Question and answer generation method, device and equipment based on commodity information

Abstract

The invention discloses a question and answer generation method, device and equipment based on commodity information. The method comprises the following steps: acquiring commodity information and portrait information of an order placing user to construct a user portrait; generating a plurality of buyer roles according to the commodity information and the user portrait by using a preset large language model; generating a plurality of initial questions based on different buyer roles and the commodity information, and supplementing and optimizing the generated initial questions to obtain updated questions; and constructing question and answer prompt words, and generating a plurality of reference answers corresponding to each updated question through the large language model according to the updated questions and the commodity information to obtain a question answer list. Quality and personalization of question and answer generation based on commodity information are improved, and more accurate, comprehensive and high-quality question and answer service is provided for users.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device and equipment for generating questions and answers based on commodity information. Background Art

[0002] In the current live e-commerce and digital human interaction field, in order to improve the interactivity and efficiency of live broadcasts, anchors or digital humans usually prepare detailed question-and-answer sessions before live broadcasts. The key to this link is to accurately predict and answer questions that the audience may ask, thereby ensuring the smoothness of the live broadcast process and the satisfaction of the audience experience. To achieve this goal, artificial intelligence (AI) technology, especially the question-and-answer system based on retrieval-augmented generation (RAG), has been widely used in this field. However, although RAG technology has improved the automation and intelligence level of the question-and-answer system to a certain extent, it still faces many challenges and limitations in practical applications, which limit its further development and application effects.

[0003] First, the complexity of input content and model hallucination problems. RAG systems usually rely on a large amount of text data as input. However, when the input content is too large and complex, large language models are prone to "hallucinations" when processing this information. That is, the model may be disturbed by irrelevant or misleading information when understanding and generating answers, resulting in errors or deviations when answering audience questions. This model hallucination problem not only reduces the accuracy of question-answering, but may also have a negative impact on the credibility of the live broadcast and the audience experience.

[0004] Secondly, personalized needs are ignored. The RAG system mainly focuses on answering known knowledge or information, such as common questions in live broadcast scripts and basic attributes of products. However, in actual live broadcast scenarios, audience questions are often more diverse and personalized, involving multiple aspects such as specific usage scenarios of products, personal preferences, and purchase motivations. Since the RAG system lacks in-depth understanding and analysis of user portraits and personalized needs, it is difficult to generate high-quality answers that meet the personalized needs of the audience. Summary of the invention

[0005] In view of this, the purpose of the present invention is to propose a question and answer generation method, device and equipment based on product information, aiming to solve the defects of the existing question and answer system in processing complex inputs and meeting personalized needs, thereby improving the quality and personalization of question and answer generation based on product information, and providing users with more accurate, comprehensive and high-quality question and answer services.

[0006] To achieve the above object, the present invention provides a method for generating questions and answers based on commodity information, the method comprising:

[0007] Obtain product information and ordering user portrait information to build user portraits;

[0008] Generate multiple buyer personas based on the product information and the user portrait using a preset large language model;

[0009] Generate multiple initial questions based on different buyer roles and product information, and supplement and optimize the generated initial questions to obtain updated questions;

[0010] Construct question and answer prompt words, generate multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model, and obtain a question answer list.

[0011] Preferably, the method further comprises:

[0012] The predicted questions corresponding to the digital human's output answers are screened out from the question answer list, and the reference answers corresponding to the predicted questions are checked for correctness and richness.

[0013] Preferably, the method further comprises:

[0014] The reference answers in the question answer list are sorted, and the best reference answer is selected as the target answer according to the sorting result, and the remaining reference answers are used as alternative answers; wherein the sorting is based on the relevance, accuracy, readability and popularity of the answers.

[0015] Preferably, the step of acquiring product information and portrait information of the ordering user to construct a user portrait includes:

[0016] Obtain the product title, attributes and product image for parsing to obtain the product information including product name, product function, brand, price, applicable population, usage scenario, style, color and design;

[0017] Obtain and analyze the user's historical purchasing behavior, browsing history, and user information portrait information to construct the user portrait.

[0018] Preferably, the generating multiple buyer personas according to the product information and the user portrait using a preset large language model includes:

[0019] Utilizing the large language model to analyze the product information and the user portrait to obtain user orientation including needs, pain points, and preferences;

[0020] A plurality of buyer roles covering different target user groups are generated according to the product information, the user portrait and the user orientation, wherein the role information of the buyer role includes role name, age, gender, occupation, consumption budget, background description and purchase motivation.

