Information generation method, device, computer equipment and medium

By obtaining product images and generating target description information, the problem of low efficiency in product description information generation in e-commerce systems is solved, efficient automatic generation is achieved, labor costs are reduced, and the market performance of products is improved.

CN117593069BActive Publication Date: 2024-11-22XINGIN INFORMATION TECH (SHANGHAI) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310890422.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-07-19
Publication Date
2024-11-22
Estimated Expiration
2043-07-19

AI Technical Summary

Technical Problem

In the e-commerce system, it takes a lot of manpower and energy to add description information to products, resulting in inefficient generation.

Method used

By obtaining product images, extracting product information and filling in preset prompts to generate templates, generating target description information, and directly automatically generating product description information to avoid tedious manual editing.

Benefits of technology

It improves the efficiency of product description information generation, reduces the cost of merchants to list and market products, and improves the quality of description information, increasing the click rate and order rate of products.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117593069B_ABST
    Figure CN117593069B_ABST
Patent Text Reader

Abstract

The embodiments of the present application disclose an information generation method, apparatus, computer equipment and medium, wherein the method comprises: obtaining a product image of a product, and obtaining product information of the information-generated product based on the product image of the information-generated product; the information-generated product information comprises a product name and product characteristics; obtaining a preset prompt generation template, filling the information-generated product information into the information-generated prompt generation template, and using the filled prompt generation template as prompt text of the information-generated product; using the information-generated prompt text as a prompt to generate reference description information of the information-generated product, and determining target description information of the information-generated product based on the information-generated reference description information; the information-generated target description information is used for online publishing on a product platform; and the generation efficiency of the description information of the product can be effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an information generation method, apparatus, computer equipment and medium. Background Art

[0002] In e-commerce systems, merchants usually need to add corresponding descriptive information to products when listing them to attract customers. Currently, product descriptions or recommendations are usually edited manually. This manual editing method requires a lot of manpower and energy to maintain product description information. Therefore, how to improve the efficiency of generating product description information has become a current research hotspot. Summary of the invention

[0003] The embodiments of the present application provide an information generation method, apparatus, computer equipment and medium, which can effectively improve the generation efficiency of product description information.

[0004] The first aspect of the embodiments of the present application discloses a method for generating information, the method comprising:

[0005] Acquire a product image of a product, and acquire product information of the product based on the product image of the product; the product information includes a product name and product characteristics;

[0006] Obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the prompt text of the product;

[0007] The reference description information of the commodity is generated using the prompt text as a prompt, and the target description information of the commodity is determined based on the reference description information; the target description information is used for online publishing on the commodity platform.

[0008] A second aspect of an embodiment of the present application discloses an information generating device, the device comprising:

[0009] An acquisition unit, configured to acquire a commodity image of a commodity, and acquire commodity information of the commodity based on the commodity image of the commodity; the commodity information includes a commodity name and commodity characteristics;

[0010] A filling unit, used for obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the prompt text of the product;

[0011] A generating unit is used to generate reference description information of the commodity using the prompt text as a prompt, and determine target description information of the commodity based on the reference description information; the target description information is used to be published online on a commodity platform.

[0012] A third aspect of an embodiment of the present application discloses a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method of the first aspect above.

[0013] A fourth aspect of an embodiment of the present application discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the method of the first aspect.

[0014] The fifth aspect of the embodiments of the present application discloses a computer program product or a computer program, wherein the computer program product or the computer program comprises program instructions, and when the program instructions are executed by a processor, the method of the first aspect is implemented.

[0015] In an embodiment of the present application, a product image of a product is obtained, and product information of the product generated by the information is obtained based on the product image of the product generated by the information; the product information generated by the information includes the product name and product characteristics; a preset prompt generation template is obtained, and the product information generated by the information is filled into the information generation prompt generation template, and the filled prompt generation template is used as the prompt text of the information generation product; the reference description information of the information generation product is generated using the information generation prompt text as a prompt, and the target description information of the information generation product is determined based on the information generation reference description information; the information generation target description information is used for online publishing on the product platform. Through the above method, the description information of the product can be automatically generated directly based on the product image of the product. Compared with the manual editing method, the tedious operation and a lot of time-consuming work caused by the manual editing of the text of the description information can be effectively avoided, without taking up a lot of manpower and energy, and effectively improving the efficiency of generating the description information. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 It is a schematic diagram of the architecture of an information generation system provided in an embodiment of the present application;

[0018] Figure 2 It is a flowchart of an information generation method provided in an embodiment of the present application;

[0019] Figure 3This is a flow chart of determining whether there is a need to generate description information for a product, provided in an embodiment of the present application;

[0020] Figure 4 It is a flowchart of another information generation method provided in an embodiment of the present application;

[0021] Figure 5 It is a structural schematic diagram of an information generating device provided in an embodiment of the present application;

[0022] Figure 6 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] The embodiment of the present application proposes an information generation scheme, which can generate target description information corresponding to a product based on the product image of the product, and the target description information can be used for online publishing on the product platform. The principle of the information generation scheme is as follows: first, the product information of the product can be obtained based on the product image of the product; the product information can include the product name and product characteristics; and then the target description information of the product can be generated based on the product information. Optionally, a preset prompt generation template can be obtained to fill the product information into the prompt generation template, and the filled prompt generation template is used as the prompt text of the product; then, the reference description information of the product can be generated using the prompt text as a prompt, and the target description information of the product can be determined based on the reference description information.

[0025] Through the above scheme, the description information of the product can be automatically generated directly based on the product image of the product. Compared with the manual editing method, it can effectively avoid the tedious operations and a lot of time-consuming text work caused by manual editing of the description information, without taking up a lot of manpower and energy, and effectively improve the efficiency of generating the description information, thereby effectively reducing the cost of listing and marketing products for merchants. In addition, the corresponding description information can be generated in combination with the prompt text to effectively improve the generation quality of the description information, and high-quality description information can also make the product get a higher probability of click-through rate and order rate.

[0026] In a specific implementation, the execution subject of the above-mentioned information generation scheme may be a computer device, which may be a terminal or a server. The terminal mentioned here may be a smart phone, tablet computer, laptop computer, desktop computer and other devices; the server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, etc.

[0027] For example, when the computer device is a server, the embodiment of the present application provides an information generation system, such as Figure 1 As shown, the information generation system may include at least one terminal and at least one server; the terminal may obtain a product image of a product and upload the obtained product image of the product to the server (i.e., a computer device) so that the server may use an information generation scheme to determine the target description information of the product.

[0028] It should be noted that the data related to user information in the embodiments of the present application are all data authorized by the user.

