A method and system for completing a search according to a fuzzy meaning label using an AI large model

Abstract

The application discloses a search method and system for completing fuzzy meaning labels by using an AI large model, relates to the technical field of fuzzy search, and comprises the following steps: a prompt word template is customized for guiding the AI large model to perform semantic matching according to user fuzzy input and labels in a RAG knowledge base by clearly defining scene requirements and targets; and the search intention of the user is recognized according to the customized prompt word template by combining the AI large model and an agent technology. The technical design of the application is more humanized and reasonable, the search result is more diverse, and the user has more flexibility when selecting a promotion strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fuzzy search technology, specifically, a method and system for completing search according to fuzzy meaning labels using AI large models. BACKGROUND

[0002] In our daily life and work, search function is ubiquitous. Whether in web search, file retrieval, or database query, we often cannot find the desired results due to spelling errors or incomplete information. Fuzzy search emerges as the times require, which helps us obtain more flexible search results by identifying similar words to the query. This article will discuss the background, principle, implementation and application of fuzzy search in detail, helping readers to better understand this basic and practical technology. The emergence of fuzzy search solves the limitations of traditional exact search. Usually, exact search will strictly match user input and data content, ignoring spelling errors, similar words or complex query conditions. However, in practical applications, data is often not completely consistent, such as name, address, product description and other fields may have spelling errors, abbreviations, pinyin differences, etc. The emergence of fuzzy search makes up for this matching gap, making the search function more "fault-tolerant". As a basic part of search technology, fuzzy search is widely used in various scenarios due to its strong fault tolerance and high efficiency. However, in today's increasingly complex demand, the limitations of fuzzy search have gradually emerged, especially in deep semantic understanding and complex queries. SUMMARY

[0003] In order to solve the above problems, the purpose of the present application is to provide a search technology according to fuzzy meaning labels using AI large models, aiming to make the search results more diverse, so as to overcome the limitations of fuzzy search.

[0004] In order to achieve the above technical purpose, the present application provides a method for completing search according to fuzzy meaning labels using AI large models, comprising the following steps:

[0005] By specifying the scene demand and target, customizing the prompt word template, which is used to guide the AI large model to perform semantic matching according to the user's fuzzy input and the labels in the RAG knowledge base;

[0006] Combined with AI large model and agent technology, the search intention of the user is identified according to the customized prompt word template.

[0007] Preferably, in customizing the prompt template, the input fuzzy search term is obtained, matched with the labels in the RAG knowledge base, the initial prompt template is customized according to the matching result, and the initial template is input into the AI large model to test its matching effect on multiple fuzzy inputs. After optimizing the prompt according to the matching result, the prompt template is generated to guide the model to complete the matching of fuzzy input to labels and obtain the most suitable label, which is associated with the red content.

[0008] Preferably, in optimizing the prompt, if the model tends to match too broadly, add constraints to the prompt to prefer the most specific label; if the model misses relevant labels, add instructions to try to expand the semantics if there is no direct match, consider synonyms or hypernyms.

[0009] Preferably, in customizing the prompt template, the prompt template is customized according to the user input, the reference of the knowledge base label, and the instruction of the matching logic.

[0010] Preferably, in identifying the search intent of the user, after preprocessing the fuzzy search term input by the user, the AI large model is used for preliminary semantic analysis to make a preliminary intent judgment in the possible semantic space, and a candidate intent is generated.

[0011] Based on the agent technology combined with the RAG knowledge base, the label matching is performed according to the candidate intent.

[0012] The AI large model is controlled to retrieve relevant red content from the knowledge base according to the label matching result and send it to the user.

[0013] Preferably, in preprocessing the fuzzy search term, the fuzzy search term is preprocessed through text cleaning, spelling correction, and language detection.

[0014] Preferably, in the label matching, the agent technology is used for task decomposition and knowledge base interaction to extract all available labels from the RAG knowledge base, and the embedding capability of the AI large model is called to convert the user input and the knowledge base label into a vector to calculate the similarity. The most relevant label is selected according to the similarity ranking, and when the user has subsequent input, the matching label is adjusted by the agent technology in combination with the context.

