A vehicle recommendation method and device based on a visual language model

Abstract

This application provides a vehicle recommendation method and device based on a visual language model. The method includes: in response to a specific user operation, acquiring an initial full-body image of the user through an image acquisition module, and performing image preprocessing on the initial full-body image to obtain a target full-body image; inputting the target full-body image into a pre-trained visual language model to determine the user's current clothing tag; obtaining preset tags corresponding to each candidate vehicle model from a pre-built vehicle feature knowledge base, calculating a matching score for each candidate vehicle model based on the preset tags corresponding to each candidate vehicle model and the user's current clothing tag, and determining a target recommended vehicle model from the candidate vehicle models based on the matching score. Through this method and device, deep style preferences that are difficult for users to express precisely in language can be captured, significantly improving the accuracy and personalization of vehicle recommendations.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a vehicle recommendation method and device based on a visual language model. Background Technology

[0002] Traditional vehicle recommendation systems primarily rely on vehicle technical parameters (such as power, size, and energy consumption), price range, and user functional needs (such as seating capacity and driving range) for matching and recommendations. This approach is one-dimensional, focusing only on parameter comparisons and neglecting the user's aesthetic style. Furthermore, its interaction method is relatively passive; users often struggle to accurately express their style preferences, leading to low communication efficiency and inaccurate recommendations. Therefore, current technology has not yet been able to intelligently match users' visual personal characteristics, especially their clothing style, with vehicle design aesthetics, making it difficult to effectively improve recommendation accuracy and user experience. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a vehicle recommendation method and device based on a visual language model, which intelligently associates and matches the user's visual aesthetic features with the vehicle's design aesthetics and brand tone. This solves the problem that existing technologies rely solely on technical parameters and explicit functional preferences for vehicle recommendation, and cannot capture the user's deep aesthetic style and visual personal characteristics. It can capture the deep style preferences that users find difficult to express precisely in language, and significantly improves the accuracy and personalization of vehicle recommendation.

[0004] Secondly, embodiments of this application provide a vehicle recommendation method based on a visual language model, the vehicle recommendation method comprising: In response to a specific user operation, the image acquisition module acquires an initial full-body image of the user and performs image preprocessing on the initial full-body image to obtain a target full-body image; The full-body image of the target is input into a pre-trained visual language model to determine the user's current outfit tag; wherein, the current outfit tag includes an outfit style tag and an outfit scene tag; The system retrieves preset tags corresponding to each candidate model from a pre-built vehicle feature knowledge base, calculates a matching score for each candidate model based on the preset tags corresponding to each candidate model and the user's current outfit tags, and determines the target recommended model from the candidate models based on the matching scores; wherein, the preset tags include model style tags and scene identity tags.

[0005] Furthermore, the step of inputting the full-body image of the target into a pre-trained visual language model to determine the user's current clothing tag includes: The visual language model extracts features and performs semantic understanding on the visual elements of the target full-body image to determine at least one candidate outfit label corresponding to the target full-body image and a confidence score for each candidate outfit label; wherein, the candidate outfit label includes candidate style label and candidate scene label; Candidate outfit tags whose confidence scores exceed a preset confidence threshold are used as the current outfit tags.

[0006] Furthermore, the calculation of the matching score for each candidate car model based on the preset tags corresponding to each candidate car model and the user's current outfit tags includes: For each candidate vehicle model, semantic matching calculation is performed between the vehicle style tag corresponding to the candidate vehicle model and the clothing style tag corresponding to the user to obtain a style matching score; The scene identity tag corresponding to the candidate vehicle model is semantically matched with the scene identity tag corresponding to the user to obtain the scene matching score; Based on the style matching score and the scene matching score, the matching score corresponding to the candidate model is determined.

[0007] Furthermore, determining the target recommended model from the candidate models based on the matching score includes: Based on the matching scores, multiple candidate models are sorted from high to low, and a preset number of candidate models are determined from the sorting results as the target recommended models.