[0021] Preferably, the generated initial question is supplemented and optimized to obtain an updated question, including:

[0022] Add a reflection mechanism to continue asking questions based on the initial question to supplement the initial question, and after each iteration, delete the repeated questions to obtain supplementary questions;

[0023] The supplementary questions are optimized according to the pertinence, diversity and authenticity of the questions to obtain the updated questions; wherein the pertinence of the questions includes whether the questions revolve around the purchase motivation of the buyer role and the core selling points of the product; the diversity of the questions includes whether the questions cover all aspects of the product information; the authenticity of the questions includes whether the questions conform to the language habits of actual purchase conversations or comments.

[0024] Preferably, the constructing of question and answer prompt words generates a plurality of reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list, including:

[0025] Using the large language model to think about the intention of the updated question, and determine whether the updated question is related to a preset reference question and answer pair, if so, determining whether an answer can be found in the reference question and answer, if not, outputting that it cannot be answered;

[0026] If it is determined that the reference question and answer can find an answer, multiple reference answers are output and the multiple reference answers are synonymous sentences. Otherwise, the corresponding reference answers are supplemented manually.

[0027] To achieve the above object, the present invention further provides a question-answer generation device based on commodity information, the device comprising:

[0028] A data acquisition unit, used to acquire product information and portrait information of ordering users to construct user portraits;

[0029] A role generation unit, used to generate multiple buyer roles according to the product information and the user portrait using a preset large language model;

[0030] A question optimization unit generates a plurality of initial questions based on the different buyer roles and the product information, and supplements and optimizes the generated initial questions to obtain updated questions;

[0031] The answer generation unit is used to construct question and answer prompt words, and generate multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list.

[0032] Preferably, the device further comprises:

[0033] The screening unit is used to screen out the predicted questions corresponding to the digital human's answers to be output from the question answer list, and to check the reference answers corresponding to the predicted questions for correctness and richness.

[0034] In order to achieve the above-mentioned purpose, the present invention also proposes a question and answer generation device based on product information, including a processor, a memory and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the steps of a question and answer generation method based on product information as described in the above-mentioned embodiment.

[0035] In order to achieve the above objectives, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. The computer program is executed by a processor to implement the steps of a method for generating questions and answers based on product information as described in the above embodiment.

[0036] In order to achieve the above-mentioned purpose, the present invention also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a method for generating questions and answers based on product information as described in the above-mentioned embodiment.

[0037] Beneficial effects:

[0038] The above scheme generates buyer personas by combining product information and user portraits, making the generated questions closer to the needs and concerns of actual buyers, enhancing the personalization of questions and answers; and the generated multiple buyer personas cover different target user groups, making the generated questions more extensive and representative. By using a large language model to automatically generate multiple initial questions and optimize them, the comprehensiveness of the questions, as well as the speed and efficiency of question and answer generation, are improved; by generating multiple reference answers, the diversity and richness of the answers to the questions are increased, which helps to provide more comprehensive information to users and meet their personalized needs.

[0039] The above scheme ensures that the questions and answers finally output to the digital human are of high quality and pertinence by screening target questions and checking the correctness and richness of answers. Outputting screened and checked answers reduces the occurrence of erroneous or inaccurate information and improves the user experience.

[0040] The above scheme selects the best answer through a sorting mechanism, ensuring the quality and relevance of the answer, while providing alternative answers to increase flexibility; sorting based on multiple dimensions (relevance, accuracy, readability, popularity) makes the choice of answer more scientific and reasonable.

[0041] The above solution constructs a more comprehensive and in-depth user portrait through detailed analysis of product information and user portraits, laying the foundation for the subsequent generation of high-quality questions and answers; accurately obtaining key information about products and user purchasing behavior helps to generate more precise questions that are closer to user needs.

[0042] The above scheme, by adding reflection mechanisms and optimization steps, allows the big model to think of more effective questions in the product field, improves the pertinence and diversity of the questions, makes the questions closer to the actual purchase scenario, and also avoids the repetition and redundancy of questions; and ensures the authenticity of the questions, making the generated questions more in line with the language habits of actual purchase conversations or comments, thereby improving the practicality and credibility of the questions and answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flowchart of a method for generating questions and answers based on product information provided by an embodiment of the present invention.

