Data processing method and device

Abstract

The embodiment of the invention provides a data processing method and device, and the method comprises the steps: determining a commodity recommendation task, and obtaining the multi-modal commodity data of a target commodity related to the commodity recommendation task; adjusting the multi-modal commodity data by using a multi-modal model, and generating a commodity recommendation information card corresponding to the target commodity according to an adjustment result; processing the task theme information of the commodity recommendation task by using an information processing model to obtain image generation information corresponding to the commodity recommendation task, and generating a recommendation task theme image according to the image generation information by using an image generation model; generating a commodity recommendation information block corresponding to the commodity recommendation task based on the recommendation task theme image and the commodity recommendation information card; the problems that the generation efficiency of the promotion data is low and the marketing promotion demand cannot be met due to the fact that the corresponding promotion data is designed for the commodity in a manual editing mode are avoided.

Description

TECHNICAL FIELD

[0001] Embodiments of the present specification relate to the technical field of artificial intelligence, and in particular to a data processing method. One or more embodiments of the present specification also relate to a data processing apparatus, a computing device, a computer-readable storage medium, and a computer program product. BACKGROUND

[0002] With the continuous development of computer technology, in the scenario of marketing and promoting goods, it is necessary to design promotion data containing product promotion copy and product promotion image for the goods, and provide the promotion data to the user, so that the user can clearly understand the selling points of the goods.

[0003] In the current process of marketing and promoting goods, artificial editing is used to design corresponding promotion data for goods, resulting in low efficiency of generating promotion data, which cannot meet the needs of marketing and promotion. Therefore, how to efficiently generate promotion data for goods has become a technical problem to be solved. SUMMARY

[0004] Therefore, the embodiments of the present specification provide a data processing method. One or more embodiments of the present specification also relate to a data processing apparatus, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects in the prior art.

[0005] According to a first aspect of the embodiments of the present specification, a data processing method is provided, comprising:

[0006] determining a product recommendation task, and obtaining multi-modal product data of a target product associated with the product recommendation task;

[0007] adjusting the multi-modal product data using a multi-modal model, and generating a product recommendation information card corresponding to the target product according to the adjustment result;

[0008] processing task theme information of the product recommendation task using an information processing model, obtaining image generation information corresponding to the product recommendation task, and generating a recommendation task theme image according to the image generation information using an image generation model;

[0009] generating a product recommendation information block corresponding to the product recommendation task based on the recommendation task theme image and the product recommendation information card.

[0010] According to a second aspect of the embodiments of the present specification, a data processing apparatus is provided, comprising:

[0011] The data determination module is configured to determine a commodity recommendation task and acquire multi-modal commodity data of a target commodity associated with the commodity recommendation task.

[0012] The information card determination module is configured to adjust the multi-modal commodity data by using a multi-modal model, and generate a commodity recommendation information card corresponding to the target commodity according to an adjustment result.

[0013] The image determination module is configured to process task theme information of the commodity recommendation task by using an information processing model, acquire image generation information corresponding to the commodity recommendation task, and generate a recommendation task theme image according to the image generation information by using an image generation model.

[0014] The information block determination module is configured to generate a commodity recommendation information block corresponding to the commodity recommendation task based on the recommendation task theme image and the commodity recommendation information card.

[0015] According to a third aspect of the embodiments of the present specification, a computing device is provided, comprising:

[0016] a memory and a processor;

[0017] The memory is configured to store computer programs / instructions, and the processor is configured to execute the computer programs / instructions, which realize the steps of the above data processing method when executed by the processor.

[0018] According to a fourth aspect of the embodiments of the present specification, a computer readable storage medium is provided, which stores computer programs / instructions, which realize the steps of the above data processing method when executed by the processor.

[0019] According to a fifth aspect of the embodiments of the present specification, a computer program product is provided, comprising computer programs / instructions, which realize the steps of the above data processing method when executed by the processor.

[0020] One or more embodiments of the specification provide a data processing method, after obtaining the multi-modal commodity data of the target commodity associated with the commodity recommendation task, in order to improve the generation efficiency and quality of the recommendation data, a plurality of models can be used for processing; first, the multi-modal commodity data can be adjusted by using a multi-modal model, and the commodity recommendation information card corresponding to the target commodity is generated according to the adjustment result; second, the task theme information of the commodity recommendation task is processed by using an information processing model, and the image generation information corresponding to the commodity recommendation task is obtained; finally, the image generation model generates the recommendation task theme image according to the image generation information; the automatic processing of the recommendation data is realized by using multiple models, thereby improving the generation efficiency of the recommendation task theme image and the commodity recommendation information card, and the fine data processing operation of multiple steps can obtain the recommendation task theme image and the commodity recommendation information card with high quality; then, based on the recommendation task theme image and the commodity recommendation information card, the high-quality commodity recommendation information block corresponding to the commodity recommendation task can be quickly generated; thereby realizing the efficient generation of the promotion data of the commodity; avoiding the problem that the manual editing method is used to design the corresponding promotion data for the commodity, resulting in low generation efficiency of the promotion data and unable to meet the needs of marketing promotion. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is an application diagram of a data processing method provided by one embodiment of the specification;

[0022] Figure 2 is a flowchart of a data processing method provided by one embodiment of the specification;

[0023] Figure 3 is a schematic diagram of screening commodity images in a data processing method provided by one embodiment of the specification;

[0024] Figure 4 is a schematic diagram of commodity images in a data processing method provided by one embodiment of the specification;

[0025] Figure 5 is a schematic diagram of a business card in a data processing method provided by one embodiment of the specification;

[0026] Figure 6 is a schematic diagram of generating a banner image in a data processing method provided by one embodiment of the specification;

[0027] Figure 7 is a schematic diagram of a banner image in a data processing method provided by one embodiment of the specification;

[0028] Figure 8 is a process flowchart of a data processing method provided by one embodiment of the specification;

[0029] Figure 9 is a structural block diagram of a computing device provided by an embodiment of the present specification. DETAILED DESCRIPTION

[0030] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present specification. However, the present specification can be practiced without the specific details, other than in the examples described herein, and it is understood that the scope of the present specification is not limited to the details below. In other instances, well-known methods associated with computing, software development, and / or data analysis have not been described in detail in order to avoid unnecessarily obscuring aspects of the present specification.

[0031] The terminology used in one or more embodiments of the present specification is for the purpose of describing particular embodiments only and is not intended to be limiting of one or more embodiments of the present specification. As used in one or more embodiments of the present specification and the accompanying claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in one or more embodiments of the present specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0032] It will be understood that, although the terms first, second, etc. can be used herein to describe various information, these terms are not intended to denote a temporal or chronological order. Rather, these terms are used solely to distinguish one from another only. For example, without departing from the scope of one or more embodiments of the present specification, first can be termed second, and similarly, second can be termed first. The term "if' as used herein, can be interpreted as meaning "when" or "in response to determining" depending on the context.

[0033] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in one or more embodiments of the present specification are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0034] 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.

[0035] 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.

[0036] First, the terms involved in one or more embodiments of this specification are explained.

[0037] A banner image is a banner-style image used to attract user attention and convey key information. It plays an important role in digital marketing and web design, often appearing as a banner ad or logo to promote a product, event, or brand image.

[0038] EDM (Email Direct Marketing) refers to email marketing. It is a way of delivering marketing information directly to potential and existing customers through emails or marketing emails.

[0039] Marketing emails: refer to emails sent by companies or individuals to target audiences for marketing purposes such as promoting products, services, brands, boosting sales, and increasing awareness.

[0040] Fatigue control: In the fields of marketing, advertising, and customer relationship management, a series of strategies and techniques are adopted to adjust and optimize the frequency, content, and form of marketing activities in order to avoid the target audience's boredom or numbness to the information of a certain brand or product. Fatigue control includes: 1. Frequency control: Limit the number of times a single user contacts email marketing within a certain period of time. Through data analysis, better display frequency is understood to avoid overexposure leading to user's aversion. 2. Creative rotation: Regularly update email marketing, including copy, images, videos, and other content to attract users' attention and stimulate their interest.