[0029] Based on the information generation scheme provided above, the embodiment of the present application provides an information generation method, which can be executed by the computer device mentioned above. Figure 2 , the information generation method includes but is not limited to the following steps:

[0030] S201, obtaining a product image of a product, and obtaining product information of the product based on the product image of the product.

[0031] Among them, the commodity can be a commodity to be published on the commodity platform, and the commodity can be any type of commodity, such as books, food, clothing, etc. The commodity information can include the commodity name and commodity characteristics; for example, assuming that the commodity is a book, the commodity name of the commodity can refer to the title of the book, and the commodity characteristics can refer to the characteristics of the book. Exemplarily, the commodity name in the commodity information of a book commodity can be "XXX Book Set", and the commodity characteristics can be "Step-by-step Climbing Learning, From 0 to Independent Reading"; the commodity name in the commodity information of a clothing commodity can be "Printed Short-sleeved Shirt", and the commodity characteristics can be "Printed, Breathable".

[0032] In one implementation, when a demand for generating description information for a product is obtained, the operation of obtaining the product image of the product in step S201 may be triggered.

[0033] Optionally, a request for generating description information for a product can be obtained on a computer device to determine the need for generating description information for the product. The generation request can be triggered by a target object (which can be any user) performing a related operation on a user operation interface. For example, when the target object needs to obtain the description information corresponding to the product, it can perform a related operation on the user operation interface output by the terminal used to send a request for generating description information for the product to the computer device. For example, see Figure 3 As shown, a user operation interface may be displayed on the terminal screen of the terminal used by the target object, and the user operation interface may at least include an input area 301 for a product image. If the target object wants to obtain the description information corresponding to the product, the relevant information of the product may be first input in the input area 301 (such as the product image of the product directly, or the link address corresponding to the product image of the product); after the terminal detects that relevant information is uploaded in the input area 301, or detects that a product image is uploaded in the area 301, the product image of the product may be obtained based on the information in the input area 301, and after the terminal obtains the product image of the product, a request for generating the description information of the product may be sent to the computer device, and the computer device may determine that the generation demand for the description information of the product is obtained. Among them, the generation request may carry the product image of the product, so that the computer device may obtain the product image of the product, and further determine the description information of the product based on the product image.

[0034] Optionally, when a generation timer task is triggered, it can be determined that the generation demand for the description information of the product is obtained. For example, a generation timer task can be set, and when the trigger condition in the generation timer task is triggered, it can be determined that the generation demand is obtained. In one embodiment, a large number of product images of products can be stored in a certain designated area, and the trigger condition can be that the current time reaches the preset processing time, or the remaining storage space of the designated storage area exceeds the preset remaining storage space, etc.

[0035] In one implementation, after obtaining the product image of the product, the product information for the product can be further obtained from the product image. The product information can be used to characterize the product. For example, the product information of a product can be one or more of the product name, product characteristics, etc. of the product. In order to more accurately represent a product, in the embodiment of the present application, the product information including the product name and product characteristics is used as an example for specific description.

[0036] Optionally, the specific implementation of obtaining the product information of a product based on the product image of the product may be: the text description information contained in the product image of the product may be first obtained to determine the product information of the product based on the text description information. The text description information may contain a large amount of information about the product, wherein the information may contain relatively important information or unimportant information. In order to ensure the reliability of the information, the required information may be filtered out from the text description information, and the filtered out information may be used as the product information of the product. For example, in an embodiment of the present application, the information to be filtered out may be the product name and product characteristics of the product. Among them, the method of determining the product information from the text description information may be any of the following.

[0037] In one embodiment, the product information of the product can be determined from the text description information based on the reference information identifier. The reference information identifier can be pre-set. The reference information identifier can be used to indicate the product information to be obtained. For example, the reference information identifier can include the product name and product characteristics. Then, the product name and information corresponding to the product characteristics can be obtained from the text description information. The obtained information is the product information of the product.

[0038] In one embodiment, the text description information can be output on a text display interface so that the target object can select the desired product information from the displayed text description information. The text display interface can be output on a terminal screen used by the target object. Based on this, after the target object selects information from the text display interface, the key information selected by the target object from the text description information displayed on the text display interface can be obtained. The key information may include the product name and product characteristics; and the key information is the product information, that is, the key information can be used as the product information of the product.

[0039] In one implementation, obtaining the text description information contained in the product image may be specifically implemented in any of the following ways.

[0040] Method 1: Multimodal machine learning technology can be used to obtain text description information in product images; multimodal machine learning is a process of machine learning using multiple data modalities (such as text, images, voice, etc.). It covers multiple disciplines, such as natural language processing, computer vision, speech recognition, etc. Based on this, it can be seen that in order to obtain the text description information of the product image to be processed, multimodal machine learning can be used to achieve it; in a specific implementation, multimodal feature extraction can be performed on the product image to obtain the multimodal features of the product image, and further, the text description information of the product can be generated based on the multimodal features. It can be understood that a product image can usually contain information in the text modality and information in the image modality, which can facilitate the multimodal feature extraction operation to obtain the features in the text modality and the features in the image modality contained in the product image, and the above-mentioned multimodal features can include features in the text modality and features in the image modality. Among them, the multimodal feature extraction operation can be implemented by using a neural network with a multimodal feature extraction function, and the generation operation of generating text description information based on multimodal features can be implemented by using a neural network with a text generation function. The neural networks used for the multimodal feature extraction operation and the generation operation of generating text description information based on multimodal features can be separate networks or an integrated network. In the embodiments of the present application, the network structure of the neural network is not specifically limited.

[0041] Method 2: You can use OCR (Optical Character Recognition) technology to obtain text description information in product images. OCR technology is a process that determines the shape of a dark and light pattern by detecting it, and then uses character recognition methods to translate the shape into computer text. That is, you can use OCR technology to obtain text in an image. Based on this, it can be realized based on OCR technology. In the specific implementation, you can extract the text content in the product image based on optical character recognition OCR, and use the text content as the text description information of the product.

[0042] Method three, in order to ensure the reliability and accuracy of information acquisition, the above-mentioned multimodal machine learning and OCR technology can also be combined to jointly realize the acquisition of text description information in product images. In the specific implementation, multimodal feature extraction can be performed on the product image to obtain the multimodal features of the product image, and the text description information of the product can be generated based on the multimodal features. For the convenience of description, the text description information here can be referred to as the first text description information; the text content in the product image can also be extracted based on the optical character recognition OCR method, and the text content can be used as text description information. For the convenience of description, the text description information here can be referred to as the second text description information. After obtaining the first text description information and the second description, the first text description information and the second text description information can be combined into the text description information contained in the product image.

[0043] It can be seen that in the embodiment of the present application, the product characteristics of the product can be combined with the multimodal machine learning algorithm and OCR analysis to fully extract the text description information contained in the product image, thereby ensuring the reliability and accuracy of the product information determined based on the text description information.

