content generation options in the target industry are determined.

[0070] Obtain the historical behavior data of the target user, determine the scenario options of the target user under the scenario label based on the historical behavior data, and use the scenario label and scenario options as scenario parameters.

[0071] The target user is the user object that the content generation method is expected to serve.

[0072] In actual implementation, a variety of methods can be used to determine the scene labels corresponding to the target content generation options under the target industry through the scene recognition model, and the embodiments of the present application do not limit this. For example, the scene recognition model is a large model; prompt information is generated based on the target content generation option, and the prompt information is input into the scene recognition model to obtain the scene labels corresponding to the target content generation options. For another example, the scene recognition model is a single-label classification model based on BERT (Bidirectional Encoder Representations from Transformers). The scene recognition model encodes the target content generation options into a vector representation, and uses a fully connected softmax (normalized exponential function) classification layer to predict the most likely scene label. Among them, the training of the scene recognition model can be specifically: a batch of high-quality content generation options and corresponding scene label annotation data are used as training data for the scene recognition model, and the cross-entropy loss function is used to optimize the model so that it learns the scene categories corresponding to different content generation options. After the training is completed, the model can output a scene label with the highest probability based on the new content generation option as the final recognition result.

[0073] The following uses the insurance industry as an example, where the content generation options for the insurance industry include topic selection, copy type, style and tone, constraints, user perspectives, and reference cases. This explains the process of using a scenario recognition model to determine the scenario labels corresponding to the target content generation options in the target industry.

[0074] The user entered the following content into the content generation options under the insurance industry:

[0075] Topic selection: Introduce the experience of the brand founder and the development of the company

[0076] Copywriting type: Public account article

[0077] Style and tone: Formal, motivational

[0078] Constraints: The opening should be captivating and limited to 1,000 words.

[0079] User perspective: Emphasize persistence in entrepreneurship and the process of building from scratch

[0080] Reference case: Brand story tweets from the Get APP

[0081] Build Prompt (prompt information for large models)

[0082] You are an intelligent tag recognition assistant, whose task is to identify the most matching content scenario tags from the copywriting requirements provided by users.

[0083] The following is the user input:

[0084] Topic selection: Introduce the experience of the brand founder and the development of the company

[0085] Copywriting type: Public account article

[0086] Style and tone: Formal, motivational

[0087] Constraints: The opening should be captivating and limited to 1,000 words.

[0088] User perspective: Emphasize persistence in entrepreneurship and the process of building from scratch

[0089] Reference case: Brand story tweets from the Get APP

[0090] Please return the most appropriate scenario label and briefly explain why.

[0091] User historical behavior data refers to the various data generated during the user's interaction with the platform or system. This data can help the platform or system understand the user's interests, preferences, and behavior patterns, thereby providing users with personalized content recommendations or generation services.

[0092] Specifically, the user's historical behavior data can include the following aspects:

[0093] Click and browsing history: The images, text, or video content that a user has previously clicked on, including the time, frequency, and duration of viewing. This data can reveal a user's preference for certain types of content.

[0094] Search history: Search queries conducted by users within the platform, which can reflect the topics, keywords or questions that users are interested in.

[0095] Interaction behavior: For example, whether a user likes, comments, shares, or collects certain content. These behaviors can help the system identify the types of content that users like.

[0096] Content consumption history: The type, length, and frequency of content consumed by the user. For example, a user may prefer to watch short or long videos, or may like a certain style of images or text.

[0097] In actual implementation, a variety of methods can be used to determine the scene options of the target user under the scene label, and this application does not impose any restrictions on this.

[0098] Taking the content to be generated as copywriting as an example, the user's historical behavior data is the historical behavior data generated when the user browses different copywriting based on the user's behavior, such as browsing time, whether to like, share, purchase the advertising products involved in the copywriting, reading completion, etc.

[0099] Optionally, the copy structure of each copy browsed by the target user can be labeled manually or through a rule engine (such as whether it combines text and pictures, whether it starts with a golden sentence, etc.), and then the target user's behavioral performance on different copy structures can be counted (such as reading completion, dwell time, like rate, etc.). By comparing the significant differences in the user's performance on copies with / without certain structural features, their preference for the structure can be judged to obtain scenario options. The copy can also be converted into a structural feature vector of the copy based on the copy structure, and the user's structural preference vector can be gradually constructed and updated based on the user's historical behavior on each copy. Each time a user interacts with the copy, the structural feature vector of the copy is multiplied by the user's behavior weight and added to the user's preference vector, thereby quantifying the user's preference for each structural feature of the copy and obtaining scenario options.