[0045] Figure 2 A schematic diagram of the structure of a question-and-answer generation device based on commodity information provided by an embodiment of the present invention.

[0046] The realization of the purpose, functional features and advantages of the invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0048] The content of the present invention is described in detail below with reference to embodiments.

[0049] Reference Figure 1 The figure is a flow chart of a method for generating questions and answers based on commodity information provided by an embodiment of the present invention.

[0050] In this embodiment, the method includes:

[0051] S11, obtain product information and portrait information of the ordering user to build a user portrait.

[0052] Furthermore, in step S11, the obtaining of product information and portrait information of the ordering user to construct a user portrait includes:

[0053] S11-1, obtaining the title, attributes and product image of the product for parsing, and obtaining the product information including product name, product function, brand, price, applicable population, usage scenario, style, color and design;

[0054] S11-2, obtaining and analyzing the user's historical purchasing behavior, browsing history, and user information portrait information to construct the user portrait.

[0055] In this embodiment, the title, attributes, description and other text information of the product are obtained on the product page, product packaging, and the anchor in the live text description, and natural language processing is performed to extract keywords, entities, attribute values, etc., that is, the product name, product type, product function, brand, specification, listing time, price, origin, applicable population, color, style, design, etc. are obtained; and the product image is identified by using image recognition technology to extract color, style, scene, and other visual information, and extract text information such as selling points. The portrait information of the ordering user (including historical purchase behavior, browsing history, search history, social media activities, and user information) is obtained from the website or other data sources for analysis to build a more comprehensive user portrait. The above-obtained product information and user portrait are structured and saved in a queryable format, such as JSON format, for subsequent data processing and analysis. When the product information, live broadcast speech or user portrait changes, there is no need to manually update a large amount of data and retrain the model, which improves the accuracy and efficiency of data processing.

[0056] S12, using a preset large language model to generate multiple buyer roles according to the product information and the user portrait.

[0057] Furthermore, in step S12, the use of a preset large language model to generate multiple buyer roles according to the product information and the user portrait includes:

[0058] S12-1, using the large language model to analyze the product information and the user portrait to obtain user orientation including needs, pain points and preferences;

[0059] S12-2, generating a plurality of buyer roles covering different target user groups based on the product information, the user portrait and the user orientation, wherein the role information of the buyer role includes role name, age, gender, occupation, consumption budget, background description and purchase motivation.

[0060] In this embodiment, the user's needs, pain points, preferences, etc. are inferred based on the product information and user portraits (such as gender, age, geographical distribution, etc.) using a large language model. Among them, needs refer to the basic functions or features that potential buyers seek when purchasing goods. For example, for a smart watch, "needs" may mean that users expect it to monitor heart rate, record steps and other health data; pain points refer to specific problems or inconveniences encountered by buyers when using similar products. For example, for a mobile power bank, "pain points" may be users complaining about slow charging speeds or insufficient battery capacity; preferences refer to the buyer's tendency to prefer specific designs, functions or brands when choosing goods. For example, for an outdoor backpack, "preferences" may be that users prefer waterproof materials, bright colors or designs with multiple compartments. The prompt example is as follows:

[0061] Based on the given product information and target user portrait, please infer the user's possible needs, pain points and preferences, and generate 1-2 natural language descriptions for each of the three aspects.

[0062] Product information:

[0063] {product_info}

[0064] Target user profile:

[0065] {user_profile}

[0066] Inference and generation requirements:

[0067] 1. Carefully analyze the product’s attributes, selling points, scenarios, and other information, and combine the buyer’s gender, age distribution, and other portrait information to infer the user’s possible needs, pain points, and preferences for the product.

[0068] 2. Generate 1-2 smooth and coherent natural language descriptions for each of the three aspects: needs, pain points, and preferences. The language should be concise and clear, and in line with people's daily expression habits.

[0069] 3. The inferred and generated content should be closely related to the given products and user characteristics. It should not be too broad or too specific.

[0070] 4. The description of each aspect starts with "Requirements:", "Pain points:", "Preferences:", and different aspects are separated by line breaks.