[0041] AIGC (AI Generated Content) technology: refers to artificial intelligence generated content, which is a technology that uses artificial intelligence to automatically generate text, images, audio, video, and other forms of content.

[0042] AI text-to-image: is a specific application of AIGC technology, which refers to a technology that allows users to input a text description and an artificial intelligence model generates an image based on the text content.

[0043] AI text-to-text: is an application of AIGC technology, which refers to a technology that allows an artificial intelligence model to generate new text content based on input text prompts.

[0044] Product information: refers to various descriptions and data related to a product, including the product's basic attributes, functional characteristics, usage methods, quality standards, prices, after-sales services, and other aspects.

[0045] With the continuous development of computer technology, in the context of marketing and promoting products, it is necessary to design promotion data containing product promotion copy and product promotion images for products and provide the promotion data to users, so that users can clearly understand the selling points of the product. In the current process of marketing and promoting products, artificial editing is used to design corresponding promotion data for products, resulting in low efficiency of generating promotion data and failing to meet the needs of marketing and promotion.

[0046] For example, the user growth EDM channel content production process mainly relies on manual experience, including event theme mining, theme category / pool matching, email content production, and delivery matching. The existing marketing email production system has the following shortcomings: 1. Theme promotion picture production mainly relies on manual operation; due to the timeliness of theme emails, high requirements are placed on tracking hot events and popular trends, and manual operation has a long time cycle and high labor cost. Secondly, the commodity information display is insufficient. On the one hand, the title and selling point text are usually very long, with an average of 30-40 words. Due to the page length-width ratio problem of the email, the title cannot be fully displayed, and only the first 4-5 words are displayed. Users cannot see the complete title during browsing, and since the original length of the commodity is relatively long, even if the title is fully displayed, users also have difficulty in understanding at a glance, and the characteristics of the commodity are not effectively conveyed. On the other hand, the commodity picture usually contains a large amount of text and LOGO, which has potential conflicts and interference with the theme content of the marketing email.

[0047] Based on this, in the present specification, a data processing method is provided, and one or more embodiments of the present specification simultaneously relate to a data processing device, 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.

[0048] Referring to Figure 1 , Figure 1 An application schematic diagram of a data processing method according to an embodiment of the present specification is shown, based on Figure 1 It can be known that in the process of marketing and promoting the target commodity, the server 104 can select the current hot theme information from the theme knowledge base, determine the target commodity based on the theme information, and adjust the multi-modal commodity data of the target commodity using a large model to obtain a commercial card of the target commodity; and input the theme information into the large model to obtain a banner prompt required for generating a banner, and then input the banner prompt into a text-to-image model for image generation to obtain a banner; finally, generate a marketing email of the target commodity based on the banner and the commercial card; then the server 104 can send the marketing email to the client 102.

[0049] Referring to Figure 2 , Figure 2 A flowchart of a data processing method according to an embodiment of the present specification is shown, which specifically includes the following steps. It should be noted that Figure 2The step 204 and the step 206 in the method 200 can be executed in parallel or in sequence. In the parallel execution, the step 204 and the step 206 can be executed simultaneously. In the sequential execution, the step 204 can be executed first or the step 206 can be executed first, which is not limited herein.

[0050] The step 202: determining a product recommendation task and obtaining multi-modal product data of a target product associated with the product recommendation task.

[0051] The product recommendation task can be understood as a task of recommending a target product. For example, the product recommendation task can be a product marketing task or a marketing promotion task for a target product. The target product associated with the product recommendation task can be one or more.

[0052] The multi-modal product data can be understood as data related to a target product and / or data describing the target product. The multi-modal product data can be product description text and product visual data. The product description text can be text information describing a target product. The product description text can be a product title, product content, or usage method of a target product. For example, the product description text can be the product information described above.

[0053] The product visual data can be understood as data showing a target product from a visual aspect. The product visual data can include product images, product videos, and / or product three-dimensional models.

[0054] In one or more embodiments provided in the specification, the determining a product recommendation task and obtaining multi-modal product data of a target product associated with the product recommendation task comprises:

[0055] Selecting task theme information from a plurality of candidate theme information stored in a theme information storage unit, and generating a product recommendation task based on the task theme information;

[0056] Processing the task theme information by using a language processing model to obtain product attribute information corresponding to the task theme information;

[0057] Based on the product attribute information, determining the target product associated with the product recommendation task from a product data storage unit, and obtaining the multi-modal product data of the target product.

[0058] The theme information storage unit can be understood as a unit for storing task theme information, which can be a database, a server, a local disk, etc. The task theme information in the theme information storage unit can be stored by the operator through the client into the theme information storage unit; or can be obtained from the Internet by the theme information acquisition server and stored into the theme information storage unit. The candidate theme information can be understood as the theme information stored in the theme information storage unit that can be selected; the task theme information can be understood as the theme information used to generate the corresponding theme information of the product recommendation task; and the task theme information can be used to determine the target product associated with the product recommendation task and the product recommendation information block corresponding to the product recommendation task. For example, the theme information storage unit can be a theme knowledge base, which stores content (task theme information) from current hot topics, local holidays, social media hot topics, etc. The theme (i.e. task theme information) can be collected in real time through external data (automatically supplemented by search, etc.); at the same time, the operator can also manually supplement through the client. For example, the theme includes but is not limited to Christmas, the American Super Bowl event, winter skiing, mountaineering camping, etc. In EDM marketing operation, the theme is the central theme of the marketing email, and each component (title, Banner picture, product list) of the email needs to be produced in line with the theme. For example, in the process of producing a product pool (a plurality of target products) for the American Super Bowl event, a batch of goods are selected as marketing content through keyword matching, category matching, algorithm recommendation, etc. The target product is generally a combination of multiple goods across categories, such as team flags, star jerseys, hats, mugs, etc.

[0059] The language processing model can be understood as a model capable of processing natural language, for example, the language processing model is a large model, a large language model, etc.

[0060] The product attribute information can be understood as information representing the product type or key content corresponding to the target product; in the case of multiple target products, the product attribute information can be multiple. For example, the product attribute information is product type information or product keywords. It should be noted that the theme category word (product attribute information) is a specific category under the theme, a product keyword, and is used to filter specific product lists. For example, the theme of the Super Bowl can be associated with sports, sports clothing, baseball gloves, etc. The category can be an international product category. For example, high-end tea sets (theme) in the international station are in the first category of home and garden, and the category path is home and garden -> tableware, coffee and wine set -> tableware -> complete tableware. This theme of outdoor camping generally includes products of multiple categories, such as the category of tents is sports and entertainment -> camping and hiking -> outdoor tents; the category of mountaineering boots is sports and entertainment -> sports shoes, sports bags and accessories -> sports shoes -> hiking shoes.

[0061] The product data storage unit can be understood as a unit for storing multimodal product data. For example, the product data storage unit can be a database or a server; for example, the product data storage unit can be a server or database corresponding to an online shopping platform or an online shopping system.

[0062] Specifically, during the product recommendation process, this method needs to determine multiple candidate topic information stored in the topic information storage unit in order to perform the product recommendation task, and select task topic information from the multiple candidate topic information to generate the product recommendation task, wherein the task topic information can be any one or more of the multiple candidate topic information; or, the task topic information can be the latest candidate topic information among the multiple candidate topic information. After obtaining the task topic information, it is necessary to generate a product recommendation task based on the task topic information so that product recommendations can be made subsequently.

[0063] After generating a product recommendation task, the task subject information needs to be input into a language processing model for processing to obtain product attribute information associated with the task subject information; then, based on the product attribute information, multiple candidate products corresponding to the product attribute information are determined from a product data storage unit; and the target product associated with the product recommendation task is selected from the multiple candidate products. It should be noted that in the process of selecting the target product associated with the product recommendation task from multiple candidate products, a candidate product with higher quality (for example, sales greater than a preset amount and / or a user evaluation score greater than a preset score threshold) can be selected as the target product; or, an algorithm recommendation strategy can be used to select the target product associated with the product recommendation task from multiple candidate products.

[0064] After determining the target product associated with the product recommendation task, multimodal product data corresponding to the target product may be obtained.