[0044] S202, obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the product prompt text.

[0045] In one implementation, after obtaining the prompt generation template, the prompt text of the product can be determined based on the prompt generation template and the product information. For example, the product information can be filled into the prompt generation template, and the filled prompt generation template is used as the prompt text of the product.

[0046] For example, assume that the prompt generation template can be "As an online store merchant, generate an informative and convincing description of {product name}, and emphasize that this product has the following characteristics: {product characteristics}". Exemplarily, assume that the product name in the product information of the product is "XXX book set", and the product characteristics are "step-by-step learning, from 0 entry and exit to independent reading"; then the prompt text generated by filling these product information into the prompt generation template is: As an online store merchant, generate an informative and convincing description of {XXX book set}, and emphasize that this product has the following characteristics: {step-by-step learning, from 0 entry and exit to independent reading}. Exemplarily, assume that the product name in the product information of the product is "printed short-sleeved shirt", and the product characteristics are "printed, breathable"; then the prompt text generated by filling these product information into the prompt generation template is: As an online store merchant, generate an informative and convincing description of {printed short-sleeved shirt}, and emphasize that this product has the following characteristics: {printed, breathable}.

[0047] In one implementation, all commodities may correspond to one prompt generation template, or commodities of different commodity types may correspond to different prompt generation templates. The description of the prompt generation template under each commodity type may be more appropriate to the description of the commodity type, thereby ensuring that the prompt text generated based on the prompt generation template is more accurate, improving the reliability and accuracy of the prompt text generation, and further ensuring the reliability and accuracy of the subsequent description information generated in combination with the prompt text. In this case, prompt generation templates corresponding to various commodity types may be pre-configured. For example, assuming that the commodity types include books, food, clothing, etc., the prompt generation module may include prompt generation templates corresponding to books, food, clothing, etc.

[0048] Based on this, it can be known that the specific implementation method of obtaining the preset prompt generation template can be: first obtain the product type of the product, and then further obtain the prompt generation template matching the product type. In one embodiment, the prompt generation templates corresponding to various different product types can be pre-associated and stored in a preset storage area, and then the prompt generation template matching the product type can be obtained from the preset storage area, so that the product information can be filled into the prompt generation template corresponding to the product type.

[0049] Among them, the commodity type of the commodity can be obtained by performing type recognition on the commodity image. It can be understood that the commodity image can usually contain an image corresponding to the commodity itself. For example, the commodity image of a book commodity can display books, the commodity image of a clothing commodity can display clothing, and the commodity image of a food commodity can display food. Then, the commodity type corresponding to the commodity can be obtained by performing type recognition on the commodity image. In this case, the commodity image can be type-recognized by calling a neural network with a type recognition function to obtain the commodity type of the commodity. Alternatively, the above-mentioned generation request can carry indication information for indicating the commodity type. After obtaining the generation request, the commodity type of the commodity can be determined directly based on the indication information in the generation request.

[0050] It can be seen from the above method that the embodiment of the present application can obtain different prompt texts for commodities of different commodity types, so as to effectively improve the flexibility, reliability and accuracy of prompt text generation.

[0051] In one implementation, the same product in different information release scenarios may correspond to a prompt generation template, or the same product in different information release scenarios may correspond to different prompt generation templates. The description of the prompt generation template in each information release scenario may be more in line with the description of the information release scenario, thereby ensuring that the prompt text generated based on the prompt generation template is more accurate, improving the reliability and accuracy of the prompt text generation, and thus ensuring the reliability and accuracy of the subsequent description information generated in combination with the prompt text. In this case, prompt generation templates corresponding to various information release scenarios may be pre-configured. For example, assuming that the information release scenario includes a product launch scenario, a product note scenario, etc., the prompt generation module may include prompt generation templates corresponding to the product launch scenario, the product note scenario, etc., respectively. Among them, the product launch scenario can be understood as a scenario in which the product is launched online and published on the product platform, and the product note scenario can be understood as a scenario in which notes are taken for the product after the product is launched. For example, in an e-commerce system, after a product is released, it is usually possible to support the publication of usage notes, sharing notes, etc., and include product links in these notes to promote the product, so that the product can get more click-through conversions and recommendations through these notes. Here, these usage notes, sharing notes, etc. can all be referred to as product notes.

[0052] For example, the prompt generation template in the product launch scenario may be "As an online store merchant, generate an informative and convincing description of {product name}, and emphasize that this product has the following characteristics: {product characteristics}". For example, assuming that the product name in the product information of the product is "XXX book set", and the product characteristics are "step-by-step learning, from zero entry and exit to independent reading"; then the prompt text generated by filling these product information into the prompt generation template in the product launch scenario is: As an online store merchant, generate an informative and convincing description of {XXX book set}, and emphasize that this product has the following characteristics: {step-by-step learning, from zero entry and exit to independent reading}.

[0053] For another example, the prompt generation template in the product note scenario can be "As a recommender, I am now going to post a note about {product name}, and emphasize that this product has the following features: {product features}. Please help me list 10 related titles, plus emoticons, and control it to 20 characters. Only 1-2 emoticons are required, and emoticons should be interspersed in the copy as much as possible." For example, assuming that the product name in the product information of the product is "XXX Book Set", and the product features are "Step-by-step learning, from 0 entry and exit to independent reading"; then the prompt text generated by filling these product information into the prompt generation template in the product note scenario is: As a recommender, I am now going to post a note about {XXX Book Set}, and emphasize that this product has the following features: {Step-by-step learning, from 0 entry and exit to independent reading}. Please help me list 10 related titles, plus emoticons, and control it to 20 characters. Only 1-2 emoticons are required, and emoticons should be interspersed in the copy as much as possible. Alternatively, the prompt generation template in the product note scenario can also be "As a recommender, generate an article to recommend {product name}, attach your own experience of use, and emphasize that this set of books has the following characteristics: {product characteristics}. The article needs to be persuasive, rich in information, provide an example, and be expressed in a familiar way, and some modal particles can be added. Some emoticons can be interspersed between different paragraphs of the article, and the number of emoticons in each paragraph should not exceed 5", etc.

[0054] Based on this, it can be known that the specific implementation method of obtaining the preset prompt generation template can be: first obtain the information release scenario of the product, and further obtain the prompt generation template matching the information release scenario. In one embodiment, the prompt generation templates corresponding to various different information release scenarios can be pre-associated and stored in a preset storage area, and then the prompt generation template matching the information release scenario can be obtained from the preset storage area, so that the product information can be filled into the prompt generation template corresponding to the information release scenario.