[0071] Inference results:

[0072] <result>

[0073] need:

[0074] Pain points:

[0075] Preference:

[0076] < / result>

[0077] Generate multiple buyer roles based on product information, user portraits, and the needs, pain points, and preferences generated above. Generating buyer roles mainly relies on the nlg capabilities of the big model. In order to avoid random generation, you need to prepare an example in advance to prompt the big model and generate specific roles according to the example and requirements. The prompt example is as follows (the user portrait takes mountaineering shoes as an example):

[0078] Please generate several typical buyer personas based on the product information provided and the target user group. Each buyer persona should include the following role information:

[0079] Role name: Use a concise word or phrase to summarize the characteristics of the role, such as "young explorer", "urban white-collar beauty", etc.

[0080] Age: The age range of the character, such as "20-30 years old", "35-50 years old", etc.

[0081] Gender: The gender of the character, which can be "Male", "Female" or "Unlimited".

[0082] Occupation: The character's professional identity or work field, such as "student", "financial practitioner", "retired person", etc. Hobbies: The character's favorite activities or things in life, which can be specific projects such as "hiking" and "yoga", or general descriptions such as "trendy culture".

[0083] Consumption budget: The price range that the character is willing to pay for the product, which can be expressed as a numerical range, such as "1000-1500 yuan".

[0084] Background description: Use one to two sentences to describe the character's living conditions, personality traits, pursuits, etc. to help you understand the character more comprehensively.

[0085] Purchase motivation: The reason or purpose for the character to purchase the product, reflecting his or her demands on the product's functions, performance, etc.

[0086] Product information:

[0087] {product_info}

[0088] Target user profile:

[0089] {user_profile}

[0090] User requirements:

[0091] {user_demands}

[0092] User pain points:

[0093] {user_pain_points}

[0094] User Preferences:

[0095] {user_preferences}

[0096] Require:

[0097] 1. Create 3-5 buyer personas that are in line with the actual situation based on the product’s attributes, functions, price, and other characteristics, as well as the target user group’s age, occupation, spending power, and other factors.

[0098] 2. Each buyer persona's information should be written on the same line, with commas separating the different fields. The order and number of fields should be consistent with the template given in the instructions.

[0099] 3. The description of the buyer role should be specific and vivid, so that people can quickly form an impression of the role. At the same time, pay attention to the distinction between different roles to avoid being too similar.

[0100] 4. When designing roles, try to cover all segments of the target user group and present a diverse user portrait.

[0101] Output:

[0102] <result>

[0103] Buyer Persona:

[0104] < / result>

[0105] S13, generating a plurality of initial questions based on the different buyer roles and the product information, and supplementing and optimizing the generated initial questions to obtain updated questions.

[0106] Furthermore, in step S13, the generated initial question is supplemented and optimized to obtain an updated question, including:

[0107] S13-1, adding a reflection mechanism, continuing to ask questions based on the initial question to supplement the initial question, and after each iteration, deleting repeated questions to obtain supplementary questions;

[0108] S13-2, optimizing the supplementary questions according to the relevance, diversity and authenticity of the questions to obtain the updated questions; wherein the relevance of the questions includes whether the questions revolve around the purchase motivations of the buyer persona and the core selling points of the products; the diversity of the questions includes whether the questions cover all aspects of the product information; the authenticity of the questions includes whether the questions conform to the language habits of actual purchase conversations or comments.

[0109] In this embodiment, multiple possible questions from real buyers are generated based on buyer roles and product information. In order to generate more targeted questions, it is necessary to ensure that the questions of each buyer role are closely aligned with the role's background characteristics and purchase motivations, and focus on the product's functions, performance, usage scenarios, brand reputation, etc., to reflect the buyer role's unique concerns and purchase demands. The following aspects can be considered:

[0110] Functional questions about the product: Ask questions about the basic functions and operation of the product;

[0111] Performance-related questions: questions about the product's technical specifications, durability, etc.

[0112] Usage scenario questions: Explore the effects of using the product in different environments;

[0113] Brand reputation questions: Ask about brand history, user reviews, etc.

[0114] After-sales service issues: understand the warranty policy, return and exchange procedures and other service details;

[0115] Personalized demand questions: Propose specific demands based on the role's preferences and pain points.

[0116] The prompt example is as follows:

[0117] Please play the role of a {buyer_persona#Specific buyer role} and ask as many questions as possible from the perspective of the buyer role based on the given product information. The questions should focus on the product's functions, performance, usage scenarios, brand reputation, etc., reflecting the buyer role's unique concerns and purchasing demands.

[0118] Buyer Persona:

[0119] {buyer_persona}

[0120] Product information:

[0121] {product_info}

[0122] Require:

[0123] 1. The questions should be closely related to the buyer’s background characteristics and purchasing motivations, reflecting the impact of factors such as age, occupation, interests, and consumption concepts on purchasing decisions.