[0100] For example, taking the content to be generated as a copy as an example, the specific process of determining the scenario options of the target user under the scenario tag based on historical behavior data is explained.

[0101] First, construct the text structure feature vector of each text that the user has browsed.

[0102] The structural features of each copy are converted into a vector, which represents the structural elements of the copy. Suppose the structural feature vector of a copy is:

[0103] Structural feature vector [1, 0, 1, 0]

[0104] The four elements of this vector represent:

[0105] The first element "1": whether the copy has an attractive opening (1 means yes, 0 means no)

[0106] The second element "0": whether the copy is combined with text and pictures (1 means yes, 0 means no)

[0107] The third element "1" |: Whether the copy uses a list structure (1 means yes, 0 means no)

[0108] The fourth element "0": whether the copy contains obvious emotional colors or emotional language, such as motivational, inflammatory, exclamatory sentences, strong emotional expressions, etc. (1 means yes, 0 means no).

[0109] Therefore, [1, 0, 1, 0] means that the copy has an attractive beginning and list structure, but lacks text and image combination, emotionally charged language, and emotional language.

[0110] Next, determine the user's behavioral weight for each copy.

[0111] The user behavior weight is calculated based on the user's interaction data on the copy, with the goal of quantifying the user's interest and engagement in a particular copy. Specifically, this weight reflects the intensity of the user's interaction with the copy, typically taking into account user behavior such as dwell time, likes, comments, and shares.

[0112] Exemplarily, a first weight is calculated for the dwell time, where the first weight = user dwell time / average dwell time of the copy.

[0113] Dwell time reflects the user's attention to the content of the copy. The longer the dwell time, the greater the user's interest in the copy, and the higher the weight can be given.

[0114] The second weight is calculated based on the user's likes, comments, and sharing behaviors.

[0115] User interactions such as likes, comments, and shares can be another indicator of user engagement. Each action can be assigned a different weight based on its influence. For example, a like might have a weight of 1, a comment 2, and a share 3.

[0116] For example: suppose a user likes a copy and comments on it once, and the weights of each behavior are: like: 1, comment: 2, share: 3, so the user's second weight for this copy can be: second weight = 1 (like) + 2 (comment) + 0 (share) = 3.

[0117] Calculate the comprehensive weight.

[0118] Multiple behavior data can be combined to calculate a comprehensive user behavior weight. The comprehensive weight takes into account the relative importance of different behaviors.

[0119] For example, the weight of the dwell time may be 50%, the weight of sharing may be 20%, the weight of commenting may be 20%, and the weight of likes may be 10%.

[0120] Comprehensive weight = stay time weight × 0.5 + like weight × 0.1 + comment weight × 0.2 + share weight × 0.2.

[0121] Next, update the user's structural preference vector.

[0122] The user's structural preference vector represents the user's preference for different copywriting structural features. Assume that initially, the user's structural preference vector is [0, 0, 0, 0] (indicating that the user has no preference for each structural feature of the copywriting, i.e., there is not enough historical behavior data to determine their preference).

[0123] When a user interacts with a piece of copy, we combine the copy's structural feature vector with the user's behavior weight to update the user's structural preference vector.

[0124] The update process is: for each copy, multiply the copy's structural feature vector by the copy's behavioral weight and add it to the user's structural preference vector. The specific steps are as follows:

[0125] User's structural preference vector = the last obtained user's structural preference vector + (structural feature vector × behavior weight

[0126] For example, if the user's behavioral weight for this copy is 0.8, then:

[0127] The updated user's structural preference vector = [0, 0, 0, 0] + [1, 0, 1, 0] × 0.8 = [0.8, 0, 0.8, 0]

[0128] If the user continues to interact with other copywriting, each interaction will update the user's structural preference vector based on the copywriting's structural features and behavioral weights.

[0129] For example, if the user's structural feature vector for another copy is [0, 1, 1, 0] and its behavior weight is 0.5, the update process is as follows:

[0130] The updated user's structural preference vector = [0.8, 0, 0.8, 0] + [0, 0.5, 0.5, 0] = [0.8, 0.5, 1.3, 0].

[0131] In this way, the user's structural preference vector is obtained through the above solution. According to the value of each dimension in the user's structural preference vector, the user's copy structure preference can be determined to obtain scenario options.

