Visual retrieval method and system based on AI

Abstract

The invention relates to the technical field of intelligent retrieval, and particularly provides an AI-based visual retrieval method and system, and the method comprises the steps: obtaining scene demand data of a user and a user portrait tag; carrying out vectorization processing on the scenarized demand data, and carrying out multi-modal retrieval through a pre-constructed multi-modal database to obtain multi-modal retrieval data; meanwhile, based on the user portrait label, processing the multi-modal retrieval data to generate a personalized retrieval result dynamically associated with the user portrait label; and mapping the personalized retrieval result to a visual map for display. According to the method, multi-modal data such as texts, images and voices can be processed at the same time, the retrieval strategy can be updated and dynamically adjusted in real time through the user portrait label, the matching degree of the retrieval result and the real demand of the user is improved, and the industrial pain points that products / schemes are difficult to accurately position, and contents cannot be recommended according to different persons are solved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent retrieval technology, specifically relating to an AI-based visual retrieval method and system. Background Technology

[0002] With the rapid development of information technology, users' demands for search systems have gradually shifted from simple keyword matching to scenario-based, personalized, and efficient interaction. However, existing search technologies still have the following significant shortcomings: Traditional search systems generally employ keyword-based matching mechanisms, which can only process text-based data and have weak support for multimodal data such as images, audio, and video, making it difficult to meet users' precise search needs. Furthermore, traditional search systems offer limited personalized recommendations, failing to tailor content to individual users, requiring them to repeatedly adjust keywords or filter results, resulting in low search efficiency. Summary of the Invention

[0003] To overcome the shortcomings of the prior art, this invention proposes an AI-based visual retrieval method and system, specifically implemented through the following technical solution: An AI-based visualization retrieval method includes: Acquire user-specific needs data and user profile tags; The scenario-based requirement data is vectorized and then multimodal retrieved using a pre-built multimodal database to obtain multimodal retrieval data. Simultaneously, based on the user profile tags, the multimodal retrieval data is processed to generate personalized retrieval results dynamically associated with the user profile tags. The personalized search results are mapped to a visualization map for display.

[0004] In one specific embodiment, the user profile tags are obtained through the following steps: Once user behavior data is acquired, it is processed and analyzed based on a large language model to identify user role types and user behavior characteristics. Based on the user role type and the user behavior characteristics, the user profile tags are dynamically generated.

[0005] In one specific embodiment, the contextualized requirement data includes label requirement data described in natural language; The step of vectorizing the scenario-based requirement data and performing multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data includes: The tag requirement data is vectorized to obtain the requirement vector to be matched; A pre-built multimodal database is invoked, which stores multiple preset label requirements and a preset label requirement vector corresponding to each preset label requirement; The required vector to be matched is matched with each of the preset tag required vectors, and multiple required vectors to be matched that meet the preset conditions are selected from the multimodal database as the multimodal retrieval data to obtain multimodal retrieval data.

[0006] In one specific embodiment, the multimodal database includes product data, semantic data, or case data; or the multimodal database includes text data, graphic data, and tabular data.

[0007] In one specific embodiment, the step of processing the multimodal retrieval data based on the user profile tags to generate personalized retrieval results dynamically associated with the user profile tags includes: Based on the user profile tags, the multimodal retrieval data is processed and analyzed using a large language model to obtain multimodal retrieval data that matches the user profile tags and response templates that correspond to the user profile tags. Based on the multimodal retrieval data matching the user profile tags and the response templates corresponding to the user profile tags, personalized retrieval results dynamically associated with the user profile tags are generated.

[0008] In one specific embodiment, different user profile tags correspond to different response templates; The process of processing and analyzing multimodal retrieval data matching the user profile tags and response templates corresponding to the user profile tags to generate personalized retrieval results dynamically associated with the user profile tags includes: Based on the multimodal retrieval data matching the user profile tags and the response templates corresponding to the user profile tags, the system processes and analyzes the data to generate product recommendation data or demand solutions that are dynamically associated with the user profile tags, and uses the product recommendation data or demand solutions as the personalized search results.

