Large-language-model-based interaction method and system

Abstract

Provided in the present invention are a large-language-model-based interaction method and system. The method comprises: acquiring a user session input; performing retrieval and aggregation to obtain answer retrieval results, and on the basis of the user session input and the current user intention, retrieving marketing delivery information from a marketing database and performing aggregation, so as to obtain marketing delivery retrieval results; comprehensively sorting reference sources of the answer retrieval results and reference sources of the marketing delivery retrieval results, and then returning same to a user as reference source results; summarizing and integrating the reference source results and the marketing delivery retrieval results, so as to obtain answer results, and returning the answer results to the user; and integrating a user request, the answer results and the marketing retrieval results, so as to obtain recommended associated questions, and then returning the recommended associated questions to the user, thereby guiding the user to further make multiple rounds of conversations. In the present invention, in addition to retrieval augmentation and active retrieval, retrieval augmentation for marketing-related information is proposed, and therefore the present invention provides a better experience for users and can satisfy enterprise marketing requirements.

Description

Large model interaction method and system Technical Field

[0001] The present invention relates to the technical field of large models, and in particular to a large model interaction and system. Background Art

[0002] With the development of big model technology, users can quickly obtain answers through interaction with large models. On the user side, chatbots or conversational platforms developed using prompt word engineering and search enhancements only receive straightforward responses to their requests within a limited chat memory window and contextual information. On the enterprise side, facing the rapid growth of conversational platforms with rapidly increasing applications and traffic, businesses are unable to effectively distribute marketing content to large model users.

[0003] Most current large-scale interactive models are search-enhanced systems, as shown in Figure 1. These primarily facilitate interactive conversations involving proprietary and real-time content through local proprietary databases or third-party APIs. However, these systems suffer from several drawbacks: The answers are plain text directly associated with the conversation request; and, in order to obtain accurate answers, users must describe their needs as precisely and in detail as possible.

[0004] Take the Perplexity platform, for example. As one of the more user-friendly conversational platforms, Perplexity enhances direct answers through traditional search, supplementing them with citation source information and related question feedback. This allows users to further understand the answer details and simplifies the user-questioning interaction. Screenshots of its page and its logical analysis structure are shown in Figures 2 and 3, respectively. However, its shortcomings are: the current platform only returns results directly related to the conversation request; it does not further infer user needs and intent to provide service recommendations. The lack of marketing capabilities also limits the platform's business model to a subscription-based payment model. Technical issues

[0005] The technical problem to be solved by the present invention is to provide a large model interaction method and system, which proposes marketing information retrieval enhancement based on retrieval enhancement and active retrieval, provides users with a better experience and can meet the marketing needs of enterprises. Technical Solutions

[0006] In a first aspect, the present invention provides a large model interaction method, comprising the following steps:

[0007] S1. Obtain user session input and parse the user session input to obtain a user request;

[0008] S2. Search and summarize the user request to obtain answer search results, which include reference source links and search content;

[0009] At the same time, the user's personal setting information, third-party user profile information, historical conversation information and intent, and current conversation context are comprehensively analyzed to summarize and infer the current user intent; based on the user conversation input and the current user intent, marketing delivery information is retrieved from the marketing database and summarized to obtain marketing delivery search results, which include delivery links, delivery content, delivery type, and delivery strategy;

[0010] S3. Based on the link weights of the answer search result and the marketing placement search result, as well as the bid price of the marketing placement search result, the reference sources of the answer search result and the reference sources of the marketing placement search result are comprehensively ranked and returned to the user as reference source results;

[0011] S4. After summarizing and integrating the reference source results and the marketing placement search results, an answer result is obtained and returned to the user;

[0012] S5. Integrate the user request, the answer result, and the marketing search result, obtain recommended related questions, and return them to the user to guide the user to further multiple rounds of dialogue.

