Vehicle value evaluation method and device based on large model, equipment and storage medium

Abstract

The application relates to the technical field of intelligent decision-making, and discloses a vehicle value evaluation method and device based on a large model, equipment and a storage medium, which comprises the following steps: generating a vehicle feature vector and determining a vehicle parameter set through a vehicle attribute large model; if the target vehicle parameter set is not a complete parameter set, generating a target supplementary parameter through a parameter evaluation large model, and evaluating the target vehicle value of the target vehicle based on the target vehicle parameter set and the target supplementary parameter. In the foregoing manner, the feature vector generated by the vehicle attribute large model is used for multidimensional matching, similar vehicle parameters are found as references, and the number of parameters that need to be maintained in advance is reduced. When the parameters are incomplete, the existing information is used for completion, and the value of a vehicle that is not directly maintained is evaluated. The application can be applied to the field of automobile financial business, reduces the artificial maintenance cost, and improves the efficiency of the value evaluation system in evaluating the vehicle value.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, and in particular to a vehicle value assessment method, apparatus, device, and storage medium based on a large model. Background Technology

[0002] In the auto finance sector, vehicles serve as core collateral or leased assets, and the accuracy of their valuation directly impacts a financial institution's risk control capabilities and asset quality. Taking auto finance leasing as an example, the lessor (leasing company) invests a lump sum to purchase the vehicle and delivers it to the lessee for use. The lessee pays rent on schedule and acquires ownership of the vehicle upon lease expiration. Throughout this process, vehicle residual value assessment is a core basis for rent pricing, risk exposure calculation, and lease scheme design.

[0003] Unlike new cars, the market price of used cars (especially commercial used cars) is influenced by multiple factors, including vehicle wear and tear, added parts and configurations, regional market supply and demand, and fluctuations in new car prices, exhibiting a significant non-linear fluctuation characteristic. Accurate used car valuation is fundamental to financial product design, risk pricing, and asset disposal. Existing solutions require appraisers to continuously track market dynamics and manually update parameter values ​​for different car models, regions, and time points. For example, maintaining daily updates of 4 million parameters requires a dedicated team of over 20 people. Even then, the timeliness of parameter updates is difficult to guarantee, highlighting the contradiction between market fluctuations and parameter lag. Limited by maintenance resources, parameter knowledge bases typically only cover "top" scenarios such as mainstream car models, core cities, and common vehicle ages. For "long-tail" scenarios such as less popular models, remote areas, and extreme vehicle ages, the system often cannot match effective parameters, leading to interruptions in the appraisal process. Statistics show that the industry average failure rate for vehicle valuation is approximately 15%-25%, with over 80% of these failures stemming from missing parameters. When key parameters are missing, existing technologies typically return an "valuation failure" message directly, forcing business personnel to switch to a manual valuation process. Manual valuation relies on individual experience, and the difficulty in standardizing evaluation criteria severely restricts business approval efficiency. Therefore, in the automotive finance sector, improving the efficiency of vehicle valuation systems in assessing vehicle value has become a pressing technical challenge. Summary of the Invention

[0004] This application provides a vehicle valuation method, apparatus, device, and storage medium based on a large model, to improve the efficiency of the valuation system in assessing vehicle value.

[0005] Firstly, this application provides a vehicle valuation method based on a large model, the method comprising: Receive vehicle attribute information of the target vehicle, and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large vehicle attribute model; Based on the target vehicle feature vector, multi-dimensional matching is performed in a preset feature vector database to determine the target vehicle parameter set of the target vehicle. Detect whether the target vehicle parameter set is a complete parameter set; If the target vehicle parameter set is not the complete parameter set, then the target supplementary parameters are generated based on the vehicle attribute information by the parameter evaluation big model, and the target vehicle value of the target vehicle is evaluated based on the target vehicle parameter set and the target supplementary parameters.

[0006] Secondly, this application also provides a vehicle value assessment device based on a large model, the device comprising: The vehicle feature vector generation module is used to receive vehicle attribute information of the target vehicle and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information through a vehicle attribute big model. The vehicle parameter set determination module is used to perform multi-dimensional matching in a preset feature vector database based on the target vehicle feature vector to determine the target vehicle parameter set of the target vehicle. The complete parameter set detection module is used to detect whether the target vehicle parameter set is a complete parameter set; The vehicle value assessment module is used to generate target supplementary parameters based on the vehicle attribute information through a parameter assessment model if the target vehicle parameter set is not the complete parameter set, and to assess the target vehicle value of the target vehicle based on the target vehicle parameter set and the target supplementary parameters.

[0007] Thirdly, this application also provides a computer device, the computer device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the vehicle value assessment method based on a large model as described above.

[0008] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the vehicle valuation method based on a large model as described above.