[0009] In one specific embodiment, the visualization map includes multiple different identification images, each of which corresponds to information about a product type or manufacturer; The step of mapping the personalized search results to a visualization map for display includes: Obtain the product type or manufacturer corresponding to the personalized search results, and select the identifier image of the visualization map based on the product type or manufacturer corresponding to the personalized search results; Based on the selected identifier image, the personalized search results are mapped and the selected identifier image is highlighted.

[0010] In one specific embodiment, it further includes: The number of times a user selects an image from the visualized map within a certain period of time is summarized. Based on the number of times a user selects an image in the visualization map within a certain period of time, the user profile tags are adjusted, and personalized search results are updated based on the adjusted user profile tags.

[0011] An AI-based visual retrieval system includes: The acquisition module is used to acquire users' contextualized needs data and user profile tags; The multimodal intelligent retrieval module is used to vectorize the scenario-based demand data and perform multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data; at the same time, based on the user profile tags, the multimodal retrieval data is processed to generate personalized retrieval results that are dynamically associated with the user profile tags; The presentation module is used to map the personalized search results to a visualization map for display.

[0012] In one specific embodiment, the acquisition module includes: The identification module is used to process and analyze user behavior data based on a large language model when user behavior data is acquired, and to identify user role types and user behavior characteristics. The generation module is used to dynamically generate the user profile tags based on the user role type and the user behavior characteristics.

[0013] The present invention has at least the following beneficial effects: This application relates to the field of intelligent retrieval technology, specifically providing an AI-based visual retrieval method and system. The method includes acquiring user-defined scenario-based demand data and user profile tags; vectorizing the scenario-based demand data and performing multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data; simultaneously processing the multimodal retrieval data based on user profile tags to generate personalized retrieval results dynamically associated with the user profile tags; and mapping the personalized retrieval results to a visualization graph for display. This application achieves personalized customization of retrieval results through a dynamic association mechanism between a multimodal database and user profile tags. Compared to traditional retrieval systems that rely solely on keyword matching, this application can simultaneously process multimodal data such as text, images, and voice, and dynamically adjust retrieval strategies through real-time updates of user profile tags, improving the matching degree between retrieval results and users' actual needs. This solves industry pain points such as difficulty in accurately locating products / solutions and the inability to recommend content tailored to individual users.

[0014] In practical applications, the AI-based visualization retrieval method and system proposed in this application are applicable to multiple user roles across the sensor industry chain, including but not limited to: sensor procurement personnel, solution engineers, researchers, system integrators, manufacturers, and OEMs. By combining natural language interaction, semantic understanding, multimodal vector retrieval, and visualization, different personalized search results are recommended to different user roles. For example, product recommendation lists and reasons are recommended to sensor procurement personnel, and solution requirements are recommended to solution engineers. This solves industry pain points such as the difficulty in accurately locating products / solutions and the inability to recommend content tailored to individual users. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The process of an AI-based visual retrieval method Figure 1 ; Figure 2 The process of an AI-based visual retrieval method Figure 2 ; Figure 3 The process of an AI-based visual retrieval method Figure 3 ; Figure 4 The process of an AI-based visual retrieval method Figure 4 ; Figure 5 This is a schematic diagram of a module of an AI-based visual retrieval system.

[0017] Figure label: 1-Acquisition module; 2-Multimodal intelligent retrieval module; 3-Presentation module. Detailed Implementation

[0018] Various embodiments of the invention will be described more fully below. The invention may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the invention to the specific embodiments disclosed herein, but rather the invention should be understood to cover all modifications, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of the invention.