[0065] Taking the application of the data processing method provided in this specification in the marketing email production scenario as an example, the data processing method is explained; with the increasing maturity of multimodal large models, AIGC technology has penetrated into various production links of users, and marketing emails need to be generated in the process of product marketing and promotion; in terms of email content production, the application of AI text-to-text and text-to-image capabilities can significantly improve the richness and novelty of the content delivered, provide more real-time and effective hot events, produce more diverse main pictures and illustrations, and more efficient expression of product selling points. Based on this, the method can obtain the subject that needs to be marketed and promoted from the subject database (i.e., task subject information), and generate a product promotion task (i.e., product recommendation task) based on the subject; then, the subject is input into the large model (i.e., language processing model) for processing to obtain the product category or keyword (i.e., product attribute information) corresponding to the subject.

[0066] Then, a batch of goods are screened as marketing goods (i.e., target goods) from the online shopping platform database through keyword matching, category matching, and / or algorithm recommendation, and the product information and product images (i.e., multi-modal product data) of the marketing goods are obtained.

[0067] Based on the above, the method can quickly capture the latest market trends by analyzing current trends and hot topics, and integrate these hot topics into the content of the marketing email. This not only improves the relevance of the email, but also enhances the customer's sense of participation and interactivity, and promotes the sales conversion rate. Moreover, based on keyword matching, category matching, and algorithm recommendation, the most suitable goods for marketing promotion can be accurately selected from a large amount of product data. In this way, the advantages of the goods can be most effectively conveyed within the limited email length, improving the efficiency of the consumer's purchase decision.

[0068] Step 204: adjusting the multi-modal product data using a multi-modal model, and generating product recommendation information cards corresponding to the target goods according to the adjustment results.

[0069] The product recommendation information card can be understood as a card containing information that needs to be recommended for the target goods. For example, the product recommendation information card can be displayed to the target user, thereby recommending the target goods to the user. The product recommendation information card can be a component of the UI.

[0070] The multi-modal model can be understood as a model capable of processing multi-modal product data. The multi-modal model can be a multi-modal large model or a model architecture containing multiple models.

[0071] Continuing with the above example, in the process of promoting goods, the AI needs to reconstruct the business card. The direction of the reconstruction of the product information can be to extract the core selling points of the goods. Through concise information expression, the user can be provided with more intuitive and easy-to-read product content, so that the user can understand the characteristics of the goods at a glance. The user's interest in the goods is stimulated through scene scripts and use experiences, thereby improving the click-through rate of the goods in the email. Therefore, the multi-modal product data needs to be adjusted by AI to obtain product information that meets the requirements of marketing promotion. In the process of AI reconstructing the business card, the model input content of the reconstruction of the business card information can be a prompt constructed based on the product title and product attributes. In the prompt, multiple tasks such as selling point extraction, keyword selection, and creative copywriting can be specified at the same time, so that the multi-modal large model (i.e., the multi-modal model) can perform multiple operations based on the model input content. Moreover, the multi-modal large model can also adjust the visual data of the goods, thereby obtaining visual data of the goods that meets the requirements of the promotion of the goods.

[0072] In one or more embodiments provided in the specification, the multi-modal model comprises a text processing model and a data evaluation model;

[0073] The adjusting the multi-modal commodity data by using the multi-modal model comprises steps one to four:

[0074] Step one: determining commodity description text and commodity visual data from the multi-modal commodity data.

[0075] The text processing model can be understood as a model capable of rewriting commodity description text, for example, the text processing model can be a large language model, a large model, etc.

[0076] The data evaluation model can be understood as a model capable of evaluating commodity visual data, which can be a visual processing model or a deep learning model.

[0077] Specifically, in the process of generating the commodity recommendation information card, the multi-modal commodity data needs to be processed by using the text processing model and the data evaluation model, so the commodity description text and the commodity visual data need to be divided from the multi-modal commodity data, so as to facilitate subsequent processing. It should be noted that the processing operations for commodity description text and commodity visual data can be executed in parallel or sequentially.

[0078] Step two: using the text processing model to rewrite the commodity description text according to the task theme information, to obtain the commodity recommendation text of the target commodity.

[0079] In the above example, in the process of AI reconstruction of the business card, the model input content of the business card information reconstruction can be a Prompt constructed based on the commodity title and the commodity attributes (business details content), and the Prompt can simultaneously specify multiple tasks such as selling point extraction, keyword selection and / or creative copy. The commodity selling point extraction can be used for the business card information reconstruction of the EDM list mail this time; the keyword selection can be used for the commodity matching model of the increment domain (user growth domain) at each level; and the creative copy can be applied to the free SNS channel, for scenarios such as social media posts and commodity image matching texts. Based on this, the Prompt is input into the large model for rewriting, and the AI refined business card content (i.e. the commodity recommendation text) can be obtained. The specific examples can refer to Table 1 below.

[0080] Table 1

[0081]

[0082]

[0083] Based on the above embodiments, it can be seen that this method can accurately identify and extract the core selling points of a product by constructing a prompt using the product title and attribute information as input. This is crucial for reconstructing business card information in EDM (email direct marketing) list emails, because the highlighted product selling points can greatly increase consumer interest, thereby improving the click-through rate and conversion rate of the email. In addition, by inputting the prompt containing multi-level tasks (such as selling point extraction, keyword selection, creative copywriting, etc.) into the large model for processing, the business card content refined by AI can be obtained. This method not only ensures the quality and relevance of the output content, but also flexibly adjusts the style and focus according to the specific marketing scenario, so that the business card information finally presented to consumers is more vivid, specific and persuasive.

[0084] Step three: using a data evaluation model to evaluate the product visual data to obtain a visual evaluation result, and determining the recommended visual data of the target product from the product visual data based on the visual evaluation result.

[0085] The visual evaluation result may refer to a score obtained by evaluating the visual data of the product, or information obtained by evaluating the visual data of the product.

[0086] The recommended visual data can be understood as the visual data required for recommending the target product. The recommended visual data can be a product recommendation image or a product recommendation video.

[0087] In one or more embodiments provided in this specification, the product visual data is a plurality of product images, and the recommended visual data is a product recommendation image;

[0088] The step of evaluating the product visual data using a data evaluation model to obtain a visual evaluation result, and determining recommended visual data of the target product from the product visual data based on the visual evaluation result, includes:

[0089] Performing image evaluation on the plurality of product images using manual text detection, text risk detection, aesthetic detection, sentiment detection, and quality detection in the data evaluation model to obtain an image score corresponding to each product image;

[0090] Based on the image score, a product recommendation image of the target product is selected from the plurality of product images.

[0091] The artificial text detection can be understood as detecting whether the commodity image contains artificially added text; the text risk detection can be understood as detecting whether the text in the commodity image is risky; the aesthetic detection can be understood as performing aesthetic evaluation on the commodity image from an aesthetic perspective; the emotion detection can be understood as detecting the mood or emotion corresponding to the commodity image.

[0092] In the process of commodity recommendation, considering that the artificial image selection scheme is time-consuming and laborious, the efficiency is low. Therefore, by means of the image understanding ability of the AI large model, high-quality pictures can be quickly screened on a large scale. This not only improves the efficiency of image selection, but also ensures that the selected pictures are more in line with the preferences of the platform and users, thereby improving the attractiveness and dissemination effect of the content. The specific technical scheme is as follows:

[0093] First, in the process of screening commodity images by AI, the commodity images can be input into the large model for evaluation to obtain image scores.

[0094] It should be noted that the large model has the following capabilities in image content recognition: 1. Accurate insight into objects and scenes (i.e., aesthetic detection): Whether it is common objects such as coffee cups and mobile phones, scenes such as conference rooms and natural scenery, or human actions such as running and shaking hands, any key elements can be accurately identified. 2. Text recognition (i.e., artificial text detection and text risk detection): Printed and handwritten text in images, such as road signs, documents, billboards, and logos, can be quickly and accurately extracted to achieve efficient text recognition (OCR). 3. Deep detail capture (quality detection): In-depth analysis of color, shape, and spatial relationships in images, such as the red vase on the left side of the table, can clearly grasp and extract hidden information from images. 4. Emotional intent analysis (emotion detection): It can sensitively capture the emotions of people in images and the atmosphere of the scene, whether it is the lively atmosphere of a happy gathering or the pressure of a tense work scene, which can be accurately identified.