[0132] In the embodiment of this specification, first, based on the target content generation option, the scene recognition model is used to determine the scene label corresponding to the target content generation option under the target industry, and a parameter, the scene label, is automatically expanded from the target content generation option. In this way, when the target content is subsequently generated, the model can quickly understand the core purpose of the copywriting content and ensure that the generated content is consistent with the expected scene. At the same time, by adding the scene label, when the prompt word template is subsequently matched from the database, the target prompt word template can be matched more quickly, thereby reducing the time of content generation and improving the efficiency of content generation. Then, the historical behavior data of the target user is obtained, and the scene options of the target user under the scene label are determined based on the historical behavior data, and the scene label and the scene option are used as scene parameters. In this way, the scene options of the user under the scene label are determined based on the user's historical behavior data. Since the scene options are determined based on the target user's historical behavior data, the scene options can reflect the user's specific preferences. Subsequently, based on the scene label and the more refined scene options that reflect the user's preferences, the target content generated can be more in line with the user's real needs and improve the accuracy of the generated content.

[0133] Step 204: Based on the target content generation options and scenario parameters, a corresponding target prompt word template is matched in the prompt word database.

[0134] The prompt word database is a structured storage system that centrally manages and maintains a large number of prompt word templates used to guide model output. These templates are typically categorized and labeled based on different industries, content generation options (such as topic selection, copywriting style, and copywriting type), and scenario tags (such as product promotion, science education, and brand storytelling). The goal is to support personalized, high-quality content generation.

[0135] In one possible implementation, the target content generation options and scenario parameters are matched to corresponding prompt word templates using a predefined mapping relationship between keywords and prompt word templates to obtain the corresponding target prompt word template. Specifically, keywords are first extracted from the target content generation options and scenario parameters. Then, for each keyword, a prompt word template matching that keyword is obtained from a prompt word database. Next, the identifiers of all appearing prompt word templates are combined, and the number of hits for each prompt word template is counted. Finally, the prompt word templates are sorted based on the number of hits, and the prompt word templates with the highest number of hits are selected as the target prompt word templates.

[0136] In an optional implementation of this embodiment, the target content generation options include a structural content generation option and a personalized content generation option, and the scenario parameters include a structural scenario parameter and a personalized scenario parameter. Based on the target content generation options and the scenario parameters, a corresponding target prompt word template is matched in the prompt word database, specifically including:

[0137] Based on the structural content generation options and the structural scenario parameters, a corresponding plurality of candidate prompt word templates are matched in a prompt word database.

[0138] Based on personalized content generation options and / or personalized scenario parameters, as well as multiple candidate prompt word templates, a target prompt word template is obtained through a template evaluation model.

[0139] It should be understood that because both the structural content generation options and the structural scenario parameters reflect the overall framework and direction of copywriting content generation, they can be matched against the prompt word database to obtain multiple candidate prompt word templates. Subsequently, the personalized content generation options and / or personalized scenario parameters, along with the multiple candidate prompt word templates, are input into a template evaluation model. The template evaluation model determines the target prompt word template based on the degree of match between the multiple candidate prompt word templates and the personalized content generation options, and / or the degree of match between the multiple candidate prompt word templates and the personalized scenario parameters.

[0140] In actual implementation, various methods can be used to match multiple corresponding candidate prompt word templates in the prompt word database based on the structural content generation options and structural scenario parameters, and this application does not impose any restrictions on this. For example, the matching degree of each template with the structural content generation options and structural scenario parameters can be calculated, and multiple candidate prompt word templates can be determined based on the matching degree. For another example, the structural content generation options and structural scenario parameters can be quickly matched with the corresponding prompt word templates through the mapping relationship between predefined keywords and prompt word templates.

[0141] The following uses a rule-based keyword matching method as an example to illustrate the specific process of matching multiple candidate prompt word templates in a prompt word database based on structural content generation options and structural scenario parameters, when the content to be generated is copywriting:

[0142] First, structural content generation options and structural scenario parameters are broken down into keywords.

[0143] For example, the structured content generation options selected by the user are:

[0144] Copywriting topic: Promotion of unemployment insurance

[0145] Copywriting type: Public account article

[0146] Style and tone: humorous, colloquial

[0147] The structural scenario parameters (scenario labels) generated according to the structural content generation options selected by the user are: "grass planting recommendation".

[0148] Extracted keywords: promotion of unemployment insurance, public account articles, humor, colloquialism, and recommendations.

[0149] Next, for each keyword, a prompt word template matching the keyword is obtained from the prompt word database.

[0150] Next, the identifiers of all the prompt word templates that appear are integrated, and the number of times each prompt word template is hit is counted.

[0151] Finally, the prompt word templates are sorted according to the number of times each prompt word template is hit, and the prompt word templates with the largest number of hits are selected as candidate prompt word templates.