[0009] This application discloses a vehicle valuation method, apparatus, device, and storage medium based on a large-scale model. The method includes receiving vehicle attribute information of a target vehicle; generating a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large-scale vehicle attribute model; performing multi-dimensional matching based on the target vehicle feature vector in a preset feature vector database to determine the target vehicle parameter set; detecting whether the target vehicle parameter set is a complete parameter set; if the target vehicle parameter set is not a complete parameter set, generating target supplementary parameters based on the vehicle attribute information using a large-scale parameter valuation model, and evaluating the target vehicle value based on the target vehicle parameter set and the target supplementary parameters. Through this method, this application uses the feature vector generated by the large-scale vehicle attribute model for multi-dimensional matching, intelligently searching for parameters of similar vehicles from existing data as a reference, reducing the number of parameters that need to be pre-maintained. When the parameters are still incomplete, the large-scale parameter valuation model completes the parameters based on existing information, valuing vehicles for which parameters have not been directly maintained. In the automotive finance field, this significantly reduces manual maintenance costs while improving the efficiency of the valuation system in assessing vehicle value. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram illustrating the application environment of a vehicle valuation method based on a large model, as provided in an embodiment of this application. Figure 2 This is a flowchart illustrating a first embodiment of a vehicle valuation method based on a large model provided in this application. Figure 3 This is a flowchart illustrating a second embodiment of a vehicle valuation method based on a large model provided in this application. Figure 4 This is a flowchart illustrating a third embodiment of a vehicle valuation method based on a large model provided in this application. Figure 5 A schematic block diagram of a vehicle valuation device based on a large model provided for embodiments of this application; Figure 6 A schematic block diagram of the structure of a computer device provided for an embodiment of this application. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The vehicle valuation method based on a large model provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can receive vehicle attribute information of the target vehicle from the client, and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large vehicle attribute model. Based on the target vehicle feature vector, multi-dimensional matching is performed in a preset feature vector database to determine the target vehicle parameter set. The server then checks whether the target vehicle parameter set is a complete parameter set. If the target vehicle parameter set is not a complete parameter set, a large parameter evaluation model generates target supplementary parameters based on the vehicle attribute information, and evaluates the target vehicle value based on the target vehicle parameter set and the target supplementary parameters. In this invention, the feature vector generated by the large vehicle attribute model is used for multi-dimensional matching, intelligently searching for parameters of similar vehicles from existing data as a reference, reducing the number of parameters that need to be pre-maintained. When the parameters are still incomplete, the large parameter evaluation model completes the parameters based on existing information, estimating the value of vehicles for which parameters have not been directly maintained. In the field of auto finance, this significantly reduces manual maintenance costs while improving the efficiency of the value assessment system in evaluating vehicle value. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0014] Please see Figure 2 , Figure 2 This is a flowchart illustrating a first embodiment of a vehicle valuation method based on a large model, provided in this application. This large model-based vehicle valuation method can be applied to valuation systems to improve the efficiency of vehicle valuation.

[0015] like Figure 2 As shown, the vehicle valuation method based on a large model specifically includes steps S10 to S40.

[0016] Step S10: Receive the vehicle attribute information of the target vehicle, and generate the target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using the vehicle attribute big model. Specifically, the system receives vehicle evaluation requests submitted by business personnel via mobile terminals or business systems, and parses and obtains the vehicle attribute information of the target vehicle. Vehicle attribute information includes structured and unstructured data. Structured data includes at least vehicle model, brand, series, variant, first registration date, current mileage, vehicle location (province / city), number of ownership transfers, maintenance record summaries, and accident history tags. Unstructured data may include exterior photos, interior photos, and text descriptions of repair work orders.

[0017] The aforementioned multimodal vehicle attribute information is input into a pre-constructed large-scale vehicle attribute model. This model is then jointly trained using masked language modeling and contrastive learning on large-scale unlabeled vehicle data. This allows for the mapping of vehicle attribute information from different modalities to the same feature space. Specifically: For structured data, an embedding layer is used to convert it into a fixed-dimensional vector representation; For image data, image features are extracted using a visual encoder; For text data, semantic features are extracted using a text encoder; By fusing the aforementioned multimodal features through a cross-modal attention mechanism, a dense vector of fixed dimensions is output, which is the target vehicle feature vector.

[0018] Step S20: Based on the target vehicle feature vector, perform multi-dimensional matching in a preset feature vector database to determine the target vehicle parameter set; Specifically, the preset feature vector database stores the vehicle feature vectors of historically evaluated vehicles and their corresponding complete parameter sets, including valuation parameters such as depreciation rate, brand popularity rate, mileage wear rate, regional adjustment coefficient, model scarcity coefficient, and seasonal fluctuation coