Content generation methods, devices, electronic equipment and storage media

Abstract

This invention relates to the field of data processing technology, and discloses a content generation method, apparatus, electronic device, and storage medium. Responding to a content generation request for a target product, the invention sequentially parses the acquired product information of the target product based on the execution states and cascading relationships of multiple preset analytical agents, obtaining multiple product features of the target product. The product features output by the preset analytical agent preceding the cascading relationship serve as input to the preset analytical agent following the cascading relationship. Based on these multiple product features, recommended content information for the target product is generated. This ensures consistency among the product features output by each preset analytical agent, improves the generation effect of recommended content information, and enhances the stability and efficiency of the overall generation task.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to content generation methods, apparatus, electronic devices, and storage media. Background Technology

[0002] Currently, large-scale model proxies are typically used to generate multimodal recommendation content by constructing a long-chain task of "competitive analysis - visual generation - text output". However, this method adopts a single-stream dialogue mode in context management, which is prone to illusions or inconsistencies between stages, such as inconsistent styles between generated scene images and detail images. Furthermore, because the entire generation process lacks a breakpoint resume mechanism, if an anomaly occurs in any stage, the entire generation task needs to be restarted. As a result, the generation of multimodal recommendation content is poor in quality, unstable, and inefficient. Summary of the Invention

[0003] This invention provides a content generation method, apparatus, electronic device, and storage medium to solve the problems of poor generation effect, low stability, and low efficiency of multimodal recommendation content information.

[0004] In a first aspect, the present invention provides a content generation method, the method comprising: In response to the content generation request for the target product, based on the execution status and cascading relationship of multiple preset analysis agents, the product information of the target product is parsed sequentially to obtain multiple product features of the target product. Among them, the product features output by the preset analysis agent located before the cascading relationship serve as the input of the preset analysis agent located after the cascading relationship. Based on multiple product features, recommended content information for the target product is generated.

[0005] In one optional implementation, based on the execution states and cascading relationships of multiple preset analytical agents, the acquired product information of the target product is sequentially parsed to obtain multiple product features of the target product, including: Based on the cascading relationship, the product features of multiple preset analytical agents are reverse-searched to determine the target analytical agent whose execution status is execution failure. Based on the cascading relationship, the target analysis agent is called sequentially to parse the product information and obtain multiple product features.

[0006] In one optional implementation, based on the cascading relationship of multiple preset analytical agents, a reverse lookup is performed on the product features of the multiple preset analytical agents to determine the target analytical agent whose execution status is execution failure, including: Based on the cascading relationship, in persistent storage, snapshots corresponding to the product features of multiple preset analytical agents are searched in reverse. If no snapshot corresponding to the product features of the preset analysis agent is found, or if the snapshot corresponding to the product features of the preset analysis agent is invalid, the execution status of the preset analysis agent is determined to be execution failure.

[0007] In one optional implementation, the plurality of preset analytical agents include a first analytical agent. If the first analytical agent is invoked, the product information is parsed, including: Image analysis is performed on the product image information in the product information to obtain the product attribute characteristics of the target product; By associating product attribute features with reference information in the product information, basic recommendation information for the target product is obtained. The reference information is used to characterize the product evaluation of the related products corresponding to the target product.

[0008] In one optional implementation, the plurality of preset analytical agents include a second analytical agent, which is cascaded with the first analytical agent. If the second analytical agent is invoked, the product information is parsed, including: Based on preset scene prompts, the basic recommendation information and reference information are parsed to obtain scene description information corresponding to the target area information; The scene description information is broken down into multiple scene identification information, and scene images corresponding to each scene identification information are generated.

[0009] In one optional implementation, the plurality of preset analytical agents include a third analytical agent, which is cascaded with the second analytical agent. If the third analytical agent is invoked, the product information is parsed, including: Based on product image information and multiple scene images, image consistency constraints are constructed. Based on basic recommendation information, reference information, and multiple scene images, generate detailed description text and marketing text for the target product; Based on image consistency constraints and detailed descriptive text, generate detailed feature images of the target product.

[0010] In one alternative implementation, the method further includes: Obtain product evaluation data for related products corresponding to the target product; Semantic analysis is performed on the product evaluation data of related products to obtain the product evaluation features of related products; Based on the statistical analysis and weighting of product evaluation characteristics of related products, reference information for the target product is determined.

[0011] In a second aspect, the present invention provides a content generation apparatus, the apparatus comprising: The information parsing module is used to respond to the content generation request of the target product. Based on the execution status and cascading relationship of multiple preset analysis agents, it sequentially parses the product information of the target product to obtain multiple product features of the target product. Among them, the product features output by the preset analysis agent located before the cascading relationship serve as the input of the preset analysis agent located after the cascading relationship. The content generation module is used to generate recommended content information for the target product based on multiple product features.