[0008] Furthermore, after determining the target recommended vehicle model, the vehicle recommendation method further includes: The system identifies dialogue fragments from the dialogue knowledge base that match the target recommended car model and the current outfit tag, and generates recommended dialogues that match the user's outfit style based on these dialogue fragments.

[0009] Furthermore, the vehicle feature knowledge base is constructed through the following steps: Acquire multi-source data for each candidate vehicle model, and perform data cleaning on the multi-source data; Based on the large language model, the model style description and scene identity description of each candidate model are extracted from the cleaned multi-source data to obtain the initial description label corresponding to each candidate model. The consistency of the initial description label corresponding to each candidate vehicle model is verified to obtain the preset label corresponding to each candidate vehicle model and store it in the vehicle feature knowledge base.

[0010] Furthermore, the image preprocessing includes de-privacy processing, pose normalization processing, and illumination correction processing.

[0011] Secondly, embodiments of this application also provide a vehicle recommendation device based on a visual language model, the vehicle recommendation device comprising: An image acquisition module is used to acquire an initial full-body image of the user in response to a specific operation by the user, and to perform image preprocessing on the initial full-body image to obtain a target full-body image; The tag determination module is used to input the full-body image of the target into a pre-trained visual language model to determine the current outfit tag corresponding to the user; wherein, the current outfit tag includes an outfit style tag and an outfit scene tag; The recommended vehicle model determination module is used to obtain preset tags corresponding to each candidate vehicle model from a pre-built vehicle feature knowledge base, calculate the matching score corresponding to each candidate vehicle model based on the preset tags corresponding to each candidate vehicle model and the user's current outfit tags, and determine the target recommended vehicle model from the candidate vehicle models according to the matching scores; wherein, the preset tags include vehicle style tags and scene identity tags.

[0012] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, the steps of the vehicle recommendation method based on the visual language model described above are performed.

[0013] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the vehicle recommendation method based on the visual language model described above.

[0014] This application provides a vehicle recommendation method and device based on a visual language model. First, in response to a specific user operation, an initial full-body image of the user is acquired via an image acquisition module, and the initial full-body image is preprocessed to obtain a target full-body image. Then, the target full-body image is input into a pre-trained visual language model to determine the user's current clothing tag; wherein the current clothing tag includes a clothing style tag and a clothing scene tag. Finally, preset tags corresponding to each candidate vehicle model are obtained from a pre-built vehicle feature knowledge base. Based on the preset tags corresponding to each candidate vehicle model and the user's current clothing tag, a matching score is calculated for each candidate vehicle model, and a target recommended vehicle model is determined from the candidate vehicle models based on the matching score; wherein the preset tags include a vehicle style tag and a scene identity tag.

[0015] This application automatically identifies a user's clothing style tags and clothing scene tags by collecting a full-body image and inputting it into a visual language model. These tags are then semantically matched with the style tags and scene identity tags of candidate car models in a vehicle feature knowledge base to ultimately determine the target recommended car model. This application overcomes the limitations of traditional car recommendations that rely solely on technical parameters and explicit user preferences. It intelligently associates and matches users' visual aesthetic characteristics with the vehicle's design aesthetics and brand tone, solving the problem that existing technologies rely solely on technical parameters and explicit preferences, failing to capture users' deep aesthetic styles and visual personal characteristics. It can capture deep style preferences that are difficult for users to express precisely in words, significantly improving the accuracy and personalization of car recommendations.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a vehicle recommendation method based on a visual language model provided in this application embodiment; Figure 2 A schematic diagram of the structure of a vehicle recommendation device based on a visual language model provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. Based on the embodiments of this application, every other embodiment obtained by those skilled in the art without inventive effort falls within the scope of protection of this application.

[0020] First, the applicable application scenarios for this application will be introduced. This application can be applied to the field of artificial intelligence technology.