[0015] Preferably, in retrieving relevant red content from the knowledge base according to the label matching result, if there is no completely matched label, the closest label is selected by the agent technology, and the AI large model generates an explanation.

[0016] Preferably, after identifying the search intention of the user, the matching effect is evaluated according to the user behavior, and if the user click rate is low, it indicates that the label matching is inaccurate, and the following optimization needs to be carried out: by introducing more synonym expansion, adjusting the semantic matching algorithm of the large model; by combining user historical preferences, improving the ranking logic of the agent; updating the labels in the RAG knowledge base, and increasing more fine-grained labels, wherein the user behavior includes: whether to click the recommended red person or subsequent input exists.

[0017] The application discloses a search system for completing fuzzy meaning label search by using an AI large model, which is used for realizing the search method for completing fuzzy meaning label search by using an AI large model mentioned above, and comprises the following steps of:

[0018] A semantic matching module is used for customizing a prompt word template by determining a scene demand and a target, and is used for guiding the AI large model to perform semantic matching according to user fuzzy input and labels in the RAG knowledge base.

[0019] An intention recognition module is used for combining the AI large model and the agent technology, recognizing the search intention of the user according to the customized prompt word template.

[0020] The application discloses the following technical effects:

[0021] The technical design of the application is more humanized and reasonable, and makes the search results more rich and diverse, and also makes the user have more flexibility when selecting a promotion strategy. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0023] Figure 1 The figure is a method flowchart of the application. DETAILED DESCRIPTION

[0024] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0025] As Figure 1 indicated, the present application provides a search technology according to fuzzy meaning labels using an AI large model, specifically including the following contents:

[0026] How to get "customized prompt word templates"?

[0027] "Customized prompt word templates" are used to guide the AI large model to perform semantic matching according to user fuzzy input (such as "pet") and labels in the RAG knowledge base (such as "cat and dog"), find the most suitable label, and associate it to the red person content. In order to get such prompt word templates, the following steps can be followed:

[0028] Clarify the scene requirements and goals:

[0029] -Scene goal: User inputs fuzzy search words (such as "pet" and "dog"), the system needs to match labels in the RAG knowledge base (such as "cat and dog"), and input the matched labels to the next business flow according to the requirements.

[0030] -The core problem that the prompt word template needs to solve: how to make the model understand the semantic of user input and establish the association with the labels in the knowledge base.

[0031] -For example, the template needs to include user input, reference to knowledge base labels, and instructions for matching logic.

[0032] Initialize the prompt word template:

[0033] -Design an initial prompt word template according to the analysis results to guide the model to complete the matching of fuzzy input to labels. For example:

[0034] As an intelligent search assistant, the user inputs a vague search term: {user input}. Please match the most relevant label from the list of labels in the RAG knowledge base: {knowledge base label list}, and explain why this label is chosen. If there is no exact match, recommend the closest label.

[0035] - Example Filling:

[0036] - User Input: "pet"

[0037] - Knowledge Base Label List: "Cat and Dog" "Pet Routine" "Dog Raising Tips"

[0038] - Model Possible Output: "The most relevant label is 'Cat and Dog' because 'pet' semantically refers to pets, which is highly related to 'Cat and Dog'."

[0039] Testing and Optimization:

[0040] - Input the initial template into the AI large model and test its matching effect on various vague inputs. For example:

[0041] - Can "dog" match "Cat and Dog" or "Dog Raising Tips"?

[0042] - Can "catlover" match "Cat and Dog"?

[0043] - According to the test results, optimize the prompt words, for example:

[0044] - If the model tends to match too broadly, add constraints to the prompt words: "Prioritize the most specific label."

[0045] - If the model misses relevant labels, add instructions: "If there is no direct match, try to expand the semantics, consider synonyms or superordinate words."