[0012] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the content generation method described in the first aspect or any corresponding embodiment thereof.

[0013] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the content generation method described in the first aspect or any corresponding embodiment thereof.

[0014] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the content generation method described in the first aspect or any corresponding embodiment thereof.

[0015] The content generation method provided in this invention cascades multiple preset analysis agents, with the product features output by the preset analysis agent preceding the cascade serving as input to the preset analysis agent following the cascade. This creates a structured and clearly dependent input-output relationship in the information transmission between the multiple preset analysis agents, thereby avoiding the possibility of conflicts between the product features output by different preset analysis agents, ensuring the consistency of the product features output by each preset analysis agent, and improving the generation effect of recommended content information. Simultaneously, based on the execution states and cascade relationships of the multiple preset analysis agents, the obtained product information of the target product is parsed sequentially to obtain multiple product features of the target product. Based on these multiple product features, recommended content information of the target product is generated. By introducing the execution states of the preset analysis agents, breakpoint location is achieved in the content generation process, allowing execution to resume from a specific preset analysis agent in a particular execution state, enabling rapid recovery of content generation and improving the stability and efficiency of the overall generation task. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the first type of content generation method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second process for a content generation method according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a content generation apparatus according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] Currently, large-scale model proxies are typically used to generate multimodal recommendation content by constructing a long-chain task of "competitive analysis - visual generation - text output". However, this method adopts a single-stream dialogue mode in context management, which is prone to illusions or inconsistencies between stages, such as inconsistent styles between generated scene images and detail images. Furthermore, because the entire generation process lacks a breakpoint resume mechanism, if an anomaly occurs in any stage, the entire generation task needs to be restarted. As a result, the generation of multimodal recommendation content is poor in quality, unstable, and inefficient.

[0022] To address the aforementioned issues, this invention provides a content generation method. In response to a content generation request for a target product, the method sequentially parses the acquired product information of the target product based on the execution states and cascading relationships of multiple preset analytical agents, thereby obtaining multiple product features of the target product. The product features output by the preset analytical agent preceding the cascading relationship serve as the input to the preset analytical agent following the cascading relationship. Based on these multiple product features, recommended content information for the target product is generated. Therefore, by cascading multiple pre-defined analytical agents, and using the product features output by the pre-defined analytical agent preceding the cascade as the input to the pre-defined analytical agent following the cascade, the information transmission between multiple pre-defined analytical agents has a structured and clearly dependent input-output relationship. This avoids the possibility of conflicts between the product features output by different pre-defined analytical agents, ensures the consistency of the product features output by each pre-defined analytical agent, and improves the generation effect of recommended content information. At the same time, based on the execution state and cascade relationship of multiple pre-defined analytical agents, the product information of the target product is parsed sequentially to obtain multiple product features of the target product. Based on these multiple product features, recommended content information of the target product is generated. By introducing the execution state of the pre-defined analytical agents, breakpoint location in the content generation process can be achieved, allowing execution to start from a pre-defined analytical agent in a specific execution state, enabling rapid recovery of content generation and improving the stability and efficiency of the overall generation task.

[0023] The content generation method provided in this embodiment of the invention can be deployed in the form of a software system or hardware architecture between the Web layer and the backend scheduling service. At the same time, it uses a hybrid communication mode of HTTP protocol and SSE (Server-Sent Events) long connection to communicate with the Web layer and the backend scheduling service.

[0024] According to an embodiment of the present invention, a content generation method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0025] This embodiment provides a content generation method. Figure 1 This is a schematic diagram of the first type of content generation method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: In response to the content generation request for the target product, based on the execution status and cascading relationship of multiple preset analytical agents, the product information of the target product is parsed in sequence to obtain multiple product features of the target product.

[0026] In this embodiment of the invention, the system includes multiple preset analytical agents. Each preset analytical agent is a smart agent pre-configured with specific analytical functions, used to perform product information parsing tasks for a corresponding stage, such as product feature recognition, scene image generation, and copywriting strategy output. These preset analytical agents are arranged in series according to a preset cascading relationship, forming an ordered dependency chain. In this chain, the product features output by the preset analytical agent preceding the cascading agent serve as the input for the preset analytical agent following the cascading agent. This creates an execution sequence with clear sequential dependencies among the multiple preset analytical agents, achieving directed information transmission and avoiding the inconsistency issues between stages in a single-stream dialogue mode.

[0027] In this embodiment of the invention, the target product is the product for which content generation is required. It can be any product in any marketing environment, such as portable energy storage power supplies, smart home devices, electronic products, etc. The product information of the target product includes, but is not limited to, product image information, product description text, target market identification, product specifications, and other product-related information that can comprehensively cover the target product from multiple dimensions.