[0021] Traditional vehicle recommendation systems primarily rely on vehicle technical parameters (such as power, size, and energy consumption), price range, and user functional needs (such as seating capacity and driving range) for matching and recommendations. This approach is one-dimensional, focusing only on parameter comparisons and neglecting the user's aesthetic style. Furthermore, its interaction method is relatively passive; users often struggle to accurately express their style preferences, leading to low communication efficiency and inaccurate recommendations. Therefore, current technology has not yet been able to intelligently match users' visual personal characteristics, especially their clothing style, with vehicle design aesthetics, making it difficult to effectively improve recommendation accuracy and user experience.

[0022] Based on this, this application provides a vehicle recommendation method based on a visual language model, which intelligently associates and matches the user's visual aesthetic features with the vehicle's design aesthetics and brand tone. This solves the problem that in the prior art, vehicle recommendation only relies on technical parameters and explicit functional preferences and cannot capture the user's deep aesthetic style and visual personal characteristics. It can capture the deep style preferences that users find difficult to express precisely in language, and significantly improves the accuracy and personalization of vehicle recommendations.

[0023] Please see Figure 1 , Figure 1 This is a flowchart illustrating a vehicle recommendation method based on a visual language model, provided as an embodiment of this application. Figure 1 As shown in the embodiments of this application, the vehicle recommendation method includes: S101, in response to a specific operation by the user, the initial full-body image of the user is acquired through the image acquisition module, and the initial full-body image is preprocessed to obtain the target full-body image.

[0024] Regarding step S101 above, in specific implementation, when responding to a specific user operation, an initial full-body image of the user is acquired through an image acquisition module, and the initial full-body image is preprocessed to obtain a target full-body image. Here, as an example, a camera deployed at the entrance of the vehicle showroom or next to the vehicle detects the user's arrival and issues an invitation prompting for vehicle recommendations based on clothing via voice prompts or screen display. After receiving the user's consent, the robot's camera acquires a full-body image of the user in a natural standing position, including their overall clothing. Here, the user can issue the consent command via voice or generate an authorization command by clicking a specific button on the screen; this application does not specifically limit this method.

[0025] Here, according to the embodiments provided in this application, image preprocessing includes de-anonymization, pose normalization, and illumination correction. After acquiring the initial full-body image, the system preprocesses the image to improve the accuracy and robustness of subsequent visual language model analysis. Specifically, the preprocessing includes: de-anonymization blurring or cropping areas in the image that can identify an individual (such as the face), protecting user privacy while extracting clothing features; pose normalization adjusting the user's pose in the image to a standard upright posture, eliminating interference from differences in user posture for clothing feature extraction, facilitating consistent feature extraction by the subsequent model; and illumination correction eliminating color cast, overexposure, or shadow problems caused by differences in ambient lighting, unifying image color and brightness, and ensuring the accuracy of the visual language model's analysis of visual elements such as clothing color and fabric texture. After the above preprocessing, the resulting full-body image, while protecting user privacy, possesses the characteristics of pose standardization and illumination consistency, laying a solid data foundation for subsequent multi-level feature extraction and semantic understanding by the visual language model.

[0026] S102, the full-body image of the target is input into a pre-trained visual language model to determine the current outfit tag corresponding to the user.

[0027] Regarding step S102 above, in specific implementation, the full-body image of the target obtained in step S101 is input into a pre-trained visual language model. This model performs in-depth analysis of the clothing information in the image to ultimately determine the user's current clothing tag. Specifically, the current clothing tag includes a clothing style tag and a clothing scene tag. The clothing style tag describes the user's clothing style, such as urban, business, sports casual, Chinese style, etc., which are not specifically limited in this application. The clothing scene tag describes the typical life scene corresponding to the user's clothing, such as business meetings, travel, family outings, commuting, etc., which are also not specifically limited in this application.

[0028] As an optional embodiment, regarding step S102 above, inputting the target full-body image into a pre-trained visual language model to determine the user's current clothing tag includes: Step 1021: The visual language model performs feature extraction and semantic understanding on the visual elements of the target full-body image to determine at least one candidate clothing label corresponding to the target full-body image and the confidence score corresponding to each candidate clothing label.