[0019] In the following, the terms “comprising” or “may include” as used in various embodiments of the invention indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of the invention, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0020] Example 1 like Figure 1 As shown, this application relates to the field of intelligent retrieval technology, specifically providing an AI-based visual retrieval method. This method includes acquiring user-defined scenario-based demand data and user profile tags; vectorizing the scenario-based demand data and performing multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data; simultaneously processing the multimodal retrieval data based on user profile tags to generate personalized retrieval results dynamically associated with the user profile tags; and mapping the personalized retrieval results to a visualization graph for display. This application achieves personalized customization of retrieval results through a dynamic association mechanism between a multimodal database and user profile tags. Compared to traditional retrieval systems that rely solely on keyword matching, this application can simultaneously process multimodal data such as text, images, and voice, and can dynamically adjust retrieval strategies through real-time updates of user profile tags, improving the matching degree between retrieval results and users' actual needs. This solves industry pain points such as difficulty in accurately locating products / solutions and the inability to recommend content tailored to individual users.

[0021] In practical applications, the AI-based visualization retrieval method and system proposed in this application are applicable to multiple user roles across the sensor industry chain, including but not limited to: sensor procurement personnel, solution engineers, researchers, system integrators, manufacturers, and OEMs. By combining natural language interaction, semantic understanding, multimodal vector retrieval, and visualization, different personalized search results are recommended to different user roles. For example, product recommendation lists and reasons are recommended to sensor procurement personnel, and solution requirements are recommended to solution engineers. This solves industry pain points such as the difficulty in accurately locating products / solutions and the inability to recommend content tailored to individual users.

[0022] like Figure 2 As shown, user profile tags are obtained through the following steps: Once user behavior data is acquired, it is processed and analyzed based on a large language model to identify user role types and user behavior characteristics. User profile tags are dynamically generated based on user role type and user behavior characteristics.

[0023] This application uses a large language model to conduct in-depth analysis of user behavior data, automatically identify user role types and behavioral characteristics, and dynamically generate differentiated tags, providing a foundation for subsequent personalized retrieval.

[0024] In practical applications, the user roles in the sensor industry chain differ significantly (e.g., procurement personnel focus on cost and supply chain stability, while researchers emphasize technical parameters and cutting-edge research). This invention utilizes a large language model for deep analysis of user behavior data to automatically identify the user's role type (procurement personnel or researchers) and extract behavioral characteristics (e.g., procurement personnel frequently search for "sensor price range," while researchers frequently download "MEMS technology white papers"). Based on user role type and behavioral characteristics, it dynamically generates user profile tags (e.g., price-sensitive, multi-project leader, high-tech hurdle adaptable, etc.) to achieve accurate multi-role identification and demand decoupling. This allows for real-time adjustment of search strategies through dynamically generated user profile tags to meet the needs of different scenarios.

[0025] Specifically, user behavior data is obtained through login information and user actions. Login information includes the user's entered role, company type, and job title; user actions include the user's browsing path and AI semantic search.

[0026] like Figure 3 As shown, the scenario-based requirement data includes label requirement data described in natural language; Contextualized requirement data is vectorized and then retrieved using a pre-built multimodal database to obtain multimodal retrieval data, including: The tag requirement data is vectorized to obtain the requirement vector to be matched; Call the pre-built multimodal database, which stores multiple preset label requirements and the preset label requirement vector corresponding to each preset label requirement; The demand vector to be matched is matched with each preset tag demand vector, and multiple demand vectors to be matched that meet the preset conditions are selected from the multimodal database as multimodal retrieval data to obtain multimodal retrieval data.

[0027] This application vectorizes the tag demand data to obtain the demand vector to be matched, thereby capturing the implicit semantic relationships in natural language. At the same time, it matches the demand vector to be matched with multi-dimensional tag demand vectors such as text, parameters, and application scenarios in a multi-modal database to obtain multi-modal retrieval data, thereby greatly improving the efficiency and accuracy of retrieval.

[0028] Among them, several unmatched demand vectors that meet the preset matching conditions are selected in descending order of their ranking.

[0029] Among them, multimodal databases include product data, semantic data, or case data; or multimodal databases include text data, graphical data, and tabular data.