[0028] In this embodiment of the invention, product features are structured data output by a pre-defined analytical agent after performing a task of parsing product information of the target product. Specifically, these features can be structured objects in JSON format to facilitate data transfer and parsing between different pre-defined analytical agents. Optionally, product features may include multiple features describing and marketing recommendations for the target product, such as marketing copy. The product features output by each pre-defined analytical agent correspond to the parsing results at a specific stage in the content generation chain and provide a data input basis for subsequent stages of the pre-defined analytical agent.

[0029] In this embodiment of the invention, the content generation request is triggered by the user through a front-end interface. In response to the content generation request, the product information of the target product is first obtained. Specifically, product data and target market information of the target product are parsed from the content generation request. The product data characterizes the product's appearance and parameters, while the target market information characterizes the expected market for the target product. The product data can be product image information, and the target market information can be a target market identifier. Simultaneously, the product data and target market information are associated with the session identifier corresponding to the current content generation request and stored in an extended parameter pool. Also, pre-acquired reference information associated with the target product is extracted from the database and stored in the extended parameter pool as well.

[0030] In this embodiment of the invention, after acquiring product information, the product information is parsed sequentially based on the execution status and cascading relationship of multiple preset analysis agents. The execution status indicates whether each preset analysis agent has successfully parsed the product information and generated valid product features; it includes at least two states: successful execution and failed execution. Before starting each level of preset analysis agent, the execution status of each agent is checked to determine the actual execution progress of the current content generation process, thereby avoiding repeated execution of completed stages and improving generation efficiency. After determining the execution status of each preset analysis agent, they are called sequentially from the first preset analysis agent according to the cascading relationship. When each level of agent performs the parsing task, its input includes the product features output by the previous level agent and corresponding data from the product information. Therefore, the product features output at each stage have a clear logical dependency relationship, effectively avoiding the possibility of conflicts between product features output by different agents and ensuring a high degree of logical and visual consistency in the generated content.

[0031] In one alternative implementation, reference information for the target product can be pre-built in the following manner: First, product review data for related products corresponding to the target product is obtained. Related products are competitors belonging to the same category or similar market segment as the target product. Product review data consists of user comments on related products across various platforms and social media channels. This data can be obtained by using a web crawler module, based on the related products' Uniform Resource Locators (URLs), to crawl basic data from the product detail pages and user reviews across platforms. Specific data sources can include product detail pages from various platforms, in-site search term rankings, and product-related posts and user comments from major external social media platforms. After crawling, the raw review data is cleaned, automatically removing meaningless short reviews with fewer than a preset word count threshold (e.g., three words) to eliminate noise interference with subsequent semantic analysis.

[0032] Then, semantic parsing is performed on the product evaluation data of related products to obtain the product evaluation features of related products. Specifically, a large language model is invoked to perform multi-dimensional semantic clustering analysis on the obtained product evaluation data of related products, classifying and aggregating the product evaluation data according to corresponding product dimensions. Optionally, multiple preset business dimensions are pre-set, and the product evaluation data is assigned to the corresponding preset business dimensions for classification and aggregation. These preset business dimensions may include multiple dimensions such as quality, logistics, installation, appearance, price, and after-sales service.

[0033] Finally, based on the statistical analysis and weighting of product evaluation features of related products, reference information for the target product is determined. The statistical analysis of product evaluation features involves statistically analyzing the execution frequency and polarity of product evaluation features in each preset business dimension. This includes calculating the frequency of positive and negative reviews for each preset business dimension to identify the strengths and weaknesses of related products in each dimension. For example, in the quality dimension, if multiple reviews mention "long battery life," this is aggregated as a positive evaluation feature; if multiple reviews mention "complex installation process," this is aggregated as a negative evaluation feature. The weighting of product evaluation features for related products is based on the degree of negative impact of each preset business dimension on the target product. For example, a higher weight is assigned to preset business dimensions that directly lead to product returns. Through the collaborative processing of statistical analysis and weighting, the final output includes a word cloud containing multiple preset numbers of the most popular positive reviews for related products (e.g., five positive review word clouds), and a negative review diagnosis report sorted by the severity of negative reviews. The aforementioned positive review word cloud and negative review diagnosis report together constitute reference information, which is stored in the extended parameter pool as a shared basic context.

[0034] Step S102: Based on multiple product features, generate recommended content information for the target product.