[0095] Secondly, based on the image scores, a commodity image with better quality (i.e., a commodity recommendation image) is selected from a plurality of commodity images, for example, the plurality of commodity images are sorted in descending order based on the image scores, and the commodity image ranked first in the commodity image sequence is selected as the commodity image with better quality.

[0096] Based on the content of the above embodiments, the method can perform accurate aesthetic detection, efficient text recognition and risk control, detailed quality detection, and precise analysis of emotional intent on the product images. Through comprehensive evaluation in various dimensions, the large model scores each product image and selects the image with the highest score as the recommended product image. This approach ensures that the selected image achieves an optimal level in terms of aesthetics, text, quality, and emotional expression, thereby improving the overall quality and success rate of marketing activities. Not only does it greatly improve the efficiency and accuracy of product image selection, but it also optimizes the visual appeal of marketing materials, helping to attract more potential customers and improve the market competitiveness of the brand.

[0097] In one or more embodiments provided in the specification, the image evaluation of the plurality of product images using the artificial text detection, text risk detection, aesthetic detection, emotional detection, and quality detection in the data evaluation model obtains image scores corresponding to each product image, including:

[0098] Inputting the target product image into the data evaluation model for artificial text detection and text risk detection, wherein the target product image is any one of the plurality of product images;

[0099] In the case where the target product image passes the artificial text detection and the text risk detection, inputting the target product image into the data evaluation model for aesthetic detection and emotional detection;

[0100] In the case where the target product image passes the aesthetic detection and the emotional detection, inputting the target product image into the data evaluation model for quality detection, obtaining an image quality score of the target product image, and determining the image quality score as the image score corresponding to each product image.

[0101] Following the above example, Figure 3 is a schematic diagram of a data processing method for screening product images provided by an embodiment of the specification, based on Figure 3 It can be seen that the steps of the process of the method for screening product images are as follows:

[0102] 1. Input the picture into the visual deep learning model for processing.

[0103] Wherein, the picture is a product picture, and the visual deep learning model is an AI visual engine.

[0104] Figure 4 is a schematic diagram of a product image in a data processing method provided by an embodiment of the specification, and the reference Figure 4 It can be seen that the product image corresponding to the Chinese knot product can be Figure 4A, B, C, D, E, F.

[0105] 2. Utilize multiple content detection modules in the visual deep learning model to respectively evaluate the image quality score of the product picture, text audit, embedded text detection, aesthetic and emotional detection, and detect the threshold value according to the output results of the visual deep learning model.

[0106] The specific detection method is:

[0107] (1) Input the product image into the visual deep learning model for embedded text detection to obtain the text probability, and determine whether the text probability is <= 0.0017. If yes, it is determined that no text is contained, and text audit is performed. If not, it is determined that text is contained, and the detection is not passed;

[0108] (2) Input the product image into the visual deep learning model for text audit to obtain a result representing whether there is illegal text in the image. In the case of determining that there is no illegal text according to the result, perform audit and emotional detection. If it is determined that there is illegal text according to the result, it is determined that the detection is not passed;

[0109] (3) Input the product image into the visual deep learning model for aesthetic and emotional detection to obtain a result representing the aesthetic and emotional of the product image. In the case of determining that there is no abnormality in aesthetic and emotion according to the result, perform the operation of evaluating the image quality. If it is determined that there is abnormality in aesthetic and emotion according to the result, it is determined that the detection is not passed;

[0110] It should be noted that aesthetic and emotional detection is completed by a multi-modal large model, which inputs the product image into the large model to judge the picture content. Aesthetic and emotion is one of the scoring items, and currently adopts a 10-point system. Low-score pictures will be filtered out.

[0111] The Prompt case corresponding to aesthetic and emotional detection is: "Please perform aesthetic and emotional detection on the following picture. First, from the perspective of aesthetics, evaluate whether the composition of the picture is reasonable, whether the color matching is harmonious, whether the light and shadow are used appropriately, whether the arrangement and combination of picture elements have aesthetic sense, and whether they exhibit unique artistic style or creativity. Then perform emotional detection to judge whether the picture conveys a positive (such as happiness, warmth, hope, etc.), negative (such as sadness, fear, anger, etc.) or neutral emotional atmosphere, and whether it can evoke a certain specific emotional resonance in the viewer. Please elaborate on your analysis process and final judgment result, and express the language as accurately, clearly and logically as possible. Score according to the 10-point system, give each item score, and give the final score."

[0112] (4) input the commodity image into a visual deep learning model to evaluate the image quality score of the commodity image. Subsequently, commodity images with an image quality score >= 0.6 can be determined as pre-screened commodity images; and the image quality score can be used to rank the pre-screened commodity images, thereby facilitating the selection of better commodity images.

[0113] It should be noted that the specific evaluation of the image quality score, text review, embedded text detection, aesthetic and emotional detection results can refer to Table 2 below.

[0114] Table 2

[0115]

[0116]

[0117] Based on the content of the above embodiments, it can be known that the AI screening method of the commodity image is applied to the commodity list production link of the mail, and the better one is selected from multiple images of the commodity. Since the number of commodity images is large and the selection model is stable and reliable, the better one is selected from multiple images of the commodity. Specifically, the method can accurately detect the aesthetics of the commodity image, efficiently recognize the text and control the risk, carefully detect the quality, and accurately analyze the emotional intent: through the comprehensive evaluation of the above various dimensions, the large model scores each commodity image, and selects the image with the highest score as the recommended commodity image. This method ensures that the selected image reaches an optimal level in terms of aesthetics, text, quality, and emotional expression, thereby improving the overall quality and success rate of marketing activities. Not only does it greatly improve the efficiency and accuracy of commodity image screening, but it also optimizes the visual performance of marketing materials, helping to attract more potential customers and improve the market competitiveness of the brand.

[0118] Step four: based on the commodity recommendation text and the recommended visual data, generate the commodity recommendation information card corresponding to the target commodity.

[0119] Specifically, the method can combine the commodity recommendation text and the recommended visual data to generate the commodity recommendation information card corresponding to the target commodity.

[0120] In one or more embodiments provided in the specification, the generation of the commodity recommendation information card corresponding to the target commodity based on the commodity recommendation text and the recommended visual data comprises:

[0121] determining the commodity purchase link of the target commodity from the modal commodity data;

[0122] combining the commodity recommendation text, the recommended visual data, and the commodity purchase link to obtain the commodity recommendation information card corresponding to the target commodity.

[0123] Among them, the product purchase link can be understood as the shopping URL, purchase website, etc. of the target product.

[0124] Using the above example, Figure 5 This is a schematic diagram of a business card in a data processing method provided in one embodiment of this specification, based on Figure 5 It can be seen that this method can generate a business card corresponding to the product based on the combination of product recommendation text, high-quality product images and product purchase links (i.e. Figure 5 (1)(2)(3) in ). It should be noted that the "Buy now" button of Shangka China corresponds to the product purchase link of the product.

[0125] Based on the content of the above embodiments, it can be seen that this method can effectively integrate the key information of the product - including recommendation text, product images and purchase links - into a whole. This integration not only improves the integrity and coherence of the information, but also ensures that users can obtain all necessary decision-making support materials at one time when receiving recommendation information, greatly simplifying the user's shopping experience. By providing intuitive product images, detailed recommendation texts and direct purchase links, this solution greatly improves the user's browsing efficiency and purchasing convenience. Users can complete the entire process from understanding to purchasing without jumping or searching between different pages, thereby improving the overall user satisfaction and loyalty; it not only optimizes the user's shopping path and improves the user experience, but also strengthens the effectiveness of marketing activities and promotes the achievement of sales goals.

[0126] Step 206: Use the information processing model to process the task theme information of the product recommendation task, obtain the image generation information corresponding to the product recommendation task, and use the image generation model to generate a recommendation task theme image based on the image generation information.