[0029] Regarding step 1021 above, in specific implementation, the user's target full-body image is input into the visual language model, which performs feature extraction and semantic understanding on the image. Specifically, the visual language model identifies and analyzes various visual elements in the image, including but not limited to clothing category, color matching, cut, accessory selection, and overall styling coordination. Based on the comprehensive analysis of the above visual elements, the visual language model outputs at least one candidate outfit label and a confidence score for each candidate outfit label. The candidate outfit labels include candidate style labels, such as "urban" with a confidence score of 0.94 and "Chinese style" with a confidence score of 0.21, and candidate scene labels, such as "business meeting" with a confidence score of 0.9 and "commuting" with a confidence score of 0.11. The confidence score ranges from 0 to 1; a higher value indicates a higher degree of confidence in the model's judgment of the label.

[0030] Step 1022: Select candidate outfit tags whose confidence scores exceed a preset confidence threshold as the current outfit tags.

[0031] Regarding step 1022 above, in specific implementation, the candidate outfit tags output in step 1021 are subjected to confidence level filtering. A confidence threshold is preset, which is used to distinguish between high-confidence tags and low-confidence tags. The confidence score corresponding to each candidate outfit tag is compared with the preset threshold. Only candidate outfit tags with confidence scores exceeding the preset threshold are retained as the final current outfit tags, while tags below the threshold, such as "Chinese style" with a confidence score of 0.21 in the example above, are filtered out and discarded. In this way, according to step 1022 above, the confidence level filtering mechanism ensures that the outfit tags finally used for car model matching have sufficient reliability and representativeness, effectively avoiding the interference of low-confidence misjudgments on the accuracy of subsequent recommendations.

[0032] S103: Obtain the preset tags corresponding to each candidate model from the pre-built vehicle feature knowledge base; calculate the matching score corresponding to each candidate model based on the preset tags corresponding to each candidate model and the user's current clothing tags; and determine the target recommended model from the candidate models based on the matching scores.

[0033] Here, the vehicle feature knowledge base is a pre-built tag database that stores preset tags corresponding to each candidate vehicle model. These preset tags include vehicle style tags and scenario identity tags. Vehicle style tags are used to describe the vehicle's design language and aesthetic tone, such as "urban minimalism" or "digital trendiness." Scenario identity tags are used to describe the typical life scenarios and user identities that the vehicle is adapted to, such as "business meeting travel" or "family long-distance road trip."

[0034] Regarding step S103 above, in specific implementation, firstly, the preset tags corresponding to each candidate model are obtained from the pre-built vehicle feature knowledge base. Based on the preset tags corresponding to each candidate model and the user's current clothing tags, the matching score corresponding to each candidate model is calculated. Then, based on the matching scores, the target recommended model suitable for the user is determined from multiple candidate models.

[0035] Specifically, the vehicle feature knowledge base is constructed through the following steps: I: Obtain multi-source data corresponding to each candidate vehicle model, and perform data cleaning on the multi-source data.

[0036] Regarding step I above, in specific implementation, multi-source data corresponding to each candidate vehicle model is acquired. As an example, the sources of multi-source data include, but are not limited to, official vehicle design documents and automotive review data; this application does not specifically limit these sources. Official vehicle design documents provide an authoritative explanation of the designer's aesthetic philosophy and style positioning; automotive review data provides an objective evaluation of the vehicle's design, functions, and applicable scenarios from a third-party perspective. After acquiring the multi-source data, data cleaning is performed. The purpose of data cleaning is to remove noise and redundant information from the data, such as eliminating duplicate content, correcting format inconsistencies, and filtering information unrelated to the vehicle's design aesthetics and scenario positioning, ensuring that the data relied upon for subsequent label extraction is of high quality and consistency.

[0037] II: Based on the large language model, extract the vehicle style description and scene identity description of each candidate vehicle from the cleaned multi-source data to obtain the initial description label corresponding to each candidate vehicle.