[0030] Traditional databases often focus on single data types (such as text or parameters only), making it difficult to meet the multi-dimensional information needs of users in the sensor industry chain. This invention integrates product data (specifications, supply chain information), semantic data (technical terms, related concepts), and case data (application scenarios, solutions) to construct a multimodal database. This allows purchasing personnel to quickly compare parameters such as range, accuracy, and price in product data; in practical applications, researchers can understand the technical relationship between MEMS sensors and the piezoelectric effect through semantic data; and solution engineers can refer to case data to obtain solutions to their needs, thus achieving precise retrieval. Similarly, the database supports unified storage and retrieval of text, graphic, and tabular data, solving the problem of traditional databases' weak support for unstructured data (such as drawings and test reports) and improving retrieval accuracy.

[0031] like Figure 4 As shown, based on user profile tags, multimodal retrieval data is processed to generate personalized retrieval results dynamically associated with user profile tags, including: Based on user profile tags, a large language model is used to process and analyze multimodal retrieval data to obtain multimodal retrieval data that matches user profile tags and response templates that correspond to user profile tags. Based on the processing and analysis of multimodal retrieval data matched with user profile tags and the response templates corresponding to user profile tags, personalized search results dynamically associated with user profile tags are generated.

[0032] This application utilizes a large language model to deeply analyze multimodal retrieval data, improving the accuracy of demand matching. Furthermore, targeting the differentiated needs of users across the sensor industry chain, the system automatically selects response templates based on user profile tags. It processes and analyzes multimodal retrieval data matching user profile tags and corresponding response templates to generate personalized search results dynamically associated with user profile tags. This solves the problems of coarse demand analysis and insufficient personalization in traditional methods, providing more accurate and efficient search services for users in the sensor industry chain, while reducing system maintenance costs and improving technical scalability.

[0033] Different user profile tags correspond to different response templates; Based on the processing and analysis of multimodal retrieval data matched with user profile tags and the response templates corresponding to those tags, personalized search results dynamically associated with user profile tags are generated, including: Based on the multimodal retrieval data matched with user profile tags and the response templates corresponding to user profile tags, the system processes and analyzes the data to generate product recommendation data or demand solutions that are dynamically associated with user profile tags, and uses the product recommendation data or demand solutions as personalized search results.

[0034] This invention solves the problems of template staticization and demand adaptation lag in traditional methods by using a dynamic association mechanism between user profile tags and response templates, combined with the deep parsing capabilities of large language models. It provides users in the sensor industry chain with more accurate and efficient personalized search services, while reducing system maintenance costs and improving technical scalability.

[0035] The visualization map includes multiple different logo images, each of which corresponds to information about the product type or manufacturer. Mapping personalized search results to a visualization graph for display includes: Obtain the product type or manufacturer corresponding to the personalized search results, and select the identifier image of the visualization map based on the product type or manufacturer corresponding to the personalized search results; Based on the selected logo image, personalized search results are mapped and the selected logo image is highlighted.

[0036] Traditional search results are presented in list format, requiring users to interpret each piece of text information. This invention maps product types or manufacturers to identifiable images using a visual graph, and combines recommendation levels with highlighting technology, allowing users to easily identify key information at a glance. This provides a more intuitive and efficient decision support tool for users in the sensor industry chain.

[0037] AI-based visualization retrieval methods also include: Summarize the number of times users select the labeled images in the visualization map within a certain period of time; Based on the number of times a user selects an image in a visualization over a certain period of time, the user profile tags are adjusted, and personalized search results are updated based on the adjusted user profile tags.

[0038] Traditional user profiling relies on static tags (such as occupation and historical behavior), which makes it difficult to capture the dynamic changes in user needs. In contrast, this invention achieves dynamic user profile calibration by aggregating the number of times users click on the visualized map icons, thus enabling real-time mapping of needs.

[0039] In practical applications, when a user clicks on the logo image A, the user profile tags are adjusted based on information such as the product type or manufacturer contained in the logo image A. The search strategy is adjusted in real time according to the user profile tags. At this time, for the user, the popularity of the product type or manufacturer of the logo image A increases by one, so the brightness of the logo image A becomes brighter.