[0127] The image generation information can be understood as information used to generate the recommended task theme image, such as the prompt or word for a cultural image. The recommended task theme image can be understood as a promotional image related to the task theme and used to recommend products. For example, the recommended task theme image can be a banner image.

[0128] In one or more embodiments provided in this specification, the image generation information is image generation prompt text;

[0129] The step of processing the task theme information of the product recommendation task using the information processing model to obtain image generation information corresponding to the product recommendation task, and generating a recommendation task theme image based on the image generation information using the image generation model includes:

[0130] The task topic information of the commodity recommendation task is processed by using a language processing model to obtain commodity attribute information and theme scene information corresponding to the task topic information, wherein the theme scene information is used to describe the application scene corresponding to the task topic information, and the theme scene information is obtained by performing scene analysis on the task topic information by using a scene analysis model;

[0131] The information processing model is used to generate image generation prompt text corresponding to the commodity recommendation task according to the task topic information, the commodity attribute information and the theme scene information;

[0132] The image generation model is used to generate the recommendation task theme image corresponding to the commodity recommendation task according to the image generation prompt text.

[0133] The image generation prompt text can be understood as a prompt for text-to-image generation; the scene analysis model can be understood as a model capable of performing scene analysis on the task topic information to obtain theme scene information, for example, the scene analysis model is a large model or a deep learning model.

[0134] In the above example, considering that the current AI text-to-image generation large model has been greatly improved in capability, such as the open source stablediffusion large model, which has good performance in text-to-image generation, therefore, in the EDM list banner image of this time, the method uses an AI text-to-image generation large model to generate a corresponding banner image for each theme, which has great efficiency improvement compared with the traditional manual design scheme, and the picture quality can also be guaranteed. In the process of executing AI theme image generation, the AI-generated theme promotion image (i.e., the recommendation task theme image) needs to match the theme and the style needs to conform to the commodity + e-commerce scene background under the theme category, therefore, in-depth research is needed on the design of the text-to-image generation prompt and the large model parameters, especially the theme banner prompt generation needs to be paid attention to: the categories included in each theme are different, so the generation prompt of the theme promotion image is dynamically changed and cannot be set in place at one step, therefore, the method adopts an AI large model to generate a theme promotion image prompt. When constructing the text-to-image generation prompt, the following issues need to be paid attention to: 1. The background needs to conform to the e-commerce picture background and cannot be too messy; 2. The relevant commodities under each theme category need to be generated, the number of which should not be too many, 3-4 is optimal, and the placement of the commodities needs to be reasonable; 3. The generation of text content needs to be controlled, because the current text-to-image generation large model is not very good in generating text, and it is easy to form blurred text,

[0135] Figure 6 is a schematic diagram for generating a banner image in a data processing method according to an embodiment of the present specification; based on Figure 6It can be seen that the specific process of generating a theme promotion image by using the AI method is as follows:

[0136] (1) The theme information is analyzed by using a large model to generate theme scene words (i.e., theme scene information). The scene words are detailed descriptions of the theme application scene and are used for Prompt construction in the text-to-image process. For example, in the theme of mountaineering and camping, the scene words include hiking, crossing grassland, and camping and picnic;

[0137] (2) According to the theme words, category information (i.e., commodity attribute information), and scene information corresponding to the theme information, a prompt for generating a banner image prompt is constructed, and the prompt is input into a text-to-text large model for processing to obtain a banner image prompt required for generating a banner image.

[0138] (3) The banner image prompt is input into a text-to-image large model for image generation to obtain a banner image corresponding to the theme. The banner image prompt is a prompt set according to the parameters of the text-to-image large model.

[0139] It should be noted that the generated banner image can be manually audited, and in the case of passing the audit, it is stored in a database for subsequent email assembly.

[0140] Figure 7 is a schematic diagram of a banner image in a data processing method according to an embodiment of the present specification; based on Figure 7 It can be seen that the banner of the theme of high-end tea sets can refer to FIG. A in Figure 7 The banner of the theme of outdoor camping can refer to FIG. B in Figure 7 The banner of the theme of romantic marriage planning can refer to FIG. C in Figure 7

[0141] Step 208: Based on the recommended task theme image and the commodity recommendation information card, a commodity recommendation information block corresponding to the commodity recommendation task is generated.

[0142] Specifically, the recommended task theme image and the commodity recommendation information card are combined to generate a commodity recommendation information block corresponding to the commodity recommendation task.

[0143] Among them, the information block refers to a to-be-displayed information set composed of multimedia information, and the multimedia information contained in the to-be-displayed information set is arranged according to a set distribution rule; the commodity recommendation information block in the embodiment is composed of a recommended task theme image and a commodity recommendation information card, and the recommended task theme image and the commodity recommendation information card are located at a set position in the information block, so as to display the commodity recommendation content to the user.​

[0144] In one or more embodiments provided in the specification, after generating the product recommendation information block corresponding to the product recommendation task based on the recommended task theme image and the product recommendation information card, the method further comprises:

[0145] generating a product recommendation email based on the product recommendation information block and sending the product recommendation email to the client of the target user.

[0146] Specifically, the method can embed the product recommendation information block as part of the email into the product recommendation email, thereby obtaining a product recommendation email (i.e., a marketing email) capable of product recommendation. Then, the product recommendation email is sent to the client of the target user, thereby making product recommendations to the target user.

[0147] As can be known from the above embodiments, the method significantly improves the quality and effectiveness of the product recommendation email through intelligent and automated means, providing a high-efficiency, accurate, and user-friendly marketing tool for enterprises. By providing valuable product recommendation information to users, not only can the satisfaction and loyalty of users be increased, but also long-term and stable customer relationships can be built.

[0148] In one or more embodiments provided in the specification, before sending the product recommendation email to the client of the target user, the method further comprises:

[0149] processing the user information of the target user using a preference detection model to obtain user preference information of the target user, and constructing a user portrait of the target user based on the user preference information and the user information;

[0150] identifying a target product recommendation task associated with the user portrait from a plurality of product recommendation tasks using an information processing model;

[0151] The method of sending the product recommendation email to the client of the target user comprises:

[0152] sending the product recommendation email corresponding to the target product recommendation task to the client of the target user, so that the target user obtains a product purchase interface displayed in the client based on a product purchase link in the product recommendation email, wherein the product purchase link is a product purchase link of the target product, and the product purchase interface is used to purchase the target product.

[0153] Preferably, the preference detection model can be understood as a model for detecting the preferences of the target user, for example, the preference detection model can be a large model, a deep learning model, etc.

[0154] The commodity purchase interface can be understood as a human-computer interaction interface for purchasing target commodities, for example, a webpage or an application interface.

[0155] In the above example, in order to accurately recommend commodities to users, the method can analyze user behavior information and user personal information by using a large model to obtain user preferences, and construct a user portrait by using the user preferences and user personal information. After constructing the user portrait, the large model can be used to determine a recommendation theme that meets the user portrait from multiple themes. Then, based on the fatigue control strategy, the marketing email of the recommendation theme is sent to the user for promotion.

[0156] After receiving the marketing email, the client will display the marketing email to the user. When the user triggers (clicks, selects, or other operations) the commodity purchase link in the marketing email, a request for obtaining a commodity purchase interface will be triggered. Based on the interface obtaining request, the corresponding commodity purchase interface is requested from the server (such as an online shopping platform or an online shopping system) corresponding to the commodity purchase link. After receiving the commodity purchase interface sent by the server, the commodity purchase interface is displayed to the target user through the client, so that the user can perform a purchase operation on the target commodity based on the commodity purchase interface.

[0157] Based on the content of the above embodiment, the method can customize the generation of commodity recommendation information blocks according to the user's historical behavior, preferences, and other related data. This personalized recommendation mechanism makes the content of the marketing email more consistent with the interests and needs of the target user, thereby improving the user's attention and click rate on the email content. Moreover, the automated email generation process reduces the need for human intervention, greatly improving the efficiency of creating and sending commodity recommendation emails. Enterprises can quickly send highly relevant marketing emails to a large number of target users according to a preset schedule or trigger conditions (such as user registration, purchase behavior, etc.), which helps to quickly respond to market changes and seize sales opportunities.