[0035] In this embodiment of the invention, after obtaining multiple product features, these features are aggregated to generate complete recommendation content information for the target product. Specifically, the product features output by each level of the pre-defined analytical agent are assembled according to a pre-defined recommendation content information template. This template is a pre-configured content organization structure that defines the arrangement order and hierarchical relationship of the components within the recommendation content information. The generated recommendation content information may include product title, product description, page text, scene display images, detailed feature images, and SEO keyword strategies, among other things.

[0036] In this embodiment of the invention, after content generation is complete, all recommended content information is encapsulated into a multimodal marketing report and pushed to the user's device in real time. Simultaneously, the generated recommended content information and its corresponding product feature snapshots are persistently stored in a database, and a version number is generated for subsequent version management and retrospective analysis. Users can view the complete marketing asset report through the front-end interface and retry or refactor unsatisfactory stages or features.

[0037] The content generation method provided in this invention cascades multiple preset analysis agents, with the product features output by the preset analysis agent preceding the cascade serving as input to the preset analysis agent following the cascade. This creates a structured and clearly dependent input-output relationship in the information transmission between the multiple preset analysis agents, thereby avoiding the possibility of conflicts between the product features output by different preset analysis agents, ensuring the consistency of the product features output by each preset analysis agent, and improving the generation effect of recommended content information. Simultaneously, based on the execution states and cascade relationships of the multiple preset analysis agents, the obtained product information of the target product is parsed sequentially to obtain multiple product features of the target product. Based on these multiple product features, recommended content information of the target product is generated. By introducing the execution states of the preset analysis agents, breakpoint location is achieved in the content generation process, allowing execution to resume from a specific preset analysis agent in a particular execution state, enabling rapid recovery of content generation and improving the stability and efficiency of the overall generation task.

[0038] This embodiment provides a content generation method. Figure 2 This is a schematic diagram of a second process for a content generation method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: In response to the content generation request for the target product, based on the execution status and cascading relationship of multiple preset analytical agents, the product information of the target product is parsed sequentially to obtain multiple product features of the target product.

[0039] Specifically, step S201 includes: Step S2011: Based on the cascading relationship, reverse lookup is performed on the product features of multiple preset analytical agents to determine the target analytical agent whose execution status is execution failure.

[0040] In this embodiment of the invention, the cascading relationship is used to characterize the ordered dependency chain formed between each preset analysis agent. For example, the cascading relationship characterizes the ordered dependency chain of the first analysis agent - the second analysis agent - the third analysis agent - the fourth analysis agent - the fifth analysis agent. According to this cascading relationship, the first analysis agent is located at the starting position of the cascading relationship. The first analysis agent is the first analysis agent to perform product information parsing. The second analysis agent is cascaded with the first analysis agent and is the second analysis agent to perform product information parsing. It performs product information parsing based on the product features parsed by the first analysis agent. The third analysis agent is cascaded with the second analysis agent and is the third analysis agent to perform product information parsing. It performs product information parsing based on the product features parsed by the second analysis agent, and so on. The execution of the analysis agent in the next stage depends on the product features output by the analysis agent in the previous stage.

[0041] In this embodiment of the invention, reverse lookup refers to searching for the product features of each preset analysis agent in reverse order of the cascading relationship, starting from the last preset analysis agent and proceeding backward along the cascading relationship. This reverse lookup allows for the precise location of the first preset analysis agent in the cascading chain to fail. Because the cascading relationship has strict dependencies, if a preset analysis agent's product feature is missing or invalid, all preset analysis agents preceding it have already been successfully executed and do not need to be repeated. This preset analysis agent and all subsequent preset analysis agents need to be re-executed. These re-executed preset analysis agents are designated as target analysis agents to re-execute their parsing tasks.

[0042] In one optional implementation, based on the cascading relationship, snapshots corresponding to product features of multiple preset analytical agents are retrieved in reverse order in persistent storage. The persistent storage is the extended parameter field (extra_params) of the session database, which stores snapshots corresponding to product features at each stage in key-value pairs. Specifically, after each preset analytical agent successfully executes, the product features output by the agent are serialized, for example, into a JSON string, and stored in the corresponding keys of the extended parameter field, such as standard_phase1_content, standard_phase2_content, and standard_phase3_content. During the search, the snapshots corresponding to the product features of each preset analytical agent can be obtained by reading the records in the extended parameter field. If no snapshot corresponding to the product feature of the preset analysis agent is found, or if the snapshot corresponding to the product feature of the preset analysis agent is invalid, the execution status of the preset analysis agent is determined to be execution failure. Specifically, if the snapshot is not found in the extended parameter field, the snapshot cannot be deserialized into the correct data structure, the timestamp of the snapshot is earlier than the preset validity period threshold, or the stage version corresponding to the snapshot is incompatible with the current version, the snapshot corresponding to the product feature of the preset analysis agent is determined to be invalid.