[0055] Optionally, here, the information release scenario including product launch and product notes is used as an example for explanation, and the product launch scenario is referred to as the first scenario, and the product note scenario is referred to as the second scenario, then a prompt generation template corresponding to the first scenario and a prompt generation template corresponding to the second scenario are pre-stored. In this case, the specific implementation method of obtaining the preset prompt generation template can be: first obtain the product information release scenario; if the information release scenario is the first scenario, then obtain the prompt generation template matching the first scenario; if the information release scenario is the second scenario, then obtain the prompt generation template matching the second scenario.

[0056] Among them, the information release scenario of the product can be determined based on the indication information carried in the generation request. For example, the generation request mentioned above can carry indication information for indicating the information release scenario. After obtaining the generation request, the information release scenario of the product can be determined directly based on the indication information in the generation request.

[0057] It can be seen from the above method that the embodiment of the present application can obtain different prompt texts for the same product in different information release scenarios, so as to effectively improve the flexibility, reliability and accuracy of prompt text generation.

[0058] In one implementation, different information release scenarios and commodities of different commodity types may correspond to different prompt generation templates. In this case, prompt generation templates corresponding to various information release scenarios and commodity types may be pre-configured. For example, assuming that the information release scenarios include commodity launch, commodity notes, etc., and commodity types include books, clothing, and food, the prompt generation module may include: prompt generation templates corresponding to commodity launch scenarios and books, prompt generation templates corresponding to commodity notes scenarios and books, prompt generation templates corresponding to commodity launch scenarios and clothing, prompt generation templates corresponding to commodity notes scenarios and clothing, etc. That is, one prompt generation template is associated with two restrictions, which are the information release scenario and commodity type.

[0059] Based on this, it can be known that the specific implementation method of obtaining the preset prompt generation template can be: first obtain the information release scenario and product type of the product, and further obtain the prompt generation template that matches the information release scenario and product type. In one embodiment, the prompt generation templates corresponding to various information release scenarios and product types can be pre-associated and stored in a preset storage area, and then the prompt generation template that matches the information release scenario and product type can be obtained from the preset storage area, so that the product information can be filled into the prompt generation template corresponding to the information release scenario and product type.

[0060] Through the above method, it can be known that for different information release scenarios, the embodiments of the present application can provide different prompt generation templates for different types of goods to generate different prompt texts, so that the prompt text is more suitable for the goods, effectively improving the flexibility, reliability and accuracy of prompt text generation.

[0061] S203: Generate reference description information of the product using the prompt text as a prompt, and determine target description information of the product based on the reference description information.

[0062] The target description information can be used for online publishing on the commodity platform, that is, it can be published on the commodity platform so that users using the commodity platform can browse the description information of the commodity. In other words, the target description information can refer to the description information ultimately required by the target object.

[0063] In one implementation, a text generation network can be called to generate reference description information of a product using a prompt text as a prompt. Among them, the text generation network can be a GPT (Generative Pre-trained Transformer) network. The GPT network is a neural network model for natural language processing (NLP) that aims to automatically complete tasks such as translation, filling in the blanks, summarization, and question-answering. GPT belongs to the Transformer architecture family and uses large-scale unsupervised learning and generative pre-training methods to fine-tune various language tasks to achieve high performance. The GPT network has high performance in text generation. The text generation network can also be other networks with text generation functions, which are not specifically limited. The embodiment of the present application takes the text generation network as a GPT network as an example for relevant elaboration. That is, in the embodiment of the present application, the GPT network can be called to generate reference description information of a product using a prompt text as a prompt.

[0064] For example, for a product with a product name of "XXX Book Set" and product characteristics of "Step-by-step learning, from zero entry to independent reading", the reference description information of the product can be as follows:

[0065] "If you are looking for a set of tools that can help your child quickly grow from zero foundation to independent reading, then the XXX book set is a good choice for you.

[0066] The best feature of this set is its step-by-step learning method, which gradually guides children to develop reading habits and skills from simple to complex, from easy to difficult. It not only covers all aspects that children need, including listening, speaking, reading and writing, but also provides children with a large number of exercises and activities, allowing children to master what they have learned through practice.

[0067] In the process of learning to read using the XXX book set, children will gradually improve their language and reading skills, thus reducing the burden on parents in educating their children. Moreover, it is applicable to a wide range of students, whether they are just beginning to learn to read or want to improve their reading level.

[0068] Finally, let's look at the key features of the XXX book set. First, it is a set of step-by-step learning tools that can help children make continuous progress in reading. Second, it includes a large number of exercises and activities, allowing children to master what they have learned through practice. Finally, it has a wide range of applications and is suitable for children of all ages.

[0069] In short, the XXX book set is an educational tool with high quality and practical value, which can help children quickly grow from zero foundation to independent readers. Hurry up and buy a copy to lay a solid reading foundation for your child's future. "

[0070] In the embodiment of the present application, the description information of the product can be automatically generated directly based on the product image of the product. Compared with the manual editing method, it can effectively avoid the tedious operations and a lot of time-consuming text work caused by manual editing of the description information, without taking up a lot of manpower and energy, and effectively improve the efficiency of generating the description information, thereby effectively reducing the cost of listing and marketing products for merchants. In addition, the corresponding description information can be generated in combination with the prompt text to effectively improve the generation quality of the description information, and high-quality description information can also make the product get a higher probability of click-through rate and order rate.

[0071] See also Figure 4 , Figure 4 is a flow chart of another information generation method provided in an embodiment of the present application. The information generation method described in this embodiment can be applied to the above-mentioned computer device, such as Figure 4 As shown, the method may include:

[0072] S401, obtaining a product image of a product, and obtaining product information of the product based on the product image of the product.

[0073] S402, obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the product prompt text.

[0074] The specific implementation of steps S401-S402 may refer to the specific implementation of steps S201-S202, which will not be described again here.

[0075] S403, obtaining a parameter combination when using a text generation network to generate description information.

[0076] Among them, different parameter combinations can be used to generate different description information of a product, that is, for a product, if the parameter combination corresponding to the text generation network is not used, the description information of the product generated is different, that is to say, different reference combinations can generate different description information. The parameter combination may include one or more parameters. The parameters in the parameter combination may be parameters used to control the diversity of generated text. For example, the parameter may include one or more of Temperature, Top-p, Top-k, etc., and may also include other parameters. This is not limited and will not be illustrated one by one here. The embodiment of the present application takes the parameter combination including Temperature, Top-p, and Top-k as an example for relevant explanation. For example, Temperature = 1, Top-p = 0.8, Top-k = 30; for another example, Temperature = 1, Top-p = 0.9, Top-k = 50; and so on.