[0046] How to implement "Identify user search intent with AI large model and Agent technology"?

[0047] To identify user search intent with AI large model and Agent technology, the core goal is to understand the semantic intent of user's vague input (such as "pet" "dog") and match the most relevant red content through the labels in the RAG knowledge base (such as "Cat and Dog"). The specific implementation steps are as follows:

[0048] User input preprocessing and preliminary analysis:

[0049] - The system receives user's vague search term (such as "pet") and preprocesses it, including:

[0050] - Text cleaning: Removing extra spaces, punctuation, etc.

[0051] - Spelling correction: Correcting "doog" to "dog".

[0052] - Language detection: Ensuring the input language is consistent with the knowledge base (e.g., English, Chinese).

[0053] - Pre-processed input is fed into an AI large model for preliminary semantic analysis.

[0054] AI large model identifies search intent:

[0055] - AI large model (based on pre-trained language models like BERT or LLaMA) performs semantic understanding of user input, making an initial judgment of intent.

[0056] - In this scenario, the intent is usually "find relevant Red Person content", but the specific sub-intent could be "find Red Person related to pets" or "find Red Person related to dogs".

[0057] - The large model maps input words (like "pet") to possible semantic spaces through its semantic understanding capabilities, generating candidate intents. For example:

[0058] - Input "pet" → Intent: Find pet-related content.

[0059] - Input "dog" → Intent: Find dog-related content.

[0060] Agent technology combined with RAG knowledge base for label matching:

[0061] - Agent module intervenes, responsible for task decomposition and knowledge base interaction. Execution steps are as follows:

[0062] - Retrieve RAG knowledge base: Agent extracts all available labels from the RAG knowledge base (e.g., "cats and dogs", "pet daily routine", "dog raising skills").

[0063] - Semantic matching: Agent calls the embedding capabilities of the large model to convert user input (e.g., "pet") and knowledge base labels into vectors, calculating similarity (e.g., cosine similarity). For example:

[0064] - The vector similarity of "pet" and "cats and dogs" may be 0.85.

[0065] - The similarity of "pet" and "dog raising skills" may be 0.65.

[0066] - Label sorting: Agent sorts according to similarity, selecting the most relevant label (e.g., "cats and dogs").

[0067] - Context Management: If the user has subsequent input (e.g., "pet" followed by "food"), the agent adjusts the matching label in context, possibly prioritizing "pet daily" over "cats and dogs."

[0068] Large Model Collaborates with Agent to Generate Final Results:

[0069] - The agent sends the matched label (e.g., "cats and dogs") back to the large model.

[0070] - The large model retrieves relevant Red Person content from the knowledge base based on the label and generates user-friendly output. For example:

[0071] - Input: "pet"

[0072] - Matched Label: "cats and dogs"

[0073] - Output: "I found 'cats and dogs' related content for you. Here are some popular Red Persons: @Pet Expert A, @Cats and Dogs Daily B, who share a lot of pet-related content!"

[0074] - If there is no exact match, the agent selects the closest label and the large model generates an explanation. For example:

[0075] - Input: "cutekitten"

[0076] - Matched Label: "cats and dogs"

[0077] - Output: "I found the closest label 'cats and dogs' for you. Here are the relevant Red Persons: @Cat Enthusiast C, @Cats and Dogs Daily B."

[0078] User Feedback and System Optimization:

[0079] - The system records user behavior (e.g., whether they click on recommended Red Persons, subsequent input) to evaluate matching effectiveness.

[0080] - If user click-through rate is low, it may indicate inaccurate label matching, which needs to be optimized:

[0081] - Adjust the semantic matching algorithm of the large model (e.g., introduce more synonym extensions).

[0082] - Improve the agent's sorting logic (e.g., incorporate user historical preferences).

[0083] - Update the RAG knowledge base with more granular labels (e.g., "cat" and "dog").