[0038] Regarding step II above, in its specific implementation, based on a large language model, the model style description and scenario identity description of each candidate vehicle are extracted from the cleaned multi-source data to generate initial description tags for each candidate vehicle. Specifically, the large language model performs semantic understanding on the cleaned text data, identifies and extracts descriptive fragments involving vehicle design aesthetics, exterior style, and interior tone, and summarizes them into structured model style descriptions. Simultaneously, the large language model also extracts descriptive content related to vehicle usage scenarios and target user profiles from the data, summarizing it into scenario identity descriptions. Through this semantic extraction process, each candidate vehicle obtains a set of initial description tags, providing basic materials for subsequent tag generation.

[0039] III: Perform consistency verification on the initial description label corresponding to each candidate vehicle model to obtain the preset label corresponding to each candidate vehicle model and store it in the vehicle feature knowledge base.

[0040] Regarding step III above, in practice, the initial description labels of each candidate vehicle model generated in step II undergo consistency verification. Once the verification passes, these labels are stored as official preset labels in the vehicle feature knowledge base. The purpose of consistency verification is to ensure that the label scale of similar vehicle models throughout the knowledge base is uniform, avoiding logical contradictions. For example, if two vehicle models with similar positioning and design styles are labeled with significantly different style labels by the large language model, the system will calibrate and adjust the labels by comparing the label distribution of similar vehicle models, ensuring that the labels maintain reasonable distinguishability and consistency in the horizontal dimension. Labels that have passed consistency verification are stored as preset labels in the knowledge base for real-time use in subsequent matching calculations.

[0041] As an optional embodiment, regarding step S103 above, calculating the matching score for each candidate car model based on the preset tag corresponding to each candidate car model and the user's current outfit tag includes: Step 1031: For each candidate vehicle model, perform semantic matching calculation between the vehicle style tag corresponding to the candidate vehicle model and the clothing style tag corresponding to the user to obtain a style matching score.

[0042] Regarding step 1031 above, in specific implementation, for each candidate vehicle model, the vehicle style tag of the candidate vehicle is semantically matched with the user's clothing style tag determined in step S102 to obtain a style matching score. The core of semantic matching calculation lies in measuring the degree of similarity between two tags in the semantic space. For example, if the user's clothing style tag is "business" and the vehicle style tag of a candidate vehicle is "business meeting travel," the two are highly related semantically, and the style matching score is high; conversely, if the vehicle style tag is "family long-distance self-driving," which has a low semantic correlation with "business," the style matching score is correspondingly low. The style matching score reflects the degree of fit between the vehicle's design aesthetics and the user's aesthetic taste.

[0043] Step 1032: Perform semantic matching calculation between the scene identity tag corresponding to the candidate vehicle model and the scene identity tag corresponding to the user to obtain the scene matching score.

[0044] Regarding step 1032 above, in specific implementation, the scenario identity tag of the candidate vehicle is semantically matched with the user's clothing scenario tag determined in step S102 to obtain a scenario matching score. The core of this is to verify the consistency between the lifestyle scenario adapted to the vehicle and the typical scenario implied by the user's clothing. For example, if the user's clothing scenario tag is "business meeting," and the candidate vehicle's scenario identity tag is also "business meeting travel," the two scenarios are highly consistent, resulting in a high scenario matching score. Through consistency verification at the scenario level, the reasonableness and actual scenario fit of the recommendation results are further improved.

[0045] Step 1033: Determine the matching score corresponding to the candidate vehicle model based on the style matching score and the scene matching score.

[0046] Regarding step 1033 above, in specific implementation, the final matching score corresponding to the candidate vehicle model is determined comprehensively based on the style matching score obtained in step 1031 and the scene matching score obtained in step 1032. In an optional embodiment, the matching score can be calculated by weighted summation of the style matching score and the scene matching score, wherein the style matching score, as the primary dimension of matching with the user's clothing, can be assigned a higher weight coefficient. The final matching score comprehensively reflects the overall degree of fit between the vehicle model and the user's clothing image in both aesthetic style and lifestyle scenarios.