[0152] In actual implementation, based on personalized content generation options and / or personalized scenario parameters, as well as multiple candidate prompt word templates, a variety of methods can be used to obtain a target prompt word template through a template evaluation model, and this application does not impose any restrictions on this. For example, through the template evaluation model, the personalized content generation options and / or personalized scenario parameters can be encoded into a user semantic vector, each of the multiple candidate prompt word templates can be encoded into a template semantic vector, and the similarity between the user semantic vector and each template semantic vector can be calculated. The multiple candidate prompt word templates can then be sorted based on the similarity, and the candidate prompt word template or templates with the highest similarity can be determined as the target prompt word template. For another example, through the template evaluation model, each candidate prompt word template is marked with a structured label (such as the copy theme, copy type, tone and style, core ideas, reference cases, etc.), and the personalized content generation options (core ideas, reference cases) and / or personalized scenario parameters (scenario options) are parsed into corresponding keyword labels; based on the structured labels of the candidate prompt word templates, as well as the keyword labels corresponding to the personalized content generation options and / or personalized scenario parameters, the matching degree of the personalized content generation options and / or personalized scenario parameters of each candidate prompt word template is counted, and finally, one or several candidate prompt word templates with the highest matching degree are determined as the target prompt word templates.

[0153] In the embodiments of this specification, first, based on the structured content generation options and structured scenario parameters, a plurality of corresponding candidate prompt word templates are matched in the prompt word database. Then, based on the personalized content generation options and / or personalized scenario parameters, as well as the plurality of candidate prompt word templates, a target prompt word template is obtained through a template evaluation model. Thus, the prompt word templates in the database are initially screened using the structured content generation options and structured scenario parameters. On the one hand, since the structured parameters (structured content generation options and structured scenario parameters) generally have clear classification criteria, it is convenient for the system to quickly match relevant templates and narrow the candidate range. On the other hand, the templates screened using the structured parameters can meet basic content and format requirements, ensuring that the generated content generally meets expectations. The templates obtained after the initial screening are then further screened using the personalized content generation options and / or personalized scenario parameters. The personalized parameters (personalized content generation options and personalized scenario parameters) can guide the model to select templates that better meet user needs. Using these templates to generate target content can enhance the relevance of the content to user needs and improve the accuracy of the generated content. Personalized screening allows the system to generate content of diverse styles based on the needs of different users, meeting a wide range of application scenarios. In addition, the screening strategy can be dynamically adjusted through parameters at different levels to adapt to changing user needs and application scenarios.

[0154] In an optional implementation of this embodiment, based on the structural content generation option and the structural scenario parameter, at least one prompt word template is matched in the prompt word database, and the at least one prompt word template is directly used as the target prompt word template.

[0155] Step 206: Generate target prompt information based on the target content generation options, scenario parameters, and target prompt word template, and generate target content under the target industry through the content generation model based on the target prompt information.

[0156] Fill the target content generation options and scenario parameters into the target prompt word template to construct complete target prompt information, and then input the target prompt information into the content generation model. Through the content generation model, the target content under the target industry is obtained.

[0157] In one possible implementation, at least one target prompt word template is provided. Accordingly, based on the target content generation options, scenario parameters, and the target prompt word template, at least one target prompt message is generated. Correspondingly, based on the target prompt message, a content generation model is used to generate at least one target content within the target industry. Each at least one target content is sent to a display device for display, and users can select different target content to publish as needed.

[0158] In an optional implementation of this embodiment, the target prompt information is at least one, and generating target content for the target industry through a content generation model based on the target prompt information includes:

[0159] For each target prompt information, the content generation model is used to generate candidate content under the target industry corresponding to the target prompt information.

[0160] For each candidate content, based on the corresponding target content generation options and / or corresponding scenario parameters, a quality score of each candidate content is generated through a quality assessment model.

[0161] Generate target content for the target industry based on the quality score of each candidate content.

[0162] In actual implementation, for each candidate content, based on the corresponding target content generation options and / or corresponding scenario parameters, a variety of methods can be used to generate a quality score for each candidate content through a quality assessment model.

[0163] For example, for each candidate content, the quality assessment model is used to encode the corresponding target content generation options and / or scene parameters into a user semantic vector, and the candidate content is encoded into a content semantic vector; the similarity between the user semantic vector and the content semantic vector corresponding to the candidate content is calculated, and the similarity is used as the quality score of the candidate content.

[0164] For example, the quality assessment model inputs each candidate content into a larger model and initiates multiple rounds of dialogue with the larger model, resulting in multiple dialogue results. The prompts for each round of dialogue are generated based on the target content generation and / or scenario parameters. The quality score for each candidate content is determined based on the similarity between each of these dialogue results and the target content generation options and / or scenario parameters.