[0043] For example, in a cascaded link containing three pre-defined analytical agents, the system first searches for standard_phase3_content (the snapshot corresponding to the product features of the third analytical agent). If it is invalid, it continues to search for standard_phase2_content (the snapshot corresponding to the product features of the second analytical agent). If it is invalid, it continues to search for standard_phase1_content (the snapshot corresponding to the product features of the first analytical agent), until it finds a snapshot corresponding to the product features of a pre-defined analytical agent in a valid state, or finds a snapshot corresponding to the product features of the first pre-defined analytical agent in an invalid state in the cascaded relationship. The pre-defined analytical agents in the cascaded relationship from the first pre-defined analytical agent in an invalid state to the pre-defined analytical agents that follow are taken as target pre-defined analytical agents.

[0044] In step S2012, based on the cascading relationship, the target analysis agent is called sequentially to parse the product information and obtain multiple product features.

[0045] In this embodiment of the invention, after determining the target analysis agent, the cascading pointer is reset to the first target analysis agent, and each level of target analysis agent is called sequentially starting from the first target analysis agent according to the ascending order of the cascading relationship. For all preset analysis agents preceding the target analysis agent, i.e., preset analysis agents that have been successfully executed, snapshots corresponding to their product features are directly read from persistent storage for reuse, without repeatedly executing the parsing task corresponding to that preset analysis agent. Specifically, snapshots corresponding to the product features of all preset analysis agents that have been successfully executed before the target analysis agent are read from persistent storage, loaded into the context of the current session, and then, according to the cascading relationship, the first target analysis agent is called, taking the product features of the preceding agent as input, parsing the product information, and generating the product features corresponding to the first target analysis agent. After the parsing task corresponding to the first target analysis agent is successfully executed, the next level of preset analysis agents is called, until all preset analysis agents in the cascading chain have been executed. Therefore, by skipping successfully executed parsing tasks and re-executing only incomplete or failed tasks, the waste of resources caused by re-executing the entire chain due to a single point of failure is avoided, thus improving the efficiency of content generation.

[0046] Step S202: Based on multiple product features, generate recommended content information for the target product. For details, please refer to [link / reference]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0047] The content generation method provided in this invention, based on a cascading relationship, reverse-searches the product features of multiple preset analytical agents to determine the target analytical agent whose execution state is failed. Based on the cascading relationship, the target analytical agent is called sequentially to parse the product information and obtain multiple product features. This constructs a reverse search and breakpoint recovery mechanism, enabling the method to accurately locate the stage where the failure occurred and resume execution from that stage when faced with network anomalies, service interruptions, or model call timeouts. This significantly improves the stability and efficiency of content generation.

[0048] In an optional implementation, as described in the above embodiments, the multiple preset analytical agents include a first analytical agent, a second analytical agent, and a third analytical agent. The second analytical agent is cascaded with the first analytical agent, and the third analytical agent is cascaded with the second analytical agent, forming a cascaded relationship of first analytical agent - second analytical agent - third analytical agent. In this cascaded relationship, the first analytical agent is located at the beginning of the cascade and is responsible for the initial analysis of the product image information; the second analytical agent receives the output results of the first analytical agent and is responsible for scene-level analysis and image generation; the third analytical agent receives the output results of the two preceding preset analytical agents and is responsible for detail supplementation and generation of global copywriting strategies. The specific analysis process when calling each level of the preset analytical agent is described in detail below.

[0049] In one optional implementation, the first analytical agent is an intelligent proxy with visual understanding capabilities, which calls a multimodal visual language model to perform image parsing tasks. If the first analytical agent is invoked, the product information is parsed, including: Step a1: Perform image analysis on the product image information in the product information to obtain the product attribute features of the target product.

[0050] In step a1, the product image information refers to product images of the target product, which may include main product images, multi-angle detail images, and other image data that can display the product's appearance and structural features. After acquiring the product image information, the first analytical agent calls a multimodal visual language model to perform structured parsing of the product image information. Specifically, the first analytical agent takes the product image information as input, loads it into the multimodal visual language model, uses the visual encoder of the multimodal visual language model to extract visual features from the image, and uses the language decoder to convert the extracted visual features into structured text descriptions, thereby identifying and outputting the product attribute features of the target product. These product attribute features may include the target product's name, product category, material composition, physical structural features, size specifications, and color—product attribute information that can be perceived from a visual perspective.

[0051] Step a2 involves associating product features with reference information in the product information to obtain basic recommendation information for the target product.