[0077] Among them, Temperature is a parameter that controls the diversity of generated text. It can affect the randomness and creativity of the text generation network when generating text. The value range of Temperature can be 0 to 1, and the larger the value, the more creative the description information generated by the text generation network for the product. That is, when the Temperature value is low, the description information generated by the text generation network is simpler and more conservative, while when the Temperature value is high, the description information generated by the text generation network is more creative and diverse. In practical applications, the Temperature value can be adjusted according to needs to obtain the best description information.

[0078] Among them, Top-p is also a parameter that controls the diversity of generated text. It differs from Temperature in that Top-p considers a subset with the highest probability when generating each word, so that the sum of the word probabilities in this subset is greater than or equal to p. In this way, during the generation process, the text generation network will only select words from this subset to obtain the final required description information. The value range of Top-p can be 0 to 1, and the larger the value, the more creative the description information generated by the text generation network for the product. That is, when the Top-p value is low, the description information generated by the text generation network is relatively simple and conservative, while when the Top-p value is high, the description information generated by the text generation network is more creative and diverse. Temperature and Top-p can be used in combination to jointly control the diversity and randomness of the generated text. Different combinations of Temperature and Top-p settings will affect the style and diversity of the generated text. In practical applications, these two parameters can be adjusted according to needs to obtain the best description information.

[0079] Among them, Top-k is also a parameter that controls the diversity of text generation. It differs from Top-p in that when Top-k generates each word, it only selects from the words with the top k probability ranking. The value range of Top-k is usually between 1 and the size of the vocabulary. The size of the vocabulary varies depending on the text generation network and language. When k=1, the text generation network only selects from the words with the highest probability, and the generated text will be more conservative and may be repeated. As the value of k increases, the generated text becomes more diverse and random. When k is equal to the vocabulary size, the model will randomly select words from the entire vocabulary, and the generated text will become more random. In practical applications, the value of Top-k can be adjusted according to needs to control diversity while ensuring the quality of generated text.

[0080] Based on this, we can see that Temperature, Top-p, and Top-k can be used in combination to jointly control the diversity and randomness of the generated description information, and different setting combinations will affect the style and diversity of the generated description information. In practical applications, these three parameters can be adjusted according to needs to obtain the best description information.

[0081] In summary, in order to generate product description information using a text generation network, it is necessary to determine a parameter combination to generate product description information under the parameter combination; as mentioned above, different parameter combinations can generate different description information, so a better parameter combination can be selected to ensure the quality of the generated description information. It should be understood that the parameter combination can also be understood as a parameter combination about the prompt text (prompt).

[0082] In one implementation, the parameter combination used when the text generation network is used to generate description information for the current product can be determined by executing the task of generating reference description information for the previous product of the product. That is, when the text generation network performs the task of generating reference description information for a certain product, the parameter combination corresponding to the text generation network can be changed dynamically in real time to ensure the generation effect of the reference description information. Optionally, when it is ensured that the previous product has the maximum resource conversion amount, the parameter combination corresponding to the reference description information is generated, and it is determined as the parameter combination when the text generation network generates description information for the current product. Among them, when it is ensured that the previous product has the maximum resource conversion amount, the parameter combination corresponding to the reference description information can be determined based on the target description information corresponding to the previous product. The principle of determining the parameter combination corresponding to the current product is similar to that of determining the parameter combination corresponding to the next product. The embodiment of the present application is specifically described by taking the determination of the parameter combination corresponding to the next product as an example, as shown in the following step S405.

[0083] Among them, the resource conversion volume may include one or more of the order resource conversion volume under the order dimension and the click resource conversion volume under the click dimension.

[0084] The order resource conversion amount can be determined based on the number of times the product is browsed and the number of times the product is browsed. For example, the order resource conversion amount can be the ratio between the number of times the product is ordered and the number of times the product is browsed. The order resource conversion amount can also be understood as an order conversion rate. It can be understood that the larger the order resource conversion amount, the better the advertising effect of the target description information of the product, and the smaller the order resource conversion amount, the worse the advertising effect of the target description information of the product.

[0085] The click resource conversion amount can be determined based on the number of times the product is clicked and the number of times the product is displayed. For example, the click resource conversion amount can be CTR (click-through rate). For example, the click resource conversion amount can be the ratio between the number of times the product is clicked and the number of times the product is displayed. It can be understood that the larger the click resource conversion amount, the better the advertising effect of the target description information of the product, and the smaller the click resource conversion amount, the worse the advertising effect of the target description information of the product.

[0086] S404, calling the text generation network to generate reference description information of the product using the prompt text as a prompt under the parameter combination.

[0087] Among them, the text generation network can be a GPT network, which is a neural network model for natural language processing (NLP). GPT belongs to the Transformer architecture family, and uses large-scale unsupervised learning and generative pre-training methods to fine-tune various language tasks to achieve high performance. The GPT network has high performance in text generation. The text generation network can also be other networks with text generation functions, which are not specifically limited. The embodiment of the present application takes the text generation network as a GPT network as an example for relevant elaboration.

[0088] In one implementation, the network parameters of the text generation network may be set to the parameter combination first, and then the prompt text may be input into the text generation network, and the output of the text generation network may be used as the reference description information of the product.

[0089] S405: Determine target description information of the product based on the reference description information.

[0090] In one implementation, after the reference description information is generated, the reference description information can be directly used as the target description information of the product; or, the reference description information can be edited again to use the edited reference description information as the target description information of the product. Optionally, after the reference description information is generated, the reference description information can be displayed on a text display interface, and the target object can browse the reference description information on the text display interface and can edit the reference description information again based on its own needs to edit it into the desired description information.

[0091] In one embodiment, the text display interface can be as follows Figure 3 In the user operation interface shown, the reference description information can be displayed in the area marked by 302; after the target object browses to the reference description information in area 302, if the target object wants to edit the reference description information, the reference description information can be edited in area 302, and the edited reference description information is the target description information; after detecting that the editing of the reference description information is completed, the target description information in area 302 can be obtained, and after the terminal obtains the target description information, the target description information can be sent to the computer device.

[0092] As mentioned above, when using a text generation network to generate reference description information of a product using a prompt text as a prompt, there is a corresponding parameter combination, and different parameter combinations can generate different description information. Based on this, it can be seen that in another embodiment, if the target object is not satisfied with the generated reference description information, the description information can be regenerated, that is, the description information generation task can be executed again. In the process of executing the description information generation task again, the parameter combination corresponding to the text generation network can change, so as to generate different reference description information using different parameter combinations; and the target object can select the required reference description information based on the generated different parameter description information.

[0093] For example, there may be a regeneration control on the text display interface, and the target object may trigger the regeneration control to trigger the regeneration of the reference description information. Figure 3 In the user operation interface shown, the regeneration control may be a control marked by 303. Then, when the computer device detects that the regeneration control is triggered (pressed, clicked), the parameter combination may be randomly adjusted to generate new reference description information, such as randomly adjusting the three parameters of Temperature, Top-p, and Top-k to generate new reference description information.