[0158] One or more embodiments of the specification provide a data processing method, after obtaining the multi-modal commodity data of the target commodity associated with the commodity recommendation task, in order to improve the generation efficiency and quality of the recommendation data, a plurality of models can be used for processing; first, the multi-modal commodity data can be adjusted by using a multi-modal model, and the commodity recommendation information card corresponding to the target commodity is generated according to the adjustment result; second, the task theme information of the commodity recommendation task is processed by using an information processing model to obtain the image generation information corresponding to the commodity recommendation task; finally, the image generation model generates the recommendation task theme image according to the image generation information; the automatic processing of the recommendation data is realized by using multiple models, thereby improving the generation efficiency of the recommendation task theme image and the commodity recommendation information card, and the fine data processing operation of multiple steps can obtain the recommendation task theme image and the commodity recommendation information card with high quality; then, based on the recommendation task theme image and the commodity recommendation information card, the high-quality commodity recommendation information block corresponding to the commodity recommendation task can be quickly generated.

[0159] The following description is combined with the Figure 8 Taking the application of the data processing method provided by the specification in the marketing email generation scene as an example, the data processing method is further described. Among them, Figure 8 A process flow diagram of a data processing method provided by one embodiment of the specification is shown, which specifically includes the following steps.

[0160] Step 802: define a theme knowledge base.

[0161] Specifically, the content stored in the theme knowledge base comes from current hot topics, local holidays, social media hot topics, etc. The theme can be collected in real time by external data (automatically supplemented by search, etc.); At the same time, the operation personnel can also manually supplement through the client.

[0162] For example: the theme includes but is not limited to: Christmas Eve, American Super Bowl event, winter skiing, mountaineering camping, etc.

[0163] It should be noted that in the EDM marketing operation, the theme is the central theme of the marketing email, and each component of the email (title, Banner image, commodity list) needs to be produced according to the theme. For example, in the process of producing the commodity pool for the American Super Bowl event, a batch of commodities are selected as marketing content through keyword matching, category matching, algorithm recommendation and other technical means, and the commodities are generally a combination of multiple commodities across categories, such as team flags, star jerseys, hats, mugs, etc.

[0164] Step 804: Al content production

[0165] Specifically, the AI content production of the method includes two parts: "theme banner image generation" and "theme product selling point information" generation.

[0166] Among them, the theme banner image generation refers to generating a theme promotion image using AI. The AI-generated theme promotion image needs to match the theme and the style needs to conform to the product + e-commerce scene background under the theme category. Therefore, the method needs in-depth research on the design of text-to-image prompt and large model parameters. Each theme contains different categories, so the generation prompt of the theme promotion image is dynamically changed and cannot be set in place at one step. Therefore, the AI large model is used to generate the theme promotion image prompt, and then the text-to-image model is used to generate the banner image.

[0167] Among them, the theme product selling point information refers to product information reconstruction. The main direction of product information reconstruction is to extract the core selling points of the product. Through concise information expression, the user is provided with more intuitive and easy-to-read product content, so that the user can understand the product features at a glance. Through scene scripts, use experience, etc., the user's interest in the product is stimulated, thereby improving the click conversion of the product in the email.

[0168] For theme product selling point information, the specific execution mode is as follows:

[0169] 1. Use a large model to generate theme category words, product keywords, etc. under the theme, which are used to filter specific product lists.

[0170] 2. Based on the theme category words, the product is filtered to obtain a product list and obtain the description information (title, detailed content) of the product in the list.

[0171] 3. Based on the prompt constructed based on the product title and product attributes (detailed content), and in the prompt, multiple tasks such as selling point extraction, keyword selection, and / or creative copy can be specified at the same time.

[0172] 4. Input the prompt into the large model, and use the large model to rewrite the product title and detailed content to obtain the rewritten product title and detailed content (i.e. theme product selling point information).

[0173] For theme banner image generation, the specific execution mode is as follows

[0174] 1. Use a large model to analyze the theme information and generate theme scene words. The scene words are detailed descriptions of the application scenarios of the theme, which are used for prompt construction in the text-to-image process. For example, in the theme of mountaineering and camping, the scene words are hiking, crossing grassland, and camping picnic.

[0175] 2. Construct a prompt for generating the banner prompt according to the subject information corresponding to the subject word, category information, and scene information, and input the prompt into the text-to-image large model for processing to obtain the banner prompt required for generating the banner.

[0176] 3. Input the banner prompt into the text-to-image large model for image generation to obtain the banner corresponding to the subject. The banner prompt is a prompt set according to the parameters of the text-to-image large model.

[0177] It should be noted that the generated banner can be manually audited, and in the case of passing the audit, it is stored in the database for subsequent mail assembly.

[0178] Step 806: Mail assembly

[0179] Specifically, the execution mode of mail assembly is:

[0180] 1. Filter the product images, specifically:

[0181] (1) Input the picture into the visual deep learning model for processing.

[0182] (2) Use multiple content detection modules in the visual deep learning model to respectively evaluate the image quality score, text audit, embedded text detection, aesthetic and emotional detection of the product picture, and according to the result detection threshold value output by the visual deep learning model.

[0183] The specific detection method is:

[0184] Input the product image into the visual deep learning model for embedded text detection to obtain a text probability, and determine whether the text probability is <= 0.0017. If yes, it is determined that no text is contained, and text audit is performed. If not, it is determined that text is contained, and the detection is not passed.

[0185] Input the product image into the visual deep learning model for text audit to obtain a result representing whether there is illegal text in the image. In the case where it is determined according to the result that there is no illegal text, perform audit and emotional detection. If it is determined according to the result that there is illegal text, it is determined that the detection is not passed.

[0186] Input the product image into the visual deep learning model for aesthetic and emotional detection to obtain a result representing the aesthetic and emotional representation of the product image. In the case where it is determined according to the result that there is no abnormality in aesthetic and emotion, perform the operation of evaluating the image quality score. If it is determined according to the result that there is an abnormality in aesthetic and emotion, it is determined that the detection is not passed.

[0187] It should be noted that aesthetic and sentiment detection is completed by a multimodal large model. Product images are input into the large model to determine the image content. Aesthetics and sentiment are one of the scoring items. A 10-point system is currently used, and low-scoring images will be filtered out.

[0188] Input the product image into the visual deep learning model to evaluate the image quality score and obtain the product image quality score. Subsequently, product images with an image quality score greater than or equal to 0.6 can be determined as product images that have passed the pre-screening;

[0189] (3) Use image quality scores to rank pre-screened product images, making it easier to select a better product image.

[0190] 3. Determine the target product corresponding to the theme, and the selling point information of the theme product corresponding to the target product (i.e. product copy), the best product image, and the product purchase link.

[0191] 4. Combine the product copy, best image, and purchase link of the target product into a business card to obtain a promotional product card.

[0192] 5. Generate marketing emails based on product cards and banner images embedded in emails

[0193] Step 808: AI intelligent matching.

[0194] 1. Use large models to analyze user behavior information and personal information to obtain user preferences;

[0195] 2. Use user preferences and personal information to build user profiles;

[0196] 3. Use the big model to determine recommended topics that match the user profile from multiple topics.

[0197] Step 810: Email sending

[0198] Based on the fatigue control strategy, a marketing email with the recommended topic is sent to the user for promotion.

[0199] Based on the above steps, the data processing method provided by the present specification provides a marketing email production system based on AIGC content generation. Based on the system, AI-generated theme promotion big pictures, AI-reconstructed commodity expression, and AI-selected high-quality product pictures are realized. Among them, the AI-generated theme promotion big picture refers to that the picture is generated by a large model, the picture background conforms to the e-commerce special effect, the theme background is prominent, the commodity relevance under the theme category is good, and the picture element composition is good. The picture output process is fully AI, reducing the cost of manual intervention. AI reconstruction of commodity expression refers to using a text-to-text large model to simplify information content, forming a title by extracting selling point keywords, efficiently providing users with more intuitive and easy-to-read commodity content, allowing users to understand the characteristics of the commodity at a glance, stimulating users' interest in the commodity through scene copy and use experience, and thus improving the click conversion of commodities in the email. AI selects high-quality product pictures: due to the uneven quality of commodity pictures, AI picture understanding ability is applied to quickly select high-quality commodity pictures, and the theme of the email is associated to make the commodity pictures and the theme style more consistent and relevant.