[0047] Furthermore, determining the target recommended model from the candidate models based on the matching score includes: Based on the matching scores, multiple candidate models are sorted from high to low, and a preset number of candidate models are determined from the sorting results as the target recommended models.

[0048] Regarding the above steps, in specific implementation, after obtaining the matching score for each candidate model, all candidate models are sorted in descending order based on the matching score. The matching score comprehensively reflects the overall degree of fit between each candidate model and the user's clothing image in terms of both aesthetic style and lifestyle scenarios. The higher the score, the stronger the match between the model and the user's personal style and potential needs. After sorting, a preset number of candidate models are determined from the sorted results as target recommended models. The preset number can be flexibly set according to the actual application scenario; for example, the preset number could be 3, selecting the top three models as the recommended results. This application does not impose a specific limitation on this. Through the above sorting and filtering mechanism, the model options that best match the user's clothing image can be presented efficiently and accurately.

[0049] As an optional embodiment, after determining the target recommended vehicle model, the vehicle recommendation method provided in this application further includes: The system identifies dialogue fragments from the dialogue knowledge base that match the target recommended car model and the current outfit tag, and generates recommended dialogues that match the user's outfit style based on these dialogue fragments.

[0050] In the implementation of the above steps, after determining the target recommended vehicle model, marketing script fragments matching the target recommended vehicle model and the user's current clothing style tags are retrieved from a pre-built script knowledge base. The script knowledge base stores a wealth of marketing script materials, each associated with a corresponding vehicle model tag and style tag index, enabling the system to accurately recall scripts based on the target recommended vehicle model and the user's clothing style tags. Then, based on retrieval enhancement generation technology, the retrieved script fragments are used as context input to a large language model, which generates natural, fluent, and personalized persuasive recommendation scripts. These recommendation scripts can correlate the vehicle model's design style, aesthetic tone, and scene adaptability with the user's clothing style. For example, they can highlight the harmony between the vehicle's lines and the user's clothing cut, and the fit between the vehicle's character and the user's personal style. This allows users to intuitively perceive the deep connection between the recommendation results and their own image, thereby enhancing the persuasiveness and appeal of the recommendation, effectively improving user experience and marketing conversion rates.

[0051] The vehicle recommendation method based on a visual language model provided in this application firstly acquires an initial full-body image of the user in response to a specific user operation, and performs image preprocessing on the initial full-body image to obtain a target full-body image. Then, the target full-body image is input into a pre-trained visual language model to determine the user's current clothing tag; wherein, the current clothing tag includes a clothing style tag and a clothing scene tag. Finally, preset tags corresponding to each candidate vehicle model are obtained from a pre-built vehicle feature knowledge base. Based on the preset tags corresponding to each candidate vehicle model and the user's current clothing tag, a matching score is calculated for each candidate vehicle model, and a target recommended vehicle model is determined from the candidate vehicle models based on the matching score; wherein, the preset tags include a vehicle style tag and a scene identity tag.

[0052] This application automatically identifies a user's clothing style tags and clothing scene tags by collecting a full-body image and inputting it into a visual language model. These tags are then semantically matched with the style tags and scene identity tags of candidate car models in a vehicle feature knowledge base to ultimately determine the target recommended car model. This application overcomes the limitations of traditional car recommendations that rely solely on technical parameters and explicit user preferences. It intelligently associates and matches users' visual aesthetic characteristics with the vehicle's design aesthetics and brand tone, solving the problem that existing technologies rely solely on technical parameters and explicit preferences, failing to capture users' deep aesthetic styles and visual personal characteristics. It can capture deep style preferences that are difficult for users to express precisely in words, significantly improving the accuracy and personalization of car recommendations.