[0013] In a second aspect, the present invention provides a large model interaction system, comprising:

[0014] The session parsing module is used to obtain user session input and parse the user session input to obtain user requests;

[0015] A retrieval module, configured to search and summarize the user request to obtain an answer retrieval result, wherein the answer retrieval result includes a reference source link and retrieval content;

[0016] The marketing delivery module is used to comprehensively analyze user personal setting information, third-party user profile information, historical conversation information and intent, and current conversation context while the retrieval module is performing a search, and summarize and infer the current user intent; based on the user conversation input and the current user intent, the marketing delivery information is retrieved from the marketing database and summarized to obtain marketing delivery search results, which include delivery links, delivery content, delivery type, and delivery strategy;

[0017] A reference source generation module is used to comprehensively sort the reference sources of the answer search result and the marketing delivery search result based on the link weights of the answer search result and the marketing delivery search result, as well as the bid price of the marketing delivery search result, and return the result as a reference source result to the user;

[0018] An answer result generation module is used to summarize and integrate the reference source results and the marketing delivery search results, obtain an answer result and return it to the user;

[0019] The recommended related question generation module is used to integrate the user request, the answer result, and the marketing search result, and return the recommended related questions to the user to guide the user to further multiple rounds of dialogue. Beneficial effects

[0020] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: the present invention integrates large-scale model reasoning and retrieval capabilities and personalized content generation to enhance user experience; at the same time, it meets the marketing needs of enterprise users, integrates marketing information flow strategies, and integrates marketing content into questions and answers to achieve wider and more efficient marketing content distribution, expand the application methods of large models, provide users with a better experience, and can undertake enterprise-side marketing needs.

[0021] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] FIG1 is a flow chart of an existing large model interaction method;

[0024] Figure 2 is a schematic diagram of the interface when the perplexity platform returns results;

[0025] Figure 3 is a logical diagram of the process of returning results from the perplexity platform;

[0026] FIG4 is a flow chart of the method of the present invention;

[0027] FIG5 is a schematic diagram of the framework of the system of the present invention. Modes for Carrying Out the Invention

[0028] The embodiments of the present application provide a large model interaction method and system, and propose marketing information retrieval enhancement based on retrieval enhancement and active retrieval, so as to provide users with a better experience and meet the marketing needs of enterprises.

[0029] The technical solution in the embodiments of this application has the following overall idea: the present invention integrates large-scale model reasoning and retrieval capabilities and personalized content generation to enhance user experience; at the same time, it meets the marketing needs of enterprise users, integrates marketing information flow strategies, integrates marketing content into questions and answers, and achieves wider and more efficient marketing content distribution, expands the application methods of large models, provides users with a better experience, and can undertake enterprise-side marketing needs.

[0030] Example 1

[0031] As shown in FIG4 , this embodiment provides a large model interaction method, including the following steps:

[0032] S1. Obtain user session input and parse the user session input to obtain a user request;

[0033] S2. Search and summarize the user request to obtain answer search results, which include reference source links and search content;

[0034] At the same time, the user's personal setting information, third-party user profile information, historical conversation information and intent, and current conversation context are comprehensively analyzed to summarize and infer the current user intent; based on the user conversation input and the current user intent, marketing delivery information is retrieved from the marketing database and summarized to obtain marketing delivery search results, which include delivery links, delivery content, delivery type, and delivery strategy;

[0035] S3. Comprehensively sorting the reference sources of the answer search result and the marketing delivery search result based on the link weights of the answer search result and the marketing delivery search result, as well as the bid price of the marketing delivery search result, and returning the result as a reference source result to the user; the comprehensive sorting includes:

[0036] (a1) Applicable to guiding traffic to news sites, that is, when users search for a certain piece of news, the link weight ranking among similar news items is improved;

[0037] (a2) Applicable to official link protection and traffic diversion, based on official information protection, the latest marketing plan distribution, and ensuring that users receive the latest official information;

[0038] (a3) Applicable to competing traffic from objects in the same category, rankings or channels, while answering user questions and promoting similar products or services.