[0165] Taking the insurance industry as the target industry, the content to be generated as copywriting, and the content generation options as topic selection, copywriting type, style and tone, constraints, user opinions, and reference cases as an example, the above-mentioned multi-round dialogue method is used to illustrate the specific process of generating the quality score of each candidate content through the quality assessment model based on the corresponding target content generation options and / or corresponding scenario parameters.

[0166] The targeted content generation options are:

[0167] Theme selection: Environmental protection cup promotion

[0168] Copywriting type: Short posts on social platforms

[0169] Style and tone: humorous, light-hearted

[0170] Core idea: Environmental protection can also be fashionable

[0171] Constraints: No more than 50 words

[0172] Reference case: None

[0173] The candidate content generated based on the above content generation method is: "Eco-friendly cups can also be prestigious! This cup is used by the fashion circle."

[0174] First, for each input dimension in the target content generation option, design a reverse question and ask the model:

[0175] “What is the theme of this copy?”

[0176] “What type of copywriting does this piece of content resemble?”

[0177] “What is the style and tone of this text?”

[0178] “What is the main point or position conveyed by the author?”

[0179] “Is this content longer than 50 words?”

[0180] Based on the above multiple rounds of dialogue, the dialogue results of each dialogue are obtained:

[0181] Theme: Promotion of Eco-friendly Cup

[0182] Copy type: More like the beginning of a short video

[0183] Style and tone: Light-hearted and humorous

[0184] Core point: Environmental protection can also be fashionable

[0185] Word count: 32

[0186] Next, for each conversation result, the similarity between each conversation result and the corresponding option in the target content generation options is determined. If the similarity is greater than a preset threshold, it is considered a "hit." If the similarity is less than the preset threshold, it is considered a "miss." Hits and misses are assigned different scores: specifically, a hit is scored as 1, and a miss is scored as 0.

[0187] Next, the comprehensive scores of the candidate contents are calculated.

[0188] Continuing with the previous example, Topic: Hit, Score: 1; Copy Type: Hit, Score: 1; Style and Tone: Hit, Score: 1; Core Idea: Hit, Score: 1; Constraints: Hit, Score: 1. The overall score is 5.

[0189] Finally, the comprehensive score of the candidate content is used as the quality score of the candidate content.

[0190] In the embodiments of this specification, for each target prompt, a content generation model is first used to generate candidate content for the target industry corresponding to the target prompt. For each candidate content, a quality score is generated based on the corresponding target content generation options and / or corresponding scenario parameters using a quality assessment model. Based on the quality scores of each candidate content, target content for the target industry is generated. In this way, the quality assessment model automatically scores the generated copywriting content, and the target content is determined based on the scores, thereby ensuring the high quality of the output content.

[0191] The embodiments of this specification do not limit the specific method of generating target content for the target industry based on the quality score of each candidate content.

[0192] In an optional implementation of this embodiment, the candidate content with the highest quality score among the candidate content is used as the target content for the target industry.

[0193] In an optional implementation of this embodiment, the target prompt word template includes at least two templates, corresponding to at least two prompt messages and at least two candidate contents. Among the at least two candidate contents, the candidate content with the top N quality scores is selected as the target content for the target industry. Users can select different target content to publish as needed, where N is a positive integer greater than 1.

[0194] In an optional implementation of this embodiment, generating target content for a target industry based on the quality score of each candidate content includes:

[0195] When the quality score of the first candidate content among the candidate contents is greater than the score threshold, the first candidate content is used as the target content under the target industry;

[0196] When the quality scores of the candidate contents are all less than the score threshold, the process returns to executing the step of generating target content under the target industry through the content generation model based on the target prompt information.

[0197] The first candidate content is any one or more candidate contents among the candidate contents.

[0198] It should be understood that the quality score of the candidate content reflects the degree of match between the candidate content and user needs. The higher the quality score, the more the candidate content matches the user needs, and the higher the accuracy of the candidate content. The score threshold is a preset minimum acceptable quality score used to screen qualified copy. When the quality score of the first candidate content is greater than the score threshold, it means that the quality of the first candidate content is qualified and can be output to the user as the target content. When the quality score of the first candidate content is greater than the score threshold, it means that the quality of the first candidate content is unqualified and cannot be directly output to the user as the target content. It is necessary to regenerate at least one candidate content based on the target prompt information through the content generation model according to the fallback mechanism until there is a candidate content among the candidate contents whose quality score is greater than the score threshold.

[0199] If there are multiple first candidate contents, the candidate content with the highest quality score can be used as the target content under the target industry.

[0200] In the embodiment of this specification, by setting a score threshold, the quality of each generated content is controlled. When the quality score of each candidate content is less than the score threshold, the content is regenerated, thereby achieving further optimization of the copy content and further ensuring the accuracy of the generated content.