[0084] For example, when we search for "brand conditioning probiotics", not only will the associated notes of "brand conditioning" and "conditioning probiotics" be displayed, but also the associated notes of "brand probiotics" and other related words will be displayed. The reason behind this change is that the search is transitioning from keyword inclusion to AI algorithm understanding inclusion. That is, whether a note is included by a keyword no longer depends on whether the note contains the keyword, but on how the AI algorithm understands the note. The present application does not need to repeatedly add keywords to ensure that the note is included, but only needs to ensure that the topic is related to the keyword, which is undoubtedly more humanized and reasonable, so that the search results are more diverse, and at the same time, the user has more flexibility when choosing a promotion strategy.

[0085] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 The device that implements the function specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the function specified in one flow or multiple flows and / or blocks

[0086] In the description of the present application, it should be understood that the terms "first", "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0087] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

Claims

1. A method for completing a search according to a vague sense label using an AI large model, characterized by, The method comprises the following steps: Customize the prompt template to guide the AI large model to perform semantic matching according to the user's fuzzy input and the labels in the RAG knowledge base by clearly defining the scene requirements and targets; Combine the AI large model and the agent technology to identify the user's search intent according to the customized prompt template; When identifying the user's search intent, the AI large model is used to perform preliminary semantic analysis on the preprocessed fuzzy search words input by the user, make a preliminary intent judgment in the possible semantic space, and generate candidate intents; Based on the agent technology and the RAG knowledge base, the candidate intents are matched with the labels; The AI large model is controlled to retrieve relevant content shared by network celebrities from the knowledge base according to the label matching results and send the content to the user; When matching the labels, the agent technology is used to perform task decomposition and knowledge base interaction, extract all available labels from the RAG knowledge base, convert the user input and the knowledge base labels into vectors by using the embedding capability of the AI large model, calculate the similarity, and select the most relevant label according to the similarity ranking. When the user has subsequent input, the agent technology is used to adjust the matching labels in combination with the context. After identifying the user's search intent, the matching effect is evaluated according to the user's behavior. If the user's click rate is low, it indicates that the label matching is inaccurate, and the following optimizations need to be performed: introduce more synonyms to expand the semantic matching algorithm of the large model; improve the sorting logic of the agent by combining the user's historical preferences; update the labels in the RAG knowledge base to add more fine-grained labels. The user's behavior includes whether to click the recommended content shared by network celebrities or have subsequent input.

2. The method according to claim 1, wherein: When customizing the prompt template, the input fuzzy search words are matched with the labels in the RAG knowledge base, an initial prompt template is customized according to the matching results, the initial prompt template is input into the AI large model to test its matching effect on various fuzzy inputs, the prompt template is generated after optimizing the prompt according to the matching results, and the prompt template is used to guide the model to complete the matching of fuzzy input to labels and obtain the most suitable label associated with the content shared by network celebrities.

3. The method according to claim 2, wherein: When optimizing the prompt, if the model tends to match too broadly, add a constraint to the prompt to prefer the most specific label; if the model misses relevant labels, add an instruction: if there is no direct match, try to expand the semantics, consider synonyms or superordinate words.

4. The method according to claim 3, wherein: When customizing the prompt template, the prompt template is customized according to the user input, the reference of the knowledge base labels, and the instructions of the matching logic.

5. The method according to claim 1, wherein the method comprises: preprocessing the fuzzy search term by text cleaning, spelling correction and language detection.

6. The method according to claim 1, wherein the method comprises: when there is no completely matched label, selecting the closest label by the agent technology, and generating an explanation by the AI large model. The method comprises: a semantic matching module for customizing a prompt word template by clearly defining the scene demand and target, and guiding the AI large model to perform semantic matching according to the user's fuzzy input and the label in the RAG knowledge base; 7.A search system for completing a search by a fuzzy meaning tag using an AI large model, for implementing the search method for completing a search by a fuzzy meaning tag using an AI large model according to claim 1, characterized by, an intent recognition module for combining the AI large model and the agent technology, and recognizing the user's search intent according to the customized prompt word template. ​ ​

Images