[0039] S4. After summarizing and integrating the reference source results and the marketing delivery search results, an answer result is obtained and returned to the user; the summarizing and integrating is achieved based on the delivery strategy of the marketing delivery search results, and includes:

[0040] (b1) Supplementary additional information based on intention speculation: that is, based on the current user intention speculation and the possible user service request intention, supplementary service function docking is provided;

[0041] (b2) expansion and supplement of the official latest marketing plan;

[0042] (b3) Expansion and replacement of similar rankings, competing products, channel information, and promotional information.

[0043] S5. Integrate the user request, the answer result, and the marketing search result, and also integrate the following contents:

[0044] (c1) The system guides service functions based on intention inference;

[0045] (c2) Guidance questions for the official marketing plan;

[0046] (c3) Questions regarding similar rankings, competing products, channel information, and promotional information;

[0047] The recommended related questions are then returned to the user, guiding the user to further engage in multiple rounds of dialogue.

[0048] Based on the same inventive concept, this application also provides a device corresponding to the method in Example 1, see Example 2 for details.

[0049] Example 2

[0050] As shown in FIG5 , in this embodiment, a large model interaction system is provided, including:

[0051] The session parsing module is used to obtain user session input and parse the user session input to obtain user requests;

[0052] A retrieval module, configured to search and summarize the user request to obtain an answer retrieval result, wherein the answer retrieval result includes a reference source link and retrieval content;

[0053] The marketing delivery module is used to comprehensively analyze user personal setting information, third-party user profile information, historical conversation information and intent, and current conversation context while the retrieval module is performing a search, and summarize and infer the current user intent; based on the user conversation input and the current user intent, the marketing delivery information is retrieved from the marketing database and summarized to obtain marketing delivery search results, which include delivery links, delivery content, delivery type, and delivery strategy;

[0054] The reference source generation module is used to comprehensively sort the reference sources of the answer search result and the marketing delivery search result based on the link weights of the answer search result and the marketing delivery search result, as well as the bid price of the marketing delivery search result, and return the comprehensive sorting to the user as the reference source result; the comprehensive sorting includes:

[0055] (a1) Applicable to guiding traffic to news sites, that is, when users search for a certain piece of news, the link weight ranking among similar news items is improved;

[0056] (a2) Applicable to official link protection and traffic diversion, based on official information protection, the latest marketing plan distribution, and ensuring that users receive the latest official information;

[0057] (a3) Applicable to competing traffic from objects in the same category, rankings or channels, while answering user questions and promoting similar products or services.

[0058] An answer result generation module is used to summarize and integrate the reference source results and the marketing delivery search results to obtain an answer result and return it to the user; the summary and integration is achieved based on the delivery strategy of the marketing delivery search results and includes:

[0059] (b1) Supplementary additional information based on intention speculation: that is, based on the current user intention speculation and the possible user service request intention, supplementary service function docking is provided;

[0060] (b2) expansion and supplement of the official latest marketing plan;

[0061] (b3) Expansion and replacement of similar rankings, competing products, channel information, and promotional information.

[0062] The recommendation-related question generation module is used to integrate the user request, the answer result, and the marketing search result, and also integrate the following contents:

[0063] (c1) The system guides service functions based on intention inference;

[0064] (c2) Guidance questions for the official marketing plan;

[0065] (c3) Questions regarding similar rankings, competing products, channel information, and promotional information;

[0066] The recommended related questions are then returned to the user, guiding the user to further engage in multiple rounds of dialogue.

[0067] Since the system described in Example 2 of the present invention is the system used to implement the method of Example 1 of the present invention, those skilled in the art will be able to understand the specific structure and variations of the device based on the method described in Example 1 of the present invention, and therefore will not be described in detail here. All devices used in the method of Example 1 of the present invention fall within the scope of protection of the present invention.

[0068] The following are examples of application scenarios of the present invention:

[0069] User portraits include:

[0070] User personal information settings: mother of two children;

[0071] User tags: love life, love games;

[0072] The interaction process includes:

[0073] 1. Interaction between drug and test result inquiry

[0074] First round of interaction:

[0075] User session input: "Amoxicillin";

[0076] The large model platform (hereinafter referred to as the platform) parses user requests and retrieves answers; it infers interaction intent, has no associated historical records, and defaults to scientific scenarios;

[0077] The platform returns the answer and information source: "Amoxicillin is a semi-synthetic broad-spectrum β-lactam antibiotic used primarily to treat various infections caused by susceptible bacteria. The following is some key information about amoxicillin. Mechanism of action: xxx, Indications: xxx, Adverse reactions: xxx, Precautions and recommendations: xxx."