[0201] In an optional implementation of this embodiment, the content generation method further includes:

[0202] Obtain historical behavior data corresponding to multiple reference users.

[0203] Based on the historical behavior data corresponding to the multiple reference users, scenario options of the multiple reference users under the scenario label are determined.

[0204] When the scenario options of multiple reference users under the scenario tags respectively meet preset conditions, the content generation options under the target industry are updated based on the scenario options of the multiple reference users under the scenario tags respectively.

[0205] The multiple reference users may include the target user or may not include the target user.

[0206] Among them, updating the content generation options under the target industry may include deleting the original content generation options, or adding new content generation options based on the original content generation options.

[0207] This application description does not limit the specific content of the preset conditions.

[0208] For example, the preset condition is that the scenario options corresponding to multiple reference users are consistent. Then, when the scenario options of the multiple reference users under the scenario tags respectively meet the preset condition, based on the scenario options of the multiple reference users under the scenario tags respectively, updating the content generation options under the target industry specifically includes:

[0209] When the scenario options of multiple reference users under the scenario tags are consistent, the content generation options under the target industry are updated based on the scenario options.

[0210] It's important to note that a user's historical behavior data can reflect their preferences. Therefore, if the scenario options generated based on the historical behavior data of multiple users are consistent, it indicates that their preferences are relatively consistent. For example, if the scenario options generated based on the historical behavior data of multiple users all indicate that the user likes to include relevant images in the copy, then an "image" option can be added to the content generation options to make the content generation options more tailored to user needs.

[0211] For example, the preset condition is that the scene options corresponding to M reference users among the multiple reference users are consistent, where M is a positive integer greater than 1. Then, if the scene options of the multiple reference users under the scene tags meet the preset condition, the content generation options under the target industry are updated based on the scene options of the multiple reference users under the scene tags, specifically including:

[0212] When the scenario options corresponding to M reference users among multiple reference users are consistent, the content generation options under the target industry are updated based on the scenario options.

[0213] In the embodiments of this specification, when the behavioral data of multiple reference users form consistency or trends in a certain direction (i.e., meet certain "preset conditions", such as quantity, concentration, and weight thresholds), the system will automatically adjust and optimize the content generation options under the target industry accordingly, so that it is closer to user needs. In this way, this solution drives the update of content generation strategies through a large number of real user behaviors, so that the generated content is more in line with the actual preferences and needs of specific industries or usage scenarios. This solution avoids static content generation options, uses the historical behavioral data of many users to update generation options, adapts to changes in user preferences and market trends, and has good adaptive capabilities.

[0214] In an optional implementation of this embodiment, the content generation method further includes: feeding back the target content of the target industry to a display device for display, and the format of the target content is adapted to the display format of different display devices.

[0215] Specifically, after a user logs in through a display device, the display device sends a content display request to the server based on the user's operation. This request includes the display device type information. Upon receiving the request, the server matches and selects the target display format corresponding to the display device type information from a variety of preset content display formats. The server then sends the target display format content to the display device for presentation.

[0216] In the embodiments of the present specification, the format of the generated target content can be adapted to different display devices, thereby enhancing the applicability of the copy content in a multi-platform environment and simplifying the process of publishing the copy content across platforms.

[0217] In an optional implementation of this embodiment, the above-mentioned content generation method further includes: obtaining feedback information from the display device regarding the target content; and updating at least one of the content generation options, scene parameters, and content generation model for the target industry based on the feedback information.

[0218] After generating the target content, the server receives feedback from display devices, either from users, system evaluation modules, or third-party evaluation mechanisms. This feedback may include content quality signals such as incomplete content presentation, offset scene parameters, and unclear content presentation. The server structures the feedback, analyzes the issues it identifies, and dynamically adjusts one or more of the following content generation components accordingly: content generation options, scene parameters, and content generation model.

[0219] For example, if the feedback information shows that the current content cannot be combined with the relevant image, an "image" option is added to the content generation options, and the user can enter the image he or she wants to add to the content to be generated in the "image" option.

[0220] If the feedback information shows that the scene parameters adapted by the content are incorrect, the system will automatically correct the scene parameters based on the feedback.

[0221] If the feedback information shows the semantic deviation of the current content, the system can fine-tune the model parameters or retrain the content generation model to improve subsequent output effects.

[0222] Ultimately, the system will regenerate the target content under the target industry based on the updated parameters to better meet user needs.