[0040] Example 2 Based on the AI-based visual retrieval method in Embodiment 1, this application also provides an AI-based visual retrieval system, which is used to implement the method of Embodiment 1, specifically: like Figure 5 As shown, an AI-based visual retrieval system includes: Module 1 is used to acquire users' contextualized needs data and user profile tags; The multimodal intelligent retrieval module 2 is used to vectorize the scenario-based demand data and perform multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data; at the same time, based on user profile tags, it processes the multimodal retrieval data to generate personalized retrieval results that are dynamically associated with user profile tags. Presentation module 3 is used to map personalized search results to a visual graph for display.

[0041] This application, through the cooperation of the acquisition module 1, the multimodal intelligent retrieval module 2, and the presentation module 3, can not only process multimodal data such as text, images, and voice simultaneously, but also dynamically adjust the retrieval strategy through real-time updates of user profile tags, thereby improving the matching degree between retrieval results and users' real needs. This solves industry pain points such as the difficulty in accurately locating products / solutions and the inability to recommend content in a personalized manner.

[0042] The acquisition module includes: The identification module is used to process and analyze user behavior data based on a large language model when user behavior data is acquired, in order to identify user role types and user behavior characteristics. The generation module is used to dynamically generate user profile tags based on user role type and user behavior characteristics.

[0043] This application automatically identifies user role types and behavioral characteristics through the cooperation between the identification module and the generation module, and dynamically generates differentiated tags to provide a foundation for subsequent personalized retrieval.

[0044] The multimodal intelligent retrieval module includes: The vectorization module is used to vectorize the tag requirement data to obtain the requirement vector to be matched. The calling module is used to call a pre-built multimodal database, which stores multiple preset label requirements and a preset label requirement vector corresponding to each preset label requirement; The matching module is used to match the demand vector to be matched with each preset tag demand vector, and select multiple demand vectors to be matched that meet the preset conditions from the multimodal database as multimodal retrieval data to obtain multimodal retrieval data; The first reasoning module is used to process and analyze multimodal retrieval data based on user profile tags and a large language model to obtain multimodal retrieval data that matches the user profile tags and response templates that correspond to the user profile tags. The second reasoning module is used to process and analyze multimodal retrieval data that matches user profile tags and response templates that correspond to user profile tags, and generate personalized retrieval results that are dynamically associated with user profile tags.

[0045] This application, through the collaboration of a vectorization module, a calling module, and a matching module, captures implicit semantic relationships in natural language. Simultaneously, it matches multi-dimensional tag demand vectors (text, parameters, application scenarios, etc.) from a multi-modal database with the demand vectors to be matched, thereby obtaining multi-modal retrieval data and significantly improving retrieval efficiency and accuracy. Furthermore, by dynamically associating the multi-modal database with user profile tags through a first and second inference module, personalized customization of retrieval results is achieved. Compared to traditional retrieval systems that rely solely on keyword matching, this application can simultaneously process multi-modal data such as text, images, and voice. Moreover, it can dynamically adjust retrieval strategies through real-time updates of user profile tags, improving the matching degree between retrieval results and users' actual needs. This addresses industry pain points such as the difficulty in accurately locating products / solutions and the inability to recommend content tailored to individual users.

[0046] AI-based visual retrieval systems also include: The update and optimization module is used to summarize the number of times users select labeled images in the visualization map within a certain period of time; based on the number of times users select labeled images in the visualization map within a certain period of time, the user profile tags are adjusted, and personalized search results are updated based on the adjusted user profile tags. This invention achieves dynamic user profile calibration and real-time mapping of needs by summarizing the number of times users click on labeled images in the visualization map.

[0047] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0048] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

[0049] The above-disclosed examples are only a few specific implementation scenarios of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. An AI-based visual retrieval method, characterized in that, include: Acquire user-specific needs data and user profile tags; The scenario-based requirement data is vectorized and then multimodal retrieved using a pre-built multimodal database to obtain multimodal retrieval data. Simultaneously, based on the user profile tags, the multimodal retrieval data is processed to generate personalized retrieval results dynamically associated with the user profile tags. The personalized search results are mapped to a visualization map for display.