[0124] 2. Questions should be phrased in a natural, colloquial manner, like questions a buyer would ask in an actual conversation or review.

[0125] 3. Different questions should have different focuses, covering different aspects of the product and reflecting the diverse concerns of buyers.

[0126] 4. In addition to common aspects such as functions and prices, you can also ask questions from the perspectives of after-sales service, user experience, social recognition, etc. to highlight the personalized demands of the character.

[0127] 5. Questions should not be too broad or general, but should be specific and focused, and able to guide the seller to give targeted answers.

[0128] Output:

[0129] <result>

[0130] List of questions:

[0131] < / result>

[0132] By further adding a reflection mechanism, the big model can think of more effective questions in the commodity field. This stage is mainly to supplement the questions generated above, because the big model is lazy to generate. The main implementation method is to continue asking questions based on the last generated questions, rather than starting over. Starting over may not necessarily generate newer questions. On the contrary, it may cause repeated questions to be generated because there is no previous context. In particular, the more iterations, the easier it is to generate repeated questions. Although continuing to ask questions can reduce repeated questions, the best way is to find repeated questions after each iteration, and tell the big model the next time you ask questions that question xx has been generated repeatedly. The prompt example is as follows:

[0133] Please evaluate the problem generated in the previous step and try to optimize it from the following perspectives:

[0134] 1. Targetedness of questions: Does the question closely follow the characteristics and demands of the buyer’s role? Does it target the core selling point of the product?

[0135] 2. Diversity of questions: Do the questions cover all important aspects of the product? Are they focused and differentiated?

[0136] 3. Depth of question: Does the question touch upon the buyer’s deep needs and concerns? Can it lead the seller to provide valuable information?

[0137] 4. Authenticity of the question Does the question match the language used in actual purchase conversations or reviews? Does it sound like a question a real person would ask?

[0138] Please give a brief evaluation of the generated questions based on the above criteria and propose 2-3 optimized questions.

[0139] Please organize the initially generated questions and the optimized questions and output the final question list.

[0140] <final_questions>

[0141] List of questions:

[0142] < / final_questions>

[0143] S14, constructing question and answer prompt words, and generating multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list.

[0144] Further, in step S14, the question and answer prompt words are constructed, and multiple reference answers corresponding to each of the updated questions are generated according to the updated questions and the product information through the large language model to obtain a question answer list, including:

[0145] S14-1, using the large language model to think about the intention of the updated question, and determine whether the updated question is related to a preset reference question and answer pair, if so, determining whether an answer can be found in the reference question and answer, if not, outputting that it cannot be answered;

[0146] S14-2, if the reference question and answer determines that an answer can be found, multiple reference answers are output and the multiple reference answers are synonymous sentences, otherwise, the corresponding reference answers are supplemented manually.

[0147] In this embodiment, the question and answer prompt words are constructed to enable the large language model to output the answer to the corresponding question based on the question and product information. At this stage, the large model can be appropriately created based on the answer to improve the richness and professionalism of the answer. The prompt example is as follows:

[0148] You are a professional e-commerce customer service representative. Your task is to answer user questions from the perspective of a professional customer service representative based on the prompted questions and answers.

[0149] #Require

[0150] The answer should be objective and truthful, without exaggeration or misleading users.

[0151] #method

[0152] Please follow these few steps:

[0153] 1. You first need to think about the user's question intention, and then determine whether the user's question is related to the reference question and answer pair;

[0154] 2. If it is not relevant, output "cannot answer", if it is relevant, go to the next step;

[0155] 3. You need to refer to the input reference question and answer pairs to see if you can find the answer. You can only answer questions within the scope of the reference question and answer pairs. Guessing questions is not allowed.

[0156] 4. If yes, provide 5 answers and output them. These five answers must be synonymous sentences;

[0157] 5. Based on the answers you think are correct, add appropriate marketing techniques to highlight the selling points and advantages of the product. The tone, style, and length of each answer should be different to suit different scenarios and users.

[0158] #Actual user questions: {question}

[0159] The following content is your reference for this question and answer session: {content}

[0160] Output:

[0161] Furthermore, the method further comprises:

[0162] S15, selecting predicted questions corresponding to the digital human's output answers from the question answer list, and checking the reference answers corresponding to the predicted questions for correctness and richness.