[0223] In the embodiments of this specification, feedback data helps to determine the effectiveness of current content generation options in specific industry scenarios, thereby adjusting or optimizing these options to make them more in line with the actual usage needs and preferences of industry users. Correlating scenario parameters with user feedback information can help the system refine the content requirements of various scenarios, so that the generated content is closer to the actual usage context. When feedback information reflects that certain generated content is not effective under specific conditions, the system can use this data to fine-tune or train the underlying content generation model to enhance the model's generalization ability and expressiveness for specific industries, scenarios or devices. In short, by continuously optimizing content quality and presentation effects, the accuracy of generated content is improved, so that users can get a consistent and high-quality experience on different devices.

[0224] In an optional implementation of the embodiment of the present application, an interactive editing function is also provided on the display device side. In addition to displaying the target content, an online editing function is also provided to allow users to fine-tune the generated copy content.

[0225] Corresponding to the above method embodiment, this specification also provides a content generation device embodiment, Figure 4 FIG. 1 shows a schematic diagram of the structure of a content generation device provided by an embodiment of this specification. Figure 4 As shown, the device includes:

[0226] Determination module 402 is configured to obtain a target content generation option corresponding to the content generation task, and determine corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry;

[0227] A matching module 404 is configured to match a corresponding target prompt word template in a prompt word database based on target content generation options and scenario parameters;

[0228] The generation module 406 is configured to generate target prompt information based on the target content generation options, scenario parameters and target prompt word template, and generate target content under the target industry through the content generation model based on the target prompt information.

[0229] Optionally, the determination module 402 is further configured to:

[0230] Determine the scenario labels corresponding to the target content generation options in the target industry through the scenario recognition model;

[0231] Obtain the historical behavior data of the target user, determine the scenario options of the target user under the scenario label based on the historical behavior data, and use the scenario label and scenario options as scenario parameters.

[0232] Optionally, the target content generation option includes a structural content generation option and a personalized content generation option, the scene parameters include a structural scene parameter and a personalized scene parameter, and the matching module 404 is further configured to:

[0233] Based on the structural content generation options and the structural scenario parameters, matching corresponding multiple candidate prompt word templates in the prompt word database;

[0234] Based on personalized content generation options and / or personalized scenario parameters, as well as multiple candidate prompt word templates, a target prompt word template is obtained through a template evaluation model.

[0235] Optionally, the target prompt information is at least one, and the generating module 406 is further configured to:

[0236] For each target prompt information, generate candidate content under the target industry corresponding to the target prompt information through the content generation model;

[0237] For each candidate content, based on the corresponding target content generation options and / or corresponding scenario parameters, a quality score for each candidate content is generated using a quality assessment model;

[0238] Generate target content for the target industry based on the quality score of each candidate content.

[0239] Optionally, the generating module 406 is further configured to:

[0240] When the quality score of the first candidate content among the candidate contents is greater than the score threshold, the first candidate content is used as the target content under the target industry;

[0241] When the quality scores of the candidate contents are all less than the score threshold, the process returns to executing the step of generating target content under the target industry through the content generation model based on the target prompt information.

[0242] Optionally, the device further includes an update module configured to:

[0243] Obtain historical behavior data corresponding to multiple reference users;

[0244] Determining scenario options for the multiple reference users under scenario tags based on historical behavior data corresponding to the multiple reference users;

[0245] When the scenario options of multiple reference users under the scenario tags respectively meet preset conditions, the content generation options under the target industry are updated based on the scenario options of the multiple reference users under the scenario tags respectively.

[0246] Optionally, the device further includes a first feedback module configured to:

[0247] The target content under the target industry is fed back to the display device for display, and the format of the target content is adapted to the display format of different display devices.

[0248] Optionally, the device further includes a second feedback module configured to:

[0249] Obtaining feedback information from the display device regarding the target content;

[0250] Based on the feedback information, at least one of the content generation options, scenario parameters, and content generation models for the target industry is updated.

[0251] The above is a schematic diagram of a content generation device according to this embodiment. It should be noted that the technical solution of the content generation device and the technical solution of the content generation method described above are based on the same concept. For details not described in detail in the technical solution of the content generation device, please refer to the description of the technical solution of the content generation method described above.

[0252] Figure 5 The block diagram of a computing device 500 according to one embodiment of the present disclosure is shown. Components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. The processor 520 is connected to the memory 510 via a bus 530, and a database 550 is used to store data.

[0253] The computing device 500 also includes an access device 540 that enables the computing device 500 to communicate via one or more networks 560. Examples of such networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0254] In one embodiment of the present specification, the above components of the computing device 500 and Figure 5 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 5 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0255] Computing device 500 may 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 computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer (PC). Computing device 500 may also be a mobile or stationary server.