2. The AI-based visual retrieval method according to claim 1, characterized in that, The user profile tags are obtained through the following steps: Once user behavior data is acquired, it is processed and analyzed based on a large language model to identify user role types and user behavior characteristics. Based on the user role type and the user behavior characteristics, the user profile tags are dynamically generated.

3. The AI-based visual retrieval method according to claim 1, characterized in that, The contextualized requirement data includes label requirement data described in natural language; The step of vectorizing the scenario-based requirement data and performing multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data includes: The tag requirement data is vectorized to obtain the requirement vector to be matched; A pre-built multimodal database is invoked, which stores multiple preset label requirements and a preset label requirement vector corresponding to each preset label requirement; The required vector to be matched is matched with each of the preset tag required vectors, and multiple required vectors to be matched that meet the preset conditions are selected from the multimodal database as the multimodal retrieval data to obtain multimodal retrieval data.

4. The AI-based visual retrieval method according to claim 1, characterized in that, The multimodal database includes product data, semantic data, or case data; or the multimodal database includes text data, graphical data, and tabular data.

5. The AI-based visual retrieval method according to claim 1, characterized in that, The process of processing the multimodal retrieval data based on the user profile tags to generate personalized retrieval results dynamically associated with the user profile tags includes: Based on the user profile tags, the multimodal retrieval data is processed and analyzed using a large language model to obtain multimodal retrieval data that matches the user profile tags and response templates that correspond to the user profile tags. Based on the multimodal retrieval data matching the user profile tags and the response templates corresponding to the user profile tags, personalized retrieval results dynamically associated with the user profile tags are generated.

6. The AI-based visual retrieval method according to claim 5, characterized in that, Different user profile tags correspond to different response templates; The process of processing and analyzing multimodal retrieval data matching the user profile tags and response templates corresponding to the user profile tags to generate personalized retrieval results dynamically associated with the user profile tags includes: Based on the multimodal retrieval data matching the user profile tags and the response templates corresponding to the user profile tags, the system processes and analyzes the data to generate product recommendation data or demand solutions that are dynamically associated with the user profile tags, and uses the product recommendation data or demand solutions as the personalized search results.

7. The AI-based visual retrieval method according to claim 1, characterized in that, The visualization map includes multiple different logo images, each of which corresponds to information about the product type or manufacturer. The step of mapping the personalized search results to a visualization map for display includes: Obtain the product type or manufacturer corresponding to the personalized search results, and select the identifier image of the visualization map based on the product type or manufacturer corresponding to the personalized search results; Based on the selected identifier image, the personalized search results are mapped and the selected identifier image is highlighted.

8. The AI-based visual retrieval method according to claim 1, characterized in that, Also includes: The number of times a user selects an image from the visualized map within a certain period of time is summarized. Based on the number of times a user selects an image in the visualization map within a certain period of time, the user profile tags are adjusted, and personalized search results are updated based on the adjusted user profile tags.

9. An AI-based visual retrieval system, characterized in that, include: The acquisition module is used to acquire users' contextualized needs data and user profile tags; The multimodal intelligent retrieval module is used to vectorize the scenario-based demand data and perform multimodal retrieval through a pre-built multimodal database to obtain multimodal retrieval data; at the same time, based on the user profile tags, the multimodal retrieval data is processed to generate personalized retrieval results that are dynamically associated with the user profile tags; The presentation module is used to map the personalized search results to a visualization map for display.

10. The AI-based visual retrieval system according to claim 9, characterized in that, The acquisition module includes: The identification module is used to process and analyze user behavior data based on a large language model when user behavior data is acquired, and to identify user role types and user behavior characteristics. The generation module is used to dynamically generate the user profile tags based on the user role type and the user behavior characteristics.

Images