[0052] In step a2, the reference information is used to characterize the product evaluations of related products corresponding to the target product. After obtaining the product features, the first analytical agent performs correlation analysis between the product features and the reference information. Specifically, the first analytical agent reads the reference information stored in the shared basic context, inputs the reference information and product features together into the large language model for fusion reasoning, identifies the differentiated advantages of the target product relative to related products, and outputs the marketing keywords or marketing fields corresponding to these differentiated advantages. For example, if the negative review diagnosis in the reference information shows that the main pain points of related products are "short battery life" and "slow charging speed," while the product features show that the target product has "large capacity battery" and "fast charging technology," then the first analytical agent correlates the above information to generate marketing fields including "long battery life" and "fast charging." After the above correlation analysis, the first analytical agent integrates the product features and marketing fields of the target product and outputs the basic recommendation information of the target product. The basic recommendation information can be a structured object in JSON format, which includes the target product's product name, product category, product attribute information, and marketing fields. Meanwhile, after generating basic recommendation information, the first analytical agent stores the basic recommendation information in the shared basic context, i.e., the extended parameter pool, as input data for subsequent stages, and pushes the SSE state update event completed in the first stage to the user terminal.

[0053] In one optional implementation, the second analytical agent is an intelligent proxy with scene planning and image generation scheduling capabilities, which takes over the output of the first analytical agent. The second analytical agent is cascaded with the first analytical agent, meaning that the execution of the second analytical agent depends on the basic recommendation information output by the first analytical agent. If the second analytical agent is invoked, the product information is parsed, including: Step b1: Based on preset scene prompts, analyze the basic recommendation information and reference information to obtain scene description information corresponding to the target area information.

[0054] In step b1, the preset scene prompts are pre-configured prompt templates used to guide the large language model in generating scene descriptions adapted to the target region information. The preset scene prompts contain multi-dimensional constraints required for scene generation, such as requiring the scene to adapt to the cultural preferences and aesthetic styles of the target region, requiring the scene to reflect the core attributes of the target product, and requiring the scene design to include content hooks that maximize consumer willingness, such as "dramatic conflict" or "expectation violation." After startup, the second analytical agent first reads the basic recommendation information output by the first analytical agent and the pre-stored reference information from the shared basic context. Then, it assembles the basic recommendation information, reference information, and target region information, and inputs them along with the preset scene prompts into the large language model. Guided by the preset scene prompts, the large language model parses the basic recommendation information and reference information to generate scene description information adapted to the target region information. The generated scene description information is structured text, which may include the scene theme, scene elements, the usage of the target product in the scene, the emotional or functional value conveyed by the scene, and the prompt descriptions required for scene image generation, etc.

[0055] Step b2 involves breaking down the scene description information into multiple scene identification information and generating scene images corresponding to each scene identification information.

[0056] In step b2, after obtaining the scene description information, the second analytical agent decomposes the scene description information into multiple individual scene identifiers. Each scene identifier corresponds to a single scene, and the scene identifier is used to uniquely identify an information unit of an independent image generation subtask. By decomposing the scene description information into multiple scene identifiers, the complex scene description containing multiple sub-scenes is transformed into a one-dimensional linear single-image task sequence, facilitating subsequent parallel image generation scheduling. After completing the decomposition of the scene identifiers, the second analytical agent calls the image task distributor to generate the corresponding scene image in parallel for each scene identifier. The image task distributor inputs the scene description corresponding to each scene identifier into the image generation model, which outputs the corresponding scene image data, including the image file itself and its corresponding access address. After all scene images are generated, the second analytical agent summarizes the scene description information and scene image data, aggregates them into scene information, stores it in the shared basic context, and simultaneously pushes the SSE state update event for the second phase to the user terminal.

[0057] In one optional implementation, the third analytical agent is an intelligent proxy with the ability to generate text and detailed images, and it takes over the output results of the preceding two analytical agents. The third analytical agent is cascaded with the second analytical agent, meaning that the execution of the third analytical agent depends on the scene information output by the second and first analytical agents. If the third analytical agent is invoked, the product information is parsed, including: Step c1: Based on product image information and multiple scene images, construct image consistency constraints.

[0058] In step c1, the image consistency constraint is used to ensure that the generated detail feature image, scene image, and product image information maintain a consistent visual style. Since the scene image and detail feature image in the recommended content information are different components of the same marketing plan, if their visual styles are inconsistent—for example, the scene image is a warm-toned outdoor style while the detail feature image is a cool-toned tech style—it will result in poor visual coordination of the overall recommended content information, reducing user experience. Therefore, an image consistency constraint is constructed to ensure consistency among the multiple images generated before and after the initial generation.