[0256] The processor 520 is configured to execute the following computer-executable instructions, which implement the steps of the above-mentioned content generation method when executed by the processor.

[0257] The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of the computing device and the technical solution of the above-mentioned content generation method are based on the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the above-mentioned content generation method.

[0258] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which implement the steps of the above-mentioned content generation method when executed by a processor.

[0259] The above is a schematic diagram of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of the storage medium and the technical solution of the content generation method described above are based on the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the content generation method described above.

[0260] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned content generation method.

[0261] The above is an illustrative solution of a computer program of this embodiment. It should be noted that the technical solution of the computer program and the technical solution of the above-mentioned content generation method are based on the same concept. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the above-mentioned content generation method.

[0262] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0263] The computer instructions include computer program code, which may be in source code form, object code form, executable file, or some intermediate form. 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 disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained 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, computer-readable media does not include electric carrier signals and telecommunication signals.

[0264] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, 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 this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0265] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0266] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A content generation method, characterized in that: include: Obtaining a target content generation option corresponding to the content generation task, and determining corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry; Based on the target content generation option and the scenario parameters, matching a corresponding target prompt word template in a prompt word database; Based on the target content generation option, the scenario parameters and the target prompt word template, target prompt information is generated, and based on the target prompt information, target content under the target industry is generated through a content generation model.

2. The content generation method according to claim 1, characterized in that: The determining of corresponding scene parameters based on the target content generation option includes: Determine, by a scene recognition model, a scene label corresponding to the target content generation option under the target industry; The historical behavior data of the target user is obtained, and the scene options of the target user under the scene label are determined based on the historical behavior data, and the scene label and the scene options are used as the scene parameters.

3. The content generation method according to claim 1, characterized in that: The target content generation options include a structural content generation option and a personalized content generation option, the scenario parameters include a structural scenario parameter and a personalized scenario parameter, and matching a corresponding target prompt word template in a prompt word database based on the target content generation options and the scenario parameters includes: Based on the structural content generation option and the structural scene parameter, matching a plurality of corresponding candidate prompt word templates in the prompt word database; Based on the personalized content generation option and / or the personalized scenario parameter, and the multiple candidate prompt word templates, the target prompt word template is obtained through a template evaluation model.

4. The content generation method according to claim 1, wherein: The target prompt information is at least one, and generating target content under the target industry through a content generation model based on the target prompt information includes: For each target prompt information, generating candidate content under the target industry corresponding to the target prompt information by using the content generation model; For each candidate content, generating a quality score for the candidate content based on the corresponding target content generation option and / or the corresponding scenario parameters using a quality assessment model; Based on the quality scores of the candidate contents, target content for the target industry is generated.

5. The content generation method according to claim 4, characterized in that: Generating target content for the target industry based on the quality scores of the candidate content includes: If the quality score of a first candidate content among the candidate contents is greater than a score threshold, use the first candidate content as the target content under the target industry; When the quality scores of the candidate contents are all less than the score threshold, the process returns to executing the step of generating the target content under the target industry through the content generation model based on the target prompt information.

6. The content generation method according to claim 2, characterized in that: The method further comprises: Obtain historical behavior data corresponding to multiple reference users; Determining, based on the historical behavior data corresponding to the multiple reference users, the scenario options of the multiple reference users under the scenario tags; In a case where the scenario options of the multiple reference users respectively under the scenario tags meet a preset condition, the content generation options under the target industry are updated based on the scenario options of the multiple reference users respectively under the scenario tags.

7. The content generation method according to any one of claims 1 to 6, characterized in that: The method further comprises: The target content under the target industry is fed back to a display device for display, and the format of the target content is adapted to the display format of different display devices.

8. The content generation method according to claim 7, characterized in that: The method further comprises: Acquiring feedback information of the display device with respect to the target content; Based on the feedback information, at least one of the content generation options for the target industry, the scenario parameters, and the content generation model is updated.

9. A content generating device, characterized in that: include: a determination module configured to obtain a target content generation option corresponding to the content generation task, and determine corresponding scenario parameters based on the target content generation option, wherein the target content generation option is a currently selected content generation option among the content generation options of at least one dimension under the target industry; a matching module configured to match a corresponding target prompt word template in a prompt word database based on the target content generation option and the scenario parameter; The generation module is configured to generate target prompt information based on the target content generation option, the scenario parameters and the target prompt word template, and generate target content under the target industry through a content generation model based on the target prompt information.

10. A computing device, characterized in that include: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the content generation method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium, characterized in that It stores computer-executable instructions, which, when executed by a processor, implement the steps of the content generation method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the content generation method according to any one of claims 1 to 8.

Images