[0053] Please see Figure 2 , Figure 2This is a schematic diagram of a vehicle recommendation device based on a visual language model, provided as an embodiment of this application. Figure 2 As shown, the vehicle recommendation device 200 includes: Image acquisition module 201 is used to acquire an initial full-body image of the user in response to a specific operation of the user, and to perform image preprocessing on the initial full-body image to obtain a target full-body image; The tag determination module 202 is used to input the full-body image of the target into a pre-trained visual language model to determine the current outfit tag corresponding to the user; wherein, the current outfit tag includes an outfit style tag and an outfit scene tag; The recommended vehicle model determination module 203 is used to obtain preset tags corresponding to each candidate vehicle model from a pre-built vehicle feature knowledge base, calculate the matching score corresponding to each candidate vehicle model based on the preset tags corresponding to each candidate vehicle model and the user's current clothing tags, and determine the target recommended vehicle model from the candidate vehicle models according to the matching scores; wherein, the preset tags include vehicle style tags and scene identity tags.

[0054] Furthermore, when the tag determination module 202 inputs the full-body image of the target into a pre-trained visual language model to determine the current clothing tag corresponding to the user, the tag determination module 202 is also used for: The visual language model extracts features and performs semantic understanding on the visual elements of the target full-body image to determine at least one candidate outfit label corresponding to the target full-body image and a confidence score for each candidate outfit label; wherein, the candidate outfit label includes candidate style label and candidate scene label; Candidate outfit tags whose confidence scores exceed a preset confidence threshold are used as the current outfit tags.

[0055] Furthermore, when the recommended vehicle model determination module 203 calculates the matching score for each candidate vehicle model based on the preset tags corresponding to each candidate vehicle model and the user's current outfit tags, the recommended vehicle model determination module 203 is also used for: For each candidate vehicle model, semantic matching calculation is performed between the vehicle style tag corresponding to the candidate vehicle model and the clothing style tag corresponding to the user to obtain a style matching score; The scene identity tag corresponding to the candidate vehicle model is semantically matched with the scene identity tag corresponding to the user to obtain the scene matching score; Based on the style matching score and the scene matching score, the matching score corresponding to the candidate model is determined.

[0056] Furthermore, when the recommended vehicle model determination module 203 is used to determine the target recommended vehicle model from the candidate vehicle models based on the matching score, the recommended vehicle model determination module 203 is also used to: Based on the matching scores, multiple candidate models are sorted from high to low, and a preset number of candidate models are determined from the sorting results as the target recommended models.

[0057] Furthermore, the vehicle recommendation device 200 also includes a script generation module. After determining the target recommended vehicle model, the script generation module is used to: The system identifies dialogue fragments from the dialogue knowledge base that match the target recommended car model and the current outfit tag, and generates recommended dialogues that match the user's outfit style based on these dialogue fragments.

[0058] Furthermore, the vehicle recommendation device 200 also includes a knowledge base construction module, which is used to construct the vehicle feature knowledge base through the following steps: Acquire multi-source data for each candidate vehicle model, and perform data cleaning on the multi-source data; Based on the large language model, the model style description and scene identity description of each candidate model are extracted from the cleaned multi-source data to obtain the initial description label corresponding to each candidate model. The consistency of the initial description label corresponding to each candidate vehicle model is verified to obtain the preset label corresponding to each candidate vehicle model and store it in the vehicle feature knowledge base.

[0059] Furthermore, the image preprocessing includes de-privacy processing, pose normalization processing, and illumination correction processing.

[0060] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 As shown, the electronic device 300 includes a processor 310, a memory 320, and a bus 330.

[0061] The memory 320 stores machine-readable instructions executable by the processor 310. When the electronic device 300 is running, the processor 310 and the memory 320 communicate via the bus 330. When the machine-readable instructions are executed by the processor 310, they can perform the operations described above. Figure 1 The steps of the vehicle recommendation method based on the visual language model in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0062] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The steps of the vehicle recommendation method based on the visual language model in the method embodiment shown are described in detail in the method embodiment, and will not be repeated here.