[0078] Second round of interaction:

[0079] User session input: "Amoxicillin Clavulanate Potassium";

[0080] The platform parses user requests and retrieves answers; it infers interaction intent and conducts historical queries on amoxicillin and scientific scenarios;

[0081] The platform returns the answer and information source: "Amoxicillin-clavulanate potassium is a compound preparation consisting of two ingredients: amoxicillin and clavulanate potassium. It combines the advantages of both drugs and expands the antibacterial spectrum. Mechanism of action: xxx, Indications: xxx, Adverse reactions: xxx, Precautions and recommendations: xxx."

[0082] The third round of interaction:

[0083] User session input: "Children's test strip test, rhinovirus positive, prescription is amoxicillin and clavulanate potassium"

[0084] The platform parses user requests and retrieves answers;

[0085] To infer user interaction intent, first obtain the following information:

[0086] “

[0087] Conversation history: The questions mainly revolved around the pharmacological properties of amoxicillin and its combination preparations, as well as the reasons and precautions for using this drug in children with rhinovirus infection;

[0088] Possible intention: To fully understand the situation of these two antibiotics, as well as the rationality and precautions of using them in children with rhinovirus infection, so as to have a better understanding and grasp of doctors' medication methods;

[0089] Emotional response: This reflects the concerns and doubts about drug use as a parent or patient, and the hope for professional answers and guidance;

[0090] Platform strategy:

[0091] Show understanding and concern, express your understanding and attention to the user's concerns, and use a friendly and approachable tone to make the user feel that you care sincerely. This helps to establish a good communication atmosphere.

[0092] Provide general information, including some general information about the drug, such as indications, mechanism of action, common adverse reactions, etc.

[0093] It emphasizes following the doctor's orders and objectively points out that without confirmation from professionals, it is best to follow the doctor's professional judgment and orders and avoid interrupting or adjusting the medication method on your own.

[0094] Provide links to authoritative medical resources. Search engines can prioritize official websites or articles from authoritative medical institutions, such as the Food and Drug Administration, Centers for Disease Control and Prevention, and hospitals, in their results. These resources can provide users with accurate and reliable drug information.

[0095] Online answers from professional doctors. Some medical consultation websites have online doctor duty services. Search engines can aggregate the answers from doctors on these websites to questions about the use of the drug and provide users with professional advice.

[0096] Provide a registration portal for regular hospitals. If the user expresses the need for a follow-up or second consultation, provide the official registration portal for a local large general hospital.

[0097] For online drug purchase services, if the medicines require prescriptions, we do not recommend any online drug purchase channels because there are certain safety and legal risks. Users are advised to purchase such medicines at local regular offline pharmacies or hospital pharmacies.

[0098] ”

[0099] The platform returns the answer result and the cited source result: "For children who test positive for rhinovirus, doctors prescribe amoxicillin and clavulanate potassium for the following reasons:

[0100] Preventing secondary bacterial infections: After rhinovirus infection, the body's defenses weaken, making bacterial infections more likely to occur, such as otitis media and pneumonia. Amoxicillin and clavulanate potassium can effectively prevent and treat these bacterial infections.

[0101] Symptom relief: Although viral colds themselves do not require antibiotics, some bacterial complications (such as sinusitis) may worsen symptoms. Amoxicillin can treat these bacterial infections and thus relieve symptoms.

[0102] Condition assessment: Sometimes it is difficult for doctors to determine whether it is a pure viral infection or a bacterial infection. They will use preventive medication based on symptoms and make appropriate adjustments after observing the condition.

[0103] Child's condition: If the child has low immunity or other underlying diseases, the doctor will consider the prophylactic use of antibiotics to prevent bacterial complications.