[0200] Based on the marketing email production system, the traffic conversion of the EDM channel is further improved, the application of AIGC technology is explored to strengthen the expression of marketing content, more efficient email information is provided to users, and the email click-in is improved. On the theme promotion picture, the picture background generated by AIGC highlights the theme scene and has high relevance; the theme variety in the picture is rich and is not affected by the product pictures in the station, and the large model also has good performance in color matching and lens position. In terms of commodity information expression, the business card title is simplified, the original business card uses the business details title, the text is more and the relevance to the theme is not high, and the key words are not fully displayed on the page. The title extracted by AI highlights the characteristics of the commodity, which is composed of 5-6 core words, and the selling points are obvious; the business card details are reconstructed, the use scene copy of the commodity is generated by AI, the selling points in the actual scene are highlighted, the selling point keywords in the title are echoed, and the relevance of the content to the theme is high.

[0201] In summary, in the production process of marketing emails, the system applies AIGC technology of multi-modal large models to realize AI of the whole process of email content, and the technical effects achieved include: 1. Through the AI text-to-image capability, automatically generate theme promotion pictures to improve the visual aesthetics of the theme background and enrich the theme product categories, breaking through the limitations of traditional manual production. 2. Through the AI text-to-text capability, reconstruct the expression of product information. The title part is more concise after being refined by the large model, highlighting the core selling points of the product; the details part generates selling point keywords and use scenario scripts according to product information, strengthening the theme consistency and attractiveness of the content. 3. Apply the large model image-text understanding capability to optimize the product image screening process, filter low-quality content, and select high-quality product images. This scheme uses AIGC technology to reconstruct the content production process of marketing emails, improves the richness and effectiveness of the content, and brings substantial service improvement and user experience optimization.

[0202] Corresponding to the method embodiments described above, the present specification also provides data processing device embodiments, Figure 8 A structural schematic diagram of a data processing device according to an embodiment of the present specification is shown. As shown in the figure, Figure 8 The device comprises:

[0203] A data determination module configured to determine a product recommendation task and obtain multi-modal product data of a target product associated with the product recommendation task;

[0204] An information card determination module configured to adjust the multi-modal product data using a multi-modal model, and generate a product recommendation information card corresponding to the target product according to the adjustment result;

[0205] An image determination module configured to process task theme information of the product recommendation task using an information processing model, obtain image generation information corresponding to the product recommendation task, and generate a recommendation task theme image according to the image generation information using an image generation model;

[0206] An information block determination module configured to generate a product recommendation information block corresponding to the product recommendation task based on the recommendation task theme image and the product recommendation information card.

[0207] Optionally, the image generation information is image generation prompt text;

[0208] The image determination module is further configured to:

[0209] process the task theme information of the product recommendation task using a language processing model to obtain product attribute information and theme scene information corresponding to the task theme information, wherein the theme scene information is used to describe the application scenario corresponding to the task theme information;

[0210] generate, according to the task theme information, the commodity attribute information and the theme scene information, an image generation prompt text corresponding to the commodity recommendation task by using the information processing model;

[0211] generate, according to the image generation prompt text, a recommendation task theme image corresponding to the commodity recommendation task by using an image generation model.

[0212] Optionally, the information card determination module is further configured to:

[0213] determine a commodity description text and commodity visual data from the multi-modal commodity data;

[0214] rewrite the commodity description text according to the task theme information by using a text processing model to obtain a commodity recommendation text of the target commodity;

[0215] evaluate the commodity visual data by using a data evaluation model to obtain a visual evaluation result, and determine recommendation visual data of the target commodity from the commodity visual data based on the visual evaluation result;

[0216] generate the commodity recommendation information card corresponding to the target commodity based on the commodity recommendation text and the recommendation visual data.

[0217] Optionally, the commodity visual data is a plurality of commodity images, and the recommendation visual data is a commodity recommendation image.

[0218] The information card determination module is further configured to:

[0219] evaluate the plurality of commodity images by using a text detection unit, a risk detection unit, an aesthetic detection unit and a quality detection unit in the data evaluation model to obtain an image score corresponding to each commodity image;

[0220] select a commodity recommendation image of the target commodity from the plurality of commodity images based on the image score.

[0221] Optionally, the image score is an image quality score.

[0222] The information card determination module is further configured to:

[0223] input the plurality of commodity images into the data evaluation model, perform text detection on the plurality of commodity images by using the text detection unit, perform risk detection on the plurality of commodity images by using the risk detection unit, and perform aesthetic detection on the plurality of commodity images by using the aesthetic detection unit;

[0224] In a case that the target commodity image passes the text detection, the risk detection and the aesthetic detection, the quality detection unit is utilized to perform quality detection on the target commodity image to obtain an image quality score of the target commodity image, wherein the target commodity image is any one of the plurality of commodity images.

[0225] Optionally, the information card determination module is further configured to:

[0226] determine a commodity purchase link of the target commodity from the modal commodity data;

[0227] combine the commodity recommendation text, the recommendation visual data and the commodity purchase link to obtain the commodity recommendation information card corresponding to the target commodity.

[0228] Optionally, the data determination module is further configured to:

[0229] obtain task theme information corresponding to the commodity recommendation task from a theme information storage unit, and process the task theme information by using a language processing model to obtain commodity attribute information corresponding to the task theme information;

[0230] determine the target commodity associated with the commodity recommendation task from a commodity data storage unit based on the commodity attribute information, and obtain the multi-modal commodity data of the target commodity.

[0231] Optionally, the email sending module of the data processing apparatus is configured to:

[0232] generate a commodity recommendation email based on the commodity recommendation information block, and send the commodity recommendation email to a client of a target user.

[0233] Optionally, the task processing module of the data processing apparatus is configured to:

[0234] process user information of the target user by using a preference detection model to obtain user preference information of the target user, and construct a user portrait of the target user based on the user preference information and the user information;

[0235] identify a target commodity recommendation task associated with the user portrait from a plurality of commodity recommendation tasks by using an information processing model;

[0236] the email sending module is further configured to:

[0237] The target user's client is sent a commodity recommendation email corresponding to the target commodity recommendation task, so that the target user obtains a commodity purchase interface displayed in the client based on a commodity purchase link in the commodity recommendation email, wherein the commodity purchase link is a commodity purchase link of the target commodity, and the commodity purchase interface is used to purchase the target commodity.

[0238] One or more embodiments of the present specification provide a data processing apparatus, after obtaining the multi-modal commodity data of the target commodity associated with the commodity recommendation task, in order to improve the generation efficiency and quality of the recommendation data, a plurality of models can be used for processing; first, the multi-modal commodity data can be adjusted by using a multi-modal model, and the commodity recommendation information card corresponding to the target commodity is generated according to the adjustment result; second, the task theme information of the commodity recommendation task is processed by using an information processing model to obtain the image generation information corresponding to the commodity recommendation task; finally, the image generation model generates the recommendation task theme image according to the image generation information; the automatic processing of the recommendation data is realized by using a plurality of models, thereby improving the generation efficiency of the recommendation task theme image and the commodity recommendation information card, and the fine data processing operation of a plurality of steps can obtain the recommendation task theme image and the commodity recommendation information card with high quality; then, based on the recommendation task theme image and the commodity recommendation information card, the high-quality commodity recommendation information block corresponding to the commodity recommendation task can be quickly generated; thereby realizing the efficient generation of the promotion data of the commodity; avoiding the problem that the manual editing method is used to design the corresponding promotion data for the commodity, resulting in low generation efficiency of the promotion data and unable to meet the needs of marketing promotion.

[0239] The above is a schematic scheme of the data processing apparatus of the present embodiment. It should be noted that the technical scheme of the data processing apparatus belongs to the same concept as the technical scheme of the data processing method described above, and the details of the technical scheme of the data processing apparatus which are not described in detail can be referred to the description of the technical scheme of the data processing method.