[0063] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0067] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vehicle recommendation method based on a visual language model, characterized in that, The vehicle recommendation method includes: In response to a specific user operation, the image acquisition module acquires an initial full-body image of the user and performs image preprocessing on the initial full-body image to obtain a target full-body image; The full-body image of the target is input into a pre-trained visual language model to determine the user's current outfit tag; wherein, the current outfit tag includes an outfit style tag and an outfit scene tag; The system retrieves preset tags corresponding to each candidate model from a pre-built vehicle feature knowledge base, calculates a matching score for each candidate model based on the preset tags corresponding to each candidate model and the user's current outfit tags, and determines the target recommended model from the candidate models based on the matching scores; wherein, the preset tags include model style tags and scene identity tags.

2. The vehicle recommendation method according to claim 1, characterized in that, The step of inputting the full-body image of the target into a pre-trained visual language model to determine the user's current clothing tag includes: The visual language model extracts features and performs semantic understanding on the visual elements of the target full-body image to determine at least one candidate outfit label corresponding to the target full-body image and a confidence score for each candidate outfit label; wherein, the candidate outfit label includes candidate style label and candidate scene label; Candidate outfit tags whose confidence scores exceed a preset confidence threshold are used as the current outfit tags.

3. The vehicle recommendation method according to claim 1, characterized in that, The process of calculating the matching score for each candidate car model based on the preset tags corresponding to each candidate car model and the user's current outfit tags includes: For each candidate vehicle model, semantic matching calculation is performed between the vehicle style tag corresponding to the candidate vehicle model and the clothing style tag corresponding to the user to obtain a style matching score; The scene identity tag corresponding to the candidate vehicle model is semantically matched with the scene identity tag corresponding to the user to obtain the scene matching score; Based on the style matching score and the scene matching score, the matching score corresponding to the candidate model is determined.

4. The vehicle recommendation method according to claim 1, characterized in that, The step of determining the target recommended model from the candidate models based on the matching score includes: Based on the matching scores, multiple candidate models are sorted from high to low, and a preset number of candidate models are determined from the sorting results as the target recommended models.

5. The vehicle recommendation method according to claim 1, characterized in that, After determining the target recommended vehicle model, the vehicle recommendation method further includes: The system identifies dialogue fragments from the dialogue knowledge base that match the target recommended car model and the current outfit tag, and generates recommended dialogues that match the user's outfit style based on these dialogue fragments.

6. The vehicle recommendation method according to claim 1, characterized in that, The vehicle feature knowledge base is constructed using the following steps: Acquire multi-source data for each candidate vehicle model, and perform data cleaning on the multi-source data; Based on the large language model, the model style description and scene identity description of each candidate model are extracted from the cleaned multi-source data to obtain the initial description label corresponding to each candidate model. The consistency of the initial description label corresponding to each candidate vehicle model is verified to obtain the preset label corresponding to each candidate vehicle model and store it in the vehicle feature knowledge base.

7. The vehicle recommendation method according to claim 1, characterized in that, The image preprocessing includes de-privacy processing, pose normalization processing, and illumination correction processing.

8. A vehicle recommendation device based on a visual language model, characterized in that, The vehicle recommendation device includes: An image acquisition module is used to acquire an initial full-body image of the user in response to a specific operation by the user, and to perform image preprocessing on the initial full-body image to obtain a target full-body image; The tag determination module is used to input the full-body image of the target into a pre-trained visual language model to determine the current outfit tag corresponding to the user; wherein, the current outfit tag includes an outfit style tag and an outfit scene tag; The recommended vehicle model determination module is used to obtain preset tags corresponding to each candidate vehicle model from a pre-built vehicle feature knowledge base, calculate the matching score corresponding to each candidate vehicle model based on the preset tags corresponding to each candidate vehicle model and the user's current outfit tags, and determine the target recommended vehicle model from the candidate vehicle models according to the matching scores; wherein, the preset tags include vehicle style tags and scene identity tags.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the vehicle recommendation method based on a visual language model as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the vehicle recommendation method based on a visual language model as described in any one of claims 1 to 7.

Images