[0059] In step c1, after the third analytical agent starts, it first reads product image information and multiple scene images generated by the second analytical agent from the shared basic context. Then, it extracts style features from the product image information and the multiple scene images. The extracted style features include tonal distribution features, such as warm and cool tones; lighting condition features, such as natural light, artificial light, and backlighting; composition style features, such as centered composition and rule of thirds composition; and background style features, such as solid color background and scene-based background. After extracting the style features, an image consistency constraint is constructed based on the extracted style features. Specifically, the similarity between the detail feature image and the style features of the scene image in each style feature is not lower than a preset threshold, and it remains compatible with the style features of the product image information.

[0060] Step c2 generates detailed description text and marketing text for the target product based on basic recommendation information, reference information, and multiple scene images.

[0061] In step c2, the third analytical agent generates detailed description text and marketing text for the target product based on basic recommendation information, reference information, and multiple scene images. Specifically, the third analytical agent reads the basic recommendation information and reference information output by the first analytical agent and the scene description information generated by the second analytical agent from the shared basic context, merges and assembles the above information, and uses it as context input for the large language model. Through the large language model, it generates detailed description text and marketing text for the target product. The detailed description text is text that describes the detailed features of the target product in a specific usage scenario. It guides the generation direction of subsequent detailed feature images, enabling the detailed feature images to highlight the key details of the product, such as specific functional components, material textures, and interactive interfaces. The marketing text is text content used to showcase the product's advantages and promote consumption; it may include the target product's title, product description, and page copy.

[0062] Step c3: Based on image consistency constraints and detailed description text, generate a detailed feature image of the target product.

[0063] In step c3, after obtaining the detailed description text, the third analytical agent inputs the detailed description text as a prompt word into the image generation model, while simultaneously applying image consistency constraints to generate a detailed feature image of the target product. Specifically, the third analytical agent inputs the detailed description text as a positive prompt word into the image generation model, and uses the style features in the image consistency constraints as conditional input parameters for the image generation model, guiding the model to maintain consistent style features with the scene image when generating the detailed feature image. If the similarity of the style features between the generated detailed feature image and the scene image is lower than a preset threshold, the generated detailed feature image is corrected or regenerated until the image consistency constraint conditions are met. After the detailed feature image is generated, the third analytical agent aggregates the detailed description text, marketing text, and detailed feature image, stores them in a shared basic context, and pushes the SSE state update event for the third stage to the user terminal.

[0064] Thus, through the cascaded parsing process of the first, second, and third analytical agents, the product feature recognition and basic recommendation content generation, scene description and scene image generation, and detailed description and detailed feature image and marketing text generation are completed sequentially. The parsing and content generation in the three stages are passed down step by step and are linked together, ensuring that there is a clear logical dependency and visual consistency between the product features output in each stage.

[0065] This embodiment also provides a content generation apparatus for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0066] This embodiment provides a content generation device, such as... Figure 3 As shown, it includes: The information parsing module 301 is used to respond to the content generation request of the target product, and parse the product information of the target product in sequence based on the execution status and cascading relationship of multiple preset analysis agents to obtain multiple product features of the target product. Among them, the product features output by the preset analysis agent located before the cascading relationship are used as the input of the preset analysis agent located after the cascading relationship. The content generation module 302 is used to generate recommended content information for the target product based on multiple product features.

[0067] In one optional implementation, the information parsing module 301 includes: The reverse lookup unit is used to perform a reverse lookup of the product features of multiple preset analytical agents based on the cascading relationship, and to determine the target analytical agent whose execution status is execution failure. The information parsing unit is used to sequentially call the target analysis agent based on the cascading relationship to parse the product information and obtain multiple product features.

[0068] In one optional implementation, the reverse lookup unit includes: The snapshot lookup subunit is used to reverse-search for snapshots corresponding to product features of multiple preset analytical agents in persistent storage based on cascading relationships. The status unit is used to determine the execution status of the preset analysis agent as execution failure if no snapshot corresponding to the product feature of the preset analysis agent is found, or if the snapshot corresponding to the product feature of the preset analysis agent is invalid.

[0069] In one optional implementation, the plurality of preset analytical agents include a first analytical agent. If the first analytical agent is invoked, the information parsing unit includes: The image parsing subunit is used to perform image parsing on the product image information in the product information to obtain the product attribute features of the target product. The information association subunit is used to associate product attribute features with reference information in product information to obtain basic recommendation information for the target product. The reference information is used to characterize the product evaluation of the associated products corresponding to the target product.

[0070] In one optional implementation, the plurality of preset analytical agents include a second analytical agent, which is cascaded with the first analytical agent. If the second analytical agent is invoked, the information parsing unit includes: The information parsing subunit is used to parse basic recommendation information and reference information based on preset scene prompts to obtain scene description information corresponding to the target area information; The scene image generation subunit is used to decompose the scene description information into multiple scene identification information and generate scene images corresponding to the multiple scene identification information respectively.