[0094] It should be noted that the regenerated reference description information may be directly used as the target description information, or the target object may manually edit the regenerated reference description information and use the edited reference description information as the target description information.

[0095] As mentioned above, the parameter combination of the text generation network can be dynamically changed, that is, to achieve the update of the parameter combination of the text generation network, the resource conversion amount generated by the target description information of each commodity determined by the text generation network can be obtained, so as to update the parameter combination of the text generation network in real time based on the resource conversion amount of the commodity. After determining the target description information of the current commodity, the parameter combination used by the text generation network when executing the generation task of the reference description information of the next commodity can be further determined, that is, the update of the parameter combination of the text generation network can be achieved.

[0096] In one implementation, with respect to the target description information for generating a product mentioned at this time, the amount of resource conversion generated by the target description information of the product can be obtained, so as to adjust the parameter combination of the text generation network based on the resource conversion amount of the product under the corresponding target description information to obtain an adjusted parameter combination; the adjusted parameter combination can be used as the parameter combination used by the text generation network when executing the task of generating reference description information for the next product.

[0097] Optionally, based on the resource conversion amount of the product under the corresponding target description information, the parameter combination of the text generation network is adjusted, and the specific implementation of obtaining the adjusted parameter combination may be:

[0098] You can first obtain the resource conversion amount generated by the previous product under the corresponding target description information; it should be understood that the target description information of the previous product is the description information of the previous product published online on the product platform. Then, you can compare the resource conversion amount of the product under the corresponding target description information with the resource conversion amount of the previous product under the corresponding target description information, so as to adjust the parameter combination of the text generation network based on the comparison result.

[0099] Specifically, if the amount of resource conversion corresponding to the corresponding target description information of the product is greater than or equal to the amount of resource conversion generated by the previous product under the corresponding target description information, it indicates that the target description information of the current product can bring greater marketing effects, and the target description information is generated based on the reference description information, and the reference description information is generated with prompt text under the corresponding parameter combination, then it can be indicated that the parameter combination corresponding to the current product can bring greater marketing effects, and the parameter combination can continue to be used in the subsequent task of generating description information of the product, that is, the parameter combination used when generating the reference description information of the product using the text generation network can be used as the adjusted parameter combination.

[0100] If the amount of resource conversion corresponding to the target description information of the product is less than the amount of resource conversion generated by the previous product under the corresponding target description information, based on the above description, the parameter combination used when generating the reference description information of the previous product using the text generation network can be used as the adjusted parameter combination.

[0101] In one embodiment, when the source conversion amount is the order resource conversion amount, if the order resource conversion amount corresponding to the corresponding target description information of the product is greater than or equal to the order resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when the reference description information of the product is generated by using the text generation network can be used as the adjusted parameter combination; if the order resource conversion amount corresponding to the corresponding target description information of the product is less than the order resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when the reference description information of the previous product is generated by using the text generation network can be used as the adjusted parameter combination.

[0102] In one embodiment, when the source conversion amount is the click resource conversion amount, if the click resource conversion amount corresponding to the corresponding target description information of the product is greater than or equal to the click resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when generating the reference description information of the product using the text generation network can be used as the adjusted parameter combination; if the click resource conversion amount corresponding to the corresponding target description information of the product is less than the click resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when generating the reference description information of the previous product using the text generation network can be used as the adjusted parameter combination.

[0103] In one embodiment, when the source conversion amount is the order resource conversion amount and the click resource conversion amount, the total resource conversion amount of the product regarding the order resource conversion amount and the click resource conversion amount can be determined first. For the convenience of description, the total resource conversion amount here can be referred to as the first total resource conversion amount; and the total resource conversion amount of the previous product regarding the order resource conversion amount and the click resource conversion amount can be determined. For the convenience of description, the total resource conversion amount here can be referred to as the second total resource conversion amount.

[0104] Finally, the first total resource conversion amount can be compared with the second total resource conversion amount to adjust the parameter combination of the text generation network based on the comparison result. Specifically, if the first total resource conversion amount corresponding to the corresponding target description information of the commodity is greater than or equal to the second total resource conversion amount generated by the previous commodity under the corresponding target description information, the parameter combination used when the reference description information of the commodity is generated by the text generation network can be used as the adjusted parameter combination; if the first total resource conversion amount corresponding to the corresponding target description information of the commodity is less than the second total resource conversion amount generated by the previous commodity under the corresponding target description information, the parameter combination used when the reference description information of the previous commodity is generated by the text generation network can be used as the adjusted parameter combination.

[0105] The determination principle of the first total resource conversion amount and the second resource conversion amount are consistent, and the determination of the first total resource conversion amount is specifically described below. In one embodiment, the sum of the order resource conversion amount and the click resource conversion amount can be used as the first total resource conversion amount. In another embodiment, the order resource conversion amount and the click resource conversion amount can be weighted and summed to take the weighted sum result as the first total resource conversion amount; wherein, the weights corresponding to the order resource conversion amount and the click resource conversion amount are not specifically limited.

[0106] In one implementation, before comparing the resource conversion amount of a product under the corresponding target description information with the resource conversion amount of the previous product under the corresponding target description information, the similarity between the reference description information and the target description information of the product can be determined first, so as to determine whether to perform the comparison of the resource conversion amount based on the similarity. The similarity between the reference description information and the target description information can be determined based on the simHash algorithm, or based on other algorithms that can be used to determine text similarity, which is not limited.

[0107] Based on this, it can be known that the similarity between the reference description information and the target description information can be obtained first; if the similarity is greater than a preset threshold, the amount of resource conversion generated by the product under the corresponding target description information can be obtained; and based on the amount of resource conversion of the product under the corresponding target description information, the parameter combination of the text generation network can be adjusted to obtain the adjusted parameter combination. Among them, the preset threshold can be set in advance, for example, the preset threshold can be set to 60%, 70%, etc. Optionally, the preset threshold can be set to a larger value to ensure that when the reference description information and the target description information have a large similarity, the resource conversion amount is compared to ensure the reliability and accuracy of the parameter combination determination.

[0108] In summary, in the embodiment of the present application, the parameter combination corresponding to the prompt text (prompt) can be fed back according to the resource conversion amount of the online product, so as to continuously optimize the prompt text and the corresponding parameter combination, and obtain a parameter combination suitable for the merchant's marketing direction, so that the product has a greater resource conversion amount on the target description information obtained by using the text generation network.