[0163] In this embodiment, the questions that need to be answered by the digital person are screened from the question answer list, and the answers are proofread. For questions with answers, the correctness of the answers is checked first, and then the style of the answers is checked to see if they match the style of the live broadcast room. The third step is to check the richness of the answers and add more details as needed. For questions without answers, the merchant can decide whether to add answers.

[0164] Furthermore, the method further comprises:

[0165] S16, sorting the reference answers in the question answer list, selecting the best reference answer as the target answer according to the sorting result, and using the remaining reference answers as alternative answers; wherein the sorting is based on the relevance, accuracy, readability and popularity of the answers.

[0166] In this embodiment, keywords in the reference answer are extracted and matched with keywords in the updated question, and the keyword overlap or similarity is calculated as a preliminary indicator of relevance. For factual answers that can be directly verified, facts are checked through external data sources (such as commodity databases, professional websites, etc.) to ensure the accuracy of the answers. The language model in natural language processing is used to evaluate the grammatical correctness and fluency of the answers. If the system already has user feedback data, the popularity of the answer can be evaluated based on historical feedback (such as likes, comments, etc.). Different weights are assigned to each factor according to the importance of relevance, accuracy, readability, and popularity, and then the comprehensive score of each answer is calculated and sorted according to the score. The highest score is considered the best reference answer.

[0167] Furthermore, the method also includes:

[0168] By collecting user feedback on the target answers, including satisfaction, improvement suggestions, etc.; fine-tuning the large language model based on user feedback to improve the accuracy and personalization of subsequent question and answer generation.

[0169] Reference Figure 2 FIG. 1 is a schematic diagram of the structure of a question-and-answer generating device based on commodity information provided by an embodiment of the present invention.

[0170] In this embodiment, the device 20 includes:

[0171] A data acquisition unit 21 is used to acquire product information and portrait information of the ordering user to construct a user portrait;

[0172] A role generating unit 22, configured to generate a plurality of buyer roles according to the product information and the user portrait using a preset large language model;

[0173] A question optimization unit 23 generates a plurality of initial questions based on the different buyer roles and the product information, and supplements and optimizes the generated initial questions to obtain updated questions;

[0174] The answer generation unit 24 is used to construct question and answer prompt words, and generate multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list.

[0175] Furthermore, the device also includes:

[0176] The screening unit 25 is used to screen out the predicted questions corresponding to the digital human's answers to be output from the question answer list, and to check the reference answers corresponding to the predicted questions for correctness and richness.

[0177] Each unit module of the device 20 can respectively execute the corresponding steps in the above method embodiment, so each unit module will not be described in detail here. Please refer to the description of the above corresponding steps for details.

[0178] The embodiment of the present invention further provides a device for generating questions and answers based on commodity information. The device includes the device for generating questions and answers based on commodity information as described above. The device for generating questions and answers based on commodity information can be used to generate questions and answers based on commodity information. Figure 2 The structure of the embodiment can be executed accordingly. Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects. For details, please refer to the relevant records in the above embodiments, which will not be repeated here.

[0179] The device includes: a mobile phone, a digital camera, a tablet computer, or other devices with a camera function, or a device with an image processing function, or a device with an image display function. The device may include components such as a memory, a processor, an input unit, a display unit, and a power supply.

[0180] Among them, the memory can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as an image playback function, etc.), etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory may also include a memory controller to provide the processor and the input unit with access to the memory.

[0181] The input unit can be used to receive input digital or character or image information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control. Specifically, the input unit of this embodiment includes not only a camera, but also a touch-sensitive surface (such as a touch display) and other input devices.

[0182] The display unit can be used to display information input by the user or information provided to the user and various graphical user interfaces of the device, which can be composed of graphics, text, icons, videos and any combination thereof. The display unit may include a display panel, and optionally, the display panel may be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc. Further, the touch-sensitive surface may cover the display panel, and when the touch-sensitive surface detects a touch operation on or near it, it is transmitted to the processor to determine the type of touch event, and then the processor provides a corresponding visual output on the display panel according to the type of touch event.

[0183] The embodiment of the present invention further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the memory in the above embodiment; or a computer-readable storage medium that exists independently and is not installed in a device. The computer-readable storage medium stores at least one instruction, which is loaded and executed by a processor to implement Figure 1 The computer-readable storage medium may be a read-only memory, a disk or an optical disk, etc.