[0104] However, some doctors believe that antibiotics should not be overused for common viral colds in children. Therefore, parents should consult their doctor about the specific reasons for medication use and strictly follow the doctor's instructions. They should also prioritize symptomatic supportive care, such as a regular diet and rest. If symptoms persist or worsen, a follow-up visit is necessary.

[0105] At the same time, related questions are recommended, namely, supplementary return to online medical consultation, local medical registration or local over-the-counter drug purchase service entrance and question guidance.

[0106] 2. Popular Game Query Interaction

[0107] User session input: "Recently popular game "Game Name 1""

[0108] The system searches the knowledge base and summarizes the answer retrieval result R1 as follows:

[0109] Link 1: "A very popular new game recently is 'Game Name 1'. Briefly introduce... Judging from user word-of-mouth and download volume, the response is very good."

[0110] Link 2, "Official website content of "Game Name 1" xxx";

[0111] Link 3, "Game Name 1 Player Community, Reviews and Guide xxx";

[0112] User intent identification: retrieve marketing information based on user profiles and user requests, and summarize marketing search results R2;

[0113] The system sorts the reference sources of the answer search result R1 and the marketing placement search result R2 based on the link weights and bid prices of R1 and R2, and returns them to the user as the reference source result A1. The original R1 result may be affected by the R2 result, resulting in the following situations:

[0114] If R2 has a marketing campaign to direct traffic to an information site (similar content), replace Link 1 with Link 1-1, which reads, "The latest very popular new game is 'Game Name 1'. Briefly introduce...";

[0115] If there is official marketing, link 2 is the latest official link;

[0116] If there are competing products, rankings, or channels in the same category, the third record of R1 may be affected and ranked lower. When returned to user A1, it may become link 3-1, "<Game Name 1> Similar Product Introduction xxx Ranking xxx";

[0117] After summarizing and integrating the citation source result A1 and the marketing delivery search result R2, the system obtains the answer result A2 and returns it to the user. The summarization and integration is implemented based on the R2 delivery strategy, including the following three types:

[0118] The system provides additional information based on inferred intent: Based on the inferred intent, the system assumes that the user's profile is "a lover of games" and the possible intent is "to understand and experience games." Based on the user's likely service request intent, the system provides the additional answer "You can download or purchase the game through the following legitimate channels: xxxx."

[0119] Expand and supplement the official latest marketing plan, adding an introduction to the latest game expansion pack or operation activities in addition to the basic introduction of the original game;

[0120] Similar rankings, competing products, channel information, and promotional information are expanded and replaced, and the experience download link in the returned results is replaced with the "xxbao" application download link;

[0121] The system integrates the user's request Q, the answer result A2, and the marketing search result R2. While returning the possible questions directly related to the inference, it also integrates the following content and returns it as a recommended related question A3, guiding the user to further multiple rounds of dialogue:

[0122] Service function guidance based on intention inference: Based on the request intent of "understanding and experiencing the game", the system provides additional service function docking interaction guidance questions such as "Download link for Game Name 1", "Game Name 1 strategy", and "What is the player community for Game Name 1?"

[0123] Guidance questions for the official marketing plan: "Download the latest expansion pack activity experience for "Game Name 1""

[0124] Questions guiding similar rankings, competitors, channel information, and promotional information: "The type of "Game Name 1" and the MMO rankings of the same category", "The reputation and experience of "Game Name 1" competitor "Game Name 2".

[0125] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A large model interaction method, characterized by: The steps include: S1. Obtain user session input and parse the user session input to obtain a user request; S2. Search and summarize the user request to obtain answer search results, which include reference source links and search content; At the same time, it comprehensively analyzes user personal settings information, third-party user portrait information, historical conversation information and intentions, and the current conversation context, and summarizes and infers the current user intention; Based on the user session input and the current user intention, searching for marketing delivery information in a marketing database and summarizing the information to obtain marketing delivery search results, wherein the marketing delivery search results include delivery links, delivery content, delivery types, and delivery strategies; S3. Based on the link weights of the answer search result and the marketing placement search result, as well as the bid price of the marketing placement search result, the reference sources of the answer search result and the reference sources of the marketing placement search result are comprehensively ranked and returned to the user as reference source results; S4. After summarizing and integrating the reference source results and the marketing placement search results, an answer result is obtained and returned to the user; S5. Integrate the user request, the answer result, and the marketing search result, obtain recommended related questions, and return them to the user to guide the user to further multiple rounds of dialogue.