[0109] For example, in the case where the information release scenario is a product note scenario, the reference description information generated above can be understood as a product note, such as it can be called a reference product note, and the target description information here can be called a target product note. Then, the similarity between the released target product note and the reference product note generated using the prompt text (prompt) can be detected based on the simHash algorithm. If the similarity is greater than 60%, the reference product note generated using the prompt text can be marked, and the CTR and order resource conversion rate can be collected based on the click conversion and order conversion generated by the online product under the target product note. Then, the Temperature, Top-p, Top-k and prompt combination with the highest resource conversion rate (CTR and order resource conversion rate) is used as the parameter combination for the next time the function (i.e., the description information generation function) is used, that is, the combination with the best resource conversion rate can be used by default to generate the description information.

[0110] It can be seen that in the embodiments of the present application, based on the GPT network and combined with multimodal machine learning algorithms and OCR technology, an assistant system for automatically generating product descriptions and product notes can be provided for objects (such as merchants, product recommenders, anchors, etc.), so as to effectively reduce the cost of merchants listing and marketing products.

[0111] In the embodiment of the present application, the description information of the product can be automatically generated directly based on the product image of the product. Compared with the manual editing method, it can effectively avoid the tedious operations and a lot of time-consuming text work caused by manual editing of the description information, without taking up a lot of manpower and energy, and effectively improve the efficiency of generating the description information, thereby effectively reducing the cost of listing and marketing products for merchants. In addition, the corresponding description information can be generated in combination with the prompt text to effectively improve the generation quality of the description information, and high-quality description information can also make the product get a higher probability of click-through rate and order rate.

[0112] See also Figure 5 , Figure 5 : is a schematic diagram of the structure of an information generating device provided in an embodiment of the present application. The information generating device described in this embodiment includes:

[0113] The acquisition unit 501 is used to acquire a product image of a product, and acquire product information of the product based on the product image of the product; the product information includes a product name and product characteristics;

[0114] A filling unit 502 is used to obtain a preset prompt generation template, fill the product information into the prompt generation template, and use the filled prompt generation template as the prompt text of the product;

[0115] The generating unit 503 is used to generate reference description information of the commodity using the prompt text as a prompt, and determine target description information of the commodity based on the reference description information; the target description information is used to be published online on the commodity platform.

[0116] In one implementation, the generating unit 503 is specifically configured to:

[0117] Obtaining a parameter combination when generating description information using a text generation network; different parameter combinations are used to generate different description information for a product;

[0118] The text generation network is called to generate reference description information of the product using the prompt text as a prompt under the parameter combination.

[0119] In one implementation, the target description information is generated after the target object performs secondary editing on the reference description information; the generating unit 503 is further configured to:

[0120] Obtaining the similarity between the reference description information and the target description information;

[0121] If the similarity is greater than a preset threshold, the resource conversion amount generated by the product under the corresponding target description information is obtained; the resource conversion amount includes one or more of the order resource conversion amount under the order dimension and the click resource conversion amount under the click dimension;

[0122] Based on the resource conversion amount of the commodity under the corresponding target description information, the parameter combination of the text generation network is adjusted to obtain an adjusted parameter combination; the adjusted parameter combination is used as the parameter combination used by the text generation network when executing the task of generating reference description information for the next commodity.

[0123] In one implementation, the generating unit 503 is specifically configured to:

[0124] Obtain the resource conversion amount generated by the previous product of the product under the corresponding target description information;

[0125] If the resource conversion amount corresponding to the target description information of the product is greater than or equal to the resource conversion amount generated by the previous product under the target description information, the parameter combination used when generating the reference description information of the product using the text generation network is used as the adjusted parameter combination;

[0126] If the resource conversion amount corresponding to the target description information of the product is less than the resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when generating the reference description information of the previous product using the text generation network will be used as the adjusted parameter combination.

[0127] In one implementation, the acquisition unit 501 is specifically configured to:

[0128] Acquire text description information contained in the product image of the product, and output the text description information on a text display interface;

[0129] Acquire key information selected by the target object from the text description information displayed on the text display interface, wherein the key information includes a product name and product characteristics;

[0130] The key information is used as the commodity information of the commodity.

[0131] In one implementation, the acquisition unit 501 is specifically configured to:

[0132] Performing multimodal feature extraction on the product image to obtain multimodal features of the product image, and generating first text description information of the product based on the multimodal features;

[0133] Extracting text content from the product image based on optical character recognition (OCR) and using the text content as second text description information;

[0134] The first text description information and the second text description information are combined into the text description information included in the product image.

[0135] In one implementation, the filling unit 502 is specifically configured to:

[0136] Acquire the information release scenario of the product, and acquire a prompt generation template matching the information release scenario;

[0137] Alternatively, obtaining the commodity type of the commodity, and obtaining a prompt generation template matching the commodity type;

[0138] Alternatively, the information release scenario and the product type of the product are obtained, and a prompt generation template matching the information release scenario and the product type is obtained.

[0139] It is understood that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. The functional units in the embodiments of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units.

[0140] See also Figure 6 , Figure 6 6 is a schematic diagram of a computer device provided in an embodiment of the present application. The computer device includes: a processor 601 and a memory 602. Optionally, the computer device may also include a network interface 603. The processor 601, the memory 602 and the network interface 603 may exchange data.

[0141] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0142] The memory 602 may include a read-only memory and a random access memory, and provide program instructions and data to the processor 601. A portion of the memory 602 may also include a non-volatile random access memory. When the processor 601 calls the program instructions, it is used to execute:

[0143] Acquire a product image of a product, and acquire product information of the product based on the product image of the product; the product information includes a product name and product characteristics;

[0144] Obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the prompt text of the product;

[0145] The reference description information of the commodity is generated using the prompt text as a prompt, and the target description information of the commodity is determined based on the reference description information; the target description information is used for online publishing on the commodity platform.

[0146] In one implementation, the processor 601 is specifically configured to:

[0147] Obtaining a parameter combination when generating description information using a text generation network; different parameter combinations are used to generate different description information for a product;

[0148] The text generation network is called to generate reference description information of the product using the prompt text as a prompt under the parameter combination.

[0149] In one implementation, the target description information is generated after the target object performs secondary editing on the reference description information; the processor 601 is further configured to:

[0150] Obtaining the similarity between the reference description information and the target description information;

[0151] If the similarity is greater than a preset threshold, the resource conversion amount generated by the product under the corresponding target description information is obtained; the resource conversion amount includes one or more of the order resource conversion amount under the order dimension and the click resource conversion amount under the click dimension;

[0152] Based on the resource conversion amount of the commodity under the corresponding target description information, the parameter combination of the text generation network is adjusted to obtain an adjusted parameter combination; the adjusted parameter combination is used as the parameter combination used by the text generation network when executing the task of generating reference description information for the next commodity.