[0184] The embodiment of the present invention also provides a computer program product, including a computer program / instruction, which is loaded and executed by a processor to implement Figure 1 A question and answer generation method based on product information is shown.

[0185] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, equipment embodiment and storage medium embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0186] Furthermore, in this document, the terms "comprises," "comprising," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element.

[0187] The above description shows and describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention, through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not depart from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A question-answer generation method based on commodity information, characterized in that: The method comprises: Obtain product information and ordering user portrait information to build user portraits; Generate multiple buyer personas based on the product information and the user portrait using a preset large language model; Generate multiple initial questions based on different buyer roles and product information, and supplement and optimize the generated initial questions to obtain updated questions; Construct question and answer prompt words, generate multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model, and obtain a question answer list.

2. A method for generating questions and answers based on commodity information according to claim 1, characterized in that: The method further comprises: The predicted questions corresponding to the digital human's output answers are screened out from the question answer list, and the reference answers corresponding to the predicted questions are checked for correctness and richness.

3. A method for generating questions and answers based on commodity information according to claim 1, characterized in that: The method further comprises: The reference answers in the question answer list are sorted, and the best reference answer is selected as the target answer according to the sorting result, and the remaining reference answers are used as alternative answers; wherein the sorting is based on the relevance, accuracy, readability and popularity of the answers.

4. A method for generating questions and answers based on commodity information according to claim 1, characterized in that: The step of obtaining product information and portrait information of the ordering user to construct a user portrait includes: Obtain the product title, attributes and product image for parsing to obtain product information including product name, product function, brand, price, applicable population, usage scenario, style, color and design; Obtain and analyze the user's historical purchasing behavior, browsing history, and user information portrait information to construct the user portrait.

5. The method for generating questions and answers based on commodity information according to claim 1, characterized in that: The method of using a preset large language model to generate multiple buyer roles according to the product information and the user portrait includes: Utilizing the large language model to analyze the product information and the user portrait to obtain user orientation including needs, pain points, and preferences; A plurality of buyer roles covering different target user groups are generated according to the product information, the user portrait and the user orientation, wherein the role information of the buyer role includes role name, age, gender, occupation, consumption budget, background description and purchase motivation.

6. A method for generating questions and answers based on commodity information according to claim 1, characterized in that: The generated initial question is supplemented and optimized to obtain an updated question, including: Add a reflection mechanism to continue asking questions based on the initial question to supplement the initial question, and after each iteration, delete the repeated questions to obtain supplementary questions; The supplementary questions are optimized according to the pertinence, diversity and authenticity of the questions to obtain the updated questions; wherein the pertinence of the questions includes whether the questions revolve around the purchase motivation of the buyer role and the core selling points of the product; the diversity of the questions includes whether the questions cover all aspects of the product information; the authenticity of the questions includes whether the questions conform to the language habits of actual purchase conversations or comments.

7. A method for generating questions and answers based on commodity information according to claim 1, characterized in that: The constructing question and answer prompt words generates a plurality of reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list, including: Using the large language model to think about the intention of the updated question, and determine whether the updated question is related to a preset reference question and answer pair, if so, determining whether an answer can be found in the reference question and answer, if not, outputting that it cannot be answered; If it is determined that the reference question and answer can find an answer, multiple reference answers are output and the multiple reference answers are synonymous sentences. Otherwise, the corresponding reference answers are supplemented manually.

8. A question-answer generation device based on commodity information, characterized in that: The device comprises: A data acquisition unit, used to acquire product information and portrait information of ordering users to construct user portraits; A role generation unit, used to generate multiple buyer roles according to the product information and the user portrait using a preset large language model; A question optimization unit generates a plurality of initial questions based on the different buyer roles and the product information, and supplements and optimizes the generated initial questions to obtain updated questions; The answer generation unit is used to construct question and answer prompt words, and generate multiple reference answers corresponding to each of the updated questions according to the updated questions and the product information through the large language model to obtain a question answer list.

9. A question-and-answer generating device based on commodity information according to claim 8, characterized in that: The device also includes: The screening unit is used to screen out the predicted questions corresponding to the digital human's answers to be output from the question answer list, and to check the reference answers corresponding to the predicted questions for correctness and richness.

10. A question-answer generation device based on commodity information, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the steps of a method for generating questions and answers based on commodity information as described in any one of claims 1 to 7.

Images