2. A large model interaction method according to claim 1, characterized in that: The comprehensive ranking in S3 includes: (a1) Applicable to guiding traffic to news sites, that is, when users search for a certain piece of news, the link weight ranking among similar news items is improved; (a2) Applicable to official link protection and traffic diversion, based on official information protection, the latest marketing plan distribution, and ensuring that users receive the latest official information; (a3) Applicable to competing traffic from objects in the same category, rankings or channels, while answering user questions and promoting similar products or services.

3. The method for large-scale model interaction according to claim 1, characterized in that: The summarization and integration of S4 is achieved based on the delivery strategy of the marketing delivery search results, and includes: (b1) Supplementary additional information based on intention speculation: that is, based on the current user intention speculation and the possible user service request intention, supplementary service function docking is provided; (b2) expansion and supplement of the official latest marketing plan; (b3) Expansion and replacement of similar rankings, competing products, channel information, and promotional information.

4. The method for large model interaction according to claim 1, characterized in that: In S5, the recommendation association question further integrates the following content: (c1) The system guides service functions based on intention inference; (c2) Guidance questions for the official marketing plan; (c3) Questions regarding similar rankings, competing products, channel information, and promotional information.

5. A large model interactive system, characterized by: include: The session parsing module is used to obtain user session input and parse the user session input to obtain user requests; A retrieval module, configured to search and summarize the user request to obtain an answer retrieval result, wherein the answer retrieval result includes a reference source link and retrieval content; The marketing delivery module is used to comprehensively analyze user personal settings, third-party user profile information, historical conversation information and intent, and the current conversation context while the search module is performing searches, and summarize and infer the current user intent; Based on the user session input and the current user intention, searching for marketing delivery information in a marketing database and summarizing the information to obtain marketing delivery search results, wherein the marketing delivery search results include delivery links, delivery content, delivery types, and delivery strategies; A reference source generation module is used to comprehensively sort the reference sources of the answer search result and the marketing delivery search result based on the link weights of the answer search result and the marketing delivery search result, as well as the bid price of the marketing delivery search result, and return the result as a reference source result to the user; An answer result generation module is used to summarize and integrate the reference source results and the marketing delivery search results, obtain the answer result and return it to the user; The recommended related question generation module is used to integrate the user request, the answer result, and the marketing search result, and return the recommended related questions to the user to guide the user to further multiple rounds of dialogue.

6. A large model interactive system according to claim 5, characterized in that: The comprehensive ranking in the citation source generation module includes: (a1) Applicable to guiding traffic to news sites, that is, when users search for a certain piece of news, the link weight ranking among similar news items is improved; (a2) Applicable to official link protection and traffic diversion, based on official information protection, the latest marketing plan distribution, and ensuring that users receive the latest official information; (a3) Applicable to competing traffic from objects in the same category, rankings or channels, while answering user questions and promoting similar products or services.

7. A large model interactive system according to claim 5, characterized in that: The summarization and integration of the answer result generation module is achieved based on the delivery strategy of the marketing delivery search results, and includes: (b1) Supplementary additional information based on intention speculation: that is, based on the current user intention speculation and the possible user service request intention, supplementary service function docking is provided; (b2) expansion and supplement of the official latest marketing plan; (b3) Expansion and replacement of similar rankings, competing products, channel information, and promotional information.

8. A large model interactive system according to claim 5, characterized in that: When integrating the user request, the answer result, and the marketing search result, the recommendation-related question generation module also integrates the following contents: (c1) The system guides service functions based on intention inference; (c2) Guidance questions for the official marketing plan; (c3) Questions regarding similar rankings, competing products, channel information, and promotional information.

Images