[0071] In one optional implementation, the plurality of preset analytical agents include a third analytical agent, which is cascaded with the second analytical agent. If the third analytical agent is invoked, the information parsing unit includes: The constraint construction subunit is used to construct image consistency constraints based on product image information and multiple scene images; The text generation subunit is used to generate detailed description text and marketing text for the target product based on basic recommendation information, reference information, and multiple scene images. The detailed feature image generation subunit is used to generate detailed feature images of the target product based on image consistency constraints and detailed description text.

[0072] In one alternative embodiment, the device further includes: The product evaluation data acquisition module is used to acquire product evaluation data of related products corresponding to the target product. The product evaluation feature determination module is used to perform semantic parsing on the product evaluation data of related products to obtain the product evaluation features of related products. The reference information determination module is used to determine the reference information of the target product based on the data statistics and weight allocation of the product evaluation characteristics of related products.

[0073] The content generation apparatus provided in this embodiment of the invention can execute the content generation method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0074] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0075] The following is a detailed reference. Figure 4This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0076] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0077] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the content generation method of the embodiments of the present invention.

[0078] Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0079] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium after being downloaded via a network. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the content generation method shown in the above embodiments is implemented.

[0080] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0081] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A content generation method, characterized in that, The method includes: In response to a content generation request for a target product, based on the execution states and cascading relationships of multiple preset analytical agents, the product information of the target product is parsed sequentially to obtain multiple product features of the target product. Among them, the product features output by the preset analytical agent located before the cascading relationship serve as the input of the preset analytical agent located after the cascading relationship. Based on multiple product features, recommended content information for the target product is generated.

2. The method according to claim 1, characterized in that, The process involves sequentially parsing the acquired product information of the target product based on the execution states and cascading relationships of multiple preset analytical agents, thereby obtaining multiple product features of the target product, including: Based on the cascading relationship, the product features of multiple preset analytical agents are reverse-searched to determine the target analytical agent whose execution status is execution failure; Based on the cascading relationship, the target analysis agent is invoked sequentially to parse the product information and obtain multiple product features.

3. The method according to claim 2, characterized in that, The step of performing a reverse lookup of the product features of multiple preset analytical agents based on the cascading relationship of these agents to determine the target analytical agent whose execution status is "execution failed" includes: Based on the cascading relationship, in persistent storage, snapshots corresponding to the product features of multiple preset analytical agents are searched in reverse. If no snapshot corresponding to the product features of the preset analysis agent is found, or if the snapshot corresponding to the product features of the preset analysis agent is invalid, the execution status of the preset analysis agent is determined to be execution failure.

4. The method according to any one of claims 1-3, characterized in that, The plurality of preset analytical agents include a first analytical agent. If the first analytical agent is invoked, the product information is parsed, including: Image analysis is performed on the product image information in the product information to obtain the product attribute features of the target product; The product attribute features are associated with the reference information in the product information to obtain the basic recommendation information of the target product, wherein the reference information is used to characterize the product evaluation of the associated products corresponding to the target product.

5. The method according to claim 4, characterized in that, The plurality of preset analytical agents include a second analytical agent, which is cascaded with the first analytical agent. If the second analytical agent is invoked, the product information is parsed, including: Based on preset scene prompts, the basic recommendation information and the reference information are parsed to obtain scene description information corresponding to the target area information; The scene description information is decomposed into multiple scene identification information, and scene images corresponding to multiple scene identification information are generated respectively.

6. The method according to claim 5, characterized in that, The plurality of preset analytical agents include a third analytical agent, which is cascaded with the second analytical agent. If the third analytical agent is invoked, the product information is parsed, including: Based on the product image information and multiple scene images, image consistency constraints are constructed. Based on the basic recommendation information, the reference information, and the multiple scene images, generate detailed description text and marketing text for the target product; Based on the image consistency constraints and the detailed description text, a detailed feature image of the target product is generated.

7. The method according to claim 4, characterized in that, The method further includes: Obtain product evaluation data for related products corresponding to the target product; Semantic parsing is performed on the product evaluation data of the associated products to obtain the product evaluation features of the associated products; Based on the statistical analysis and weighting of the product evaluation characteristics of the related products, reference information for the target product is determined.

8. A content generation apparatus, characterized in that, The device includes: The information parsing module is used to respond to the content generation request of the target product, and based on the execution status and cascading relationship of multiple preset analysis agents, sequentially parse the product information of the target product to obtain multiple product features of the target product. Among them, the product features output by the preset analysis agent located before the cascading relationship are used as the input of the preset analysis agent located after the cascading relationship. The content generation module is used to generate recommended content information for the target product based on multiple product features.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the content generation method of any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the content generation method according to any one of claims 1 to 7.

Images