[0240] Figure 9 A structural block diagram of a computing device 900 according to an embodiment of the present specification is shown. The components of the computing device 900 include but are not limited to a memory 910 and a processor 920. The processor 920 is connected to the memory 910 through a bus 930, and a database 950 is used to save data.

[0241] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. 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 940 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.

[0242] In one embodiment of the present specification, the above components of the computing device 900 and Figure 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 9 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.

[0243] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a 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 a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0244] The processor 920 is configured to execute the following computer-executable instructions, which implement the steps of the above-mentioned data processing method when executed by the processor.

[0245] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts among the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, the computing device embodiment is described simply because it is basically similar to the data processing method embodiment, and the relevant part can be referred to the description of the data processing method embodiment.

[0246] An embodiment of the specification also provides a computer readable storage medium storing computer programs / instructions, which are executed by a processor to implement the steps of the above data processing method.

[0247] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts among the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, the computer readable storage medium embodiment is described simply because it is basically similar to the data processing method embodiment, and the relevant part can be referred to the description of the data processing method embodiment.

[0248] An embodiment of the specification also provides a computer program product including computer programs / instructions, which are executed by a processor to implement the steps of the above data processing method.

[0249] The above is a schematic scheme of a computer program product of the embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the above data processing method belong to the same concept, and the details of the technical scheme of the computer program product which are not described in detail can be referred to the description of the technical scheme of the data processing method.

[0250] The above describes specific embodiments of the specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps in a claim can be performed in an order different than the order in which the acts or steps are recited and still accomplish the desired results. Also, the process depicted in the accompanying figures does not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0251] The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, software distribution medium, etc. It should be noted that the computer readable medium can include appropriate additions or deletions according to the requirements of patent practice, for example, according to the patent practice in some regions, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0252] It should be noted that, for the foregoing method embodiments, in order to facilitate description, they are all expressed as a combination of a series of actions, but those skilled in the art should know that the embodiments of the present specification are not limited by the order of the described actions, because according to the embodiments of the present specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of the present specification.

[0253] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0254] The preferred embodiments of the present specification disclosed above are only used to help explain the present specification. The alternative embodiments do not describe all the details and limit the invention to the specific embodiments described. Obviously, according to the content of the embodiments of the present specification, many modifications and changes can be made. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of the present specification, so that those skilled in the art can well understand and use the present specification. The present specification is limited only by the claims and their entire scope and equivalents.

Claims

1. A data processing method, comprising: Determine a product recommendation task, and obtain multimodal product data of a target product associated with the product recommendation task; Adjusting the multimodal product data using a multimodal model, and generating a product recommendation information card corresponding to the target product based on the adjustment result; Using an information processing model to process the task theme information of the product recommendation task, obtaining image generation information corresponding to the product recommendation task, and using an image generation model to generate a recommendation task theme image based on the image generation information; Based on the recommendation task theme image and the product recommendation information card, a product recommendation information block corresponding to the product recommendation task is generated.

2. The data processing method according to claim 1, wherein the image generation information is image generation prompt text; The step of processing the task theme information of the product recommendation task using the information processing model to obtain image generation information corresponding to the product recommendation task, and generating a recommendation task theme image based on the image generation information using the image generation model includes: Processing the task theme information of the product recommendation task using a language processing model to obtain product attribute information and theme scenario information corresponding to the task theme information, wherein the theme scenario information is used to describe the application scenario corresponding to the task theme information; generating an image generation prompt text corresponding to the product recommendation task using the information processing model according to the task theme information, the product attribute information, and the theme scene information; An image generation model is used to generate prompt text according to the image, and a recommendation task theme image corresponding to the product recommendation task is generated.

3. The data processing method according to claim 1, wherein the multimodal model comprises a text processing model and a data evaluation model; The adjusting the multimodal product data using a multimodal model and generating a product recommendation information card corresponding to the target product according to the adjustment result includes: Determining product description text and product visual data from the multimodal product data; Using a text processing model, rewriting the product description text according to the task subject information to obtain a product recommendation text for the target product; Evaluating the product visual data using a data evaluation model to obtain a visual evaluation result, and determining recommended visual data of the target product from the product visual data based on the visual evaluation result; The product recommendation information card corresponding to the target product is generated based on the product recommendation text and the recommendation visual data.

4. The data processing method according to claim 3, wherein the product visual data is a plurality of product images, and the recommended visual data is a product recommendation image; The step of evaluating the product visual data using a data evaluation model to obtain a visual evaluation result, and determining recommended visual data of the target product from the product visual data based on the visual evaluation result, includes: Performing image evaluation on the plurality of product images using manual text detection, text risk detection, aesthetic detection, sentiment detection, and quality detection in the data evaluation model to obtain an image score corresponding to each product image; Based on the image score, a product recommendation image of the target product is selected from the plurality of product images.

5. The data processing method according to claim 4, wherein the step of performing image evaluation on the plurality of product images using manual text detection, text risk detection, aesthetic detection, sentiment detection, and quality detection in the data evaluation model to obtain an image score corresponding to each product image comprises: Inputting a target product image into the data assessment model for manual text detection and text risk detection, wherein the target product image is any one of the multiple product images; If it is determined that the target product image passes the manual text detection and the text risk detection, inputting the target product image into the data evaluation model for aesthetic detection and sentiment detection; When it is determined that the target product image passes the aesthetic detection and the emotion detection, the target product image is input into the data evaluation model for quality detection to obtain an image quality score of the target product image, and the image quality score is determined as the image score corresponding to each product image.

6. The data processing method according to claim 3, wherein generating the product recommendation information card corresponding to the target product based on the product recommendation text and the recommendation visual data comprises: Determining a product purchase link for the target product from the modal product data; The product recommendation text, the recommendation visual data and the product purchase link are combined to obtain the product recommendation information card corresponding to the target product.

7. The data processing method according to any one of claims 1 to 6, wherein determining a product recommendation task and obtaining multimodal product data of a target product associated with the product recommendation task comprises: Selecting task topic information from a plurality of candidate topic information stored in a topic information storage unit, and generating a product recommendation task based on the task topic information; Processing the task subject information using a language processing model to obtain commodity attribute information corresponding to the task subject information; Based on the product attribute information, the target product associated with the product recommendation task is determined from a product data storage unit, and the multimodal product data of the target product is acquired.

8. The data processing method according to any one of claims 1 to 6, further comprising: after generating a product recommendation information block corresponding to the product recommendation task based on the recommendation task theme image and the product recommendation information card; A product recommendation email is generated based on the product recommendation information block, and the product recommendation email is sent to a client of a target user.

9. The data processing method according to claim 8, before sending the product recommendation email to the client of the target user, further comprising: Processing the user information of the target user using a preference detection model to obtain user preference information of the target user, and constructing a user profile of the target user based on the user preference information and the user information; Using an information processing model, identifying a target product recommendation task associated with the user profile from a plurality of product recommendation tasks; The sending of the product recommendation email to the client of the target user includes: The product recommendation email corresponding to the target product recommendation task is sent to the client of the target user, so that the target user obtains the product purchase interface displayed in the client based on the product purchase link in the product recommendation email, wherein the product purchase link is the product purchase link of the target product, and the product purchase interface is used to purchase the target product.

10. A data processing device comprising: a data determination module configured to determine a product recommendation task and obtain multimodal product data of a target product associated with the product recommendation task; an information card determination module, configured to adjust the multimodal product data using a multimodal model, and generate a product recommendation information card corresponding to the target product according to the adjustment result; an image determination module configured to process the task theme information of the product recommendation task using an information processing model, obtain image generation information corresponding to the product recommendation task, and generate a recommendation task theme image based on the image generation information using an image generation model; The information block determination module is configured to generate a product recommendation information block corresponding to the product recommendation task based on the recommendation task theme image and the product recommendation information card.

11. A computing device comprising: memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium storing a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the steps of the method according to any one of claims 1 to 9.

13. A computer program product comprising a computer program / instruction, which implements the steps of the method according to any one of claims 1 to 9 when executed by a processor.

Images