[0153] In one implementation, the processor 601 is specifically configured to:

[0154] Obtain the resource conversion amount generated by the previous product of the product under the corresponding target description information;

[0155] If the resource conversion amount corresponding to the target description information of the product is greater than or equal to the resource conversion amount generated by the previous product under the target description information, the parameter combination used when generating the reference description information of the product using the text generation network is used as the adjusted parameter combination;

[0156] If the resource conversion amount corresponding to the target description information of the product is less than the resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when generating the reference description information of the previous product using the text generation network will be used as the adjusted parameter combination.

[0157] In one implementation, the processor 601 is specifically configured to:

[0158] Acquire text description information contained in the product image of the product, and output the text description information on a text display interface;

[0159] Acquire key information selected by the target object from the text description information displayed on the text display interface, wherein the key information includes a product name and product characteristics;

[0160] The key information is used as the commodity information of the commodity.

[0161] In one implementation, the processor 601 is specifically configured to:

[0162] Performing multimodal feature extraction on the product image to obtain multimodal features of the product image, and generating first text description information of the product based on the multimodal features;

[0163] Extracting text content from the product image based on optical character recognition (OCR) and using the text content as second text description information;

[0164] The first text description information and the second text description information are combined into the text description information included in the product image.

[0165] In one implementation, the processor 601 is specifically configured to:

[0166] Acquire the information release scenario of the product, and acquire a prompt generation template matching the information release scenario;

[0167] Alternatively, obtaining the commodity type of the commodity, and obtaining a prompt generation template matching the commodity type;

[0168] Alternatively, the information release scenario and the product type of the product are obtained, and a prompt generation template matching the information release scenario and the product type is obtained.

[0169] The present application also provides a computer storage medium in which program instructions are stored. When the program is executed, it may include: Figure 2 or Figure 4 Part or all of the steps of the information generation method in the corresponding embodiment.

[0170] It should be noted that, for the above-mentioned various method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0171] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0172] The embodiment of the present application also provides a computer program product or computer program, which includes program instructions, and when the program instructions are executed by a processor, some or all of the steps in the above method can be implemented. For example, the program instructions are stored in a computer-readable storage medium. The processor of the computer device reads the program instructions from the computer-readable storage medium, and the processor executes the program instructions, so that the computer device executes the steps performed in the embodiments of the above methods.

[0173] The above is a detailed introduction to an information generation method, device, computer equipment and medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for generating information, It is characterized in that The method comprises: Acquire a product image of a product, and acquire product information of the product based on the product image of the product; the product information includes a product name and product characteristics; Obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the prompt text of the product; Generate reference description information of the commodity using the prompt text as a prompt, and determine target description information of the commodity based on the reference description information; the target description information is used for online publishing on the commodity platform; Among them, the generating reference description information of the commodity using the prompt text as a prompt includes: obtaining a parameter combination when generating description information using a text generation network; different parameter combinations are used to generate different description information of a commodity; calling the text generation network under the parameter combination to generate reference description information of the commodity using the prompt text as a prompt; the parameter combination is adjusted based on the resource conversion amount of the previous commodity of the commodity under the corresponding target description information; the resource conversion amount includes one or more of the order resource conversion amount under the order dimension and the click resource conversion amount under the click dimension.

2. The method according to claim 1, It is characterized in that The target description information is generated after the target object performs secondary editing on the reference description information; and further includes: Obtaining the similarity between the reference description information and the target description information; If the similarity is greater than a preset threshold, the resource conversion amount generated by the commodity under the corresponding target description information is obtained; Based on the resource conversion amount of the commodity under the corresponding target description information, the parameter combination of the text generation network is adjusted to obtain an adjusted parameter combination; the adjusted parameter combination is used as the parameter combination used by the text generation network when executing the task of generating reference description information for the next commodity.

3. The method according to claim 2, It is characterized in that The adjusting the parameter combination of the text generation network based on the resource conversion amount corresponding to the corresponding target description information of the commodity to obtain the adjusted parameter combination includes: Obtain the resource conversion amount generated by the previous product of the product under the corresponding target description information; If the resource conversion amount corresponding to the target description information of the product is greater than or equal to the resource conversion amount generated by the previous product under the target description information, the parameter combination used when generating the reference description information of the product using the text generation network is used as the adjusted parameter combination; If the resource conversion amount corresponding to the target description information of the product is less than the resource conversion amount generated by the previous product under the corresponding target description information, the parameter combination used when generating the reference description information of the previous product using the text generation network will be used as the adjusted parameter combination.

4. The method according to claim 1, It is characterized in that The acquiring the commodity information of the commodity based on the commodity image of the commodity includes: Acquire text description information contained in the product image of the product, and output the text description information on a text display interface; Acquire key information selected by the target object from the text description information displayed on the text display interface, wherein the key information includes a product name and product characteristics; The key information is used as the commodity information of the commodity.

5. The method according to claim 4, It is characterized in that The obtaining of text description information contained in the product image includes: Performing multimodal feature extraction on the product image to obtain multimodal features of the product image, and generating first text description information of the product based on the multimodal features; Extracting text content from the product image based on optical character recognition (OCR) and using the text content as second text description information; The first text description information and the second text description information are combined into the text description information included in the product image.

6. The method according to claim 1, It is characterized in that The step of obtaining a preset prompt generation template includes: Acquire the information release scenario of the product, and acquire a prompt generation template matching the information release scenario; Alternatively, obtaining the commodity type of the commodity, and obtaining a prompt generation template matching the commodity type; Alternatively, the information release scenario and the product type of the product are obtained, and a prompt generation template matching the information release scenario and the product type is obtained.

7. An information generating device, It is characterized in that include: An acquisition unit, used to acquire a commodity image of a commodity, and acquire commodity information of the commodity based on the commodity image of the commodity; The product information includes product name and product characteristics; A filling unit, used for obtaining a preset prompt generation template, filling the product information into the prompt generation template, and using the filled prompt generation template as the prompt text of the product; a generating unit, configured to generate reference description information of the commodity using the prompt text as a prompt, and determine target description information of the commodity based on the reference description information; The target description information is used for online publishing on the commodity platform; Among them, the generating reference description information of the commodity using the prompt text as a prompt includes: obtaining a parameter combination when generating description information using a text generation network; different parameter combinations are used to generate different description information of a commodity; calling the text generation network under the parameter combination to generate reference description information of the commodity using the prompt text as a prompt; the parameter combination is adjusted based on the resource conversion amount of the previous commodity of the commodity under the corresponding target description information; the resource conversion amount includes one or more of the order resource conversion amount under the order dimension and the click resource conversion amount under the click dimension.

8. A computer device, It is characterized in that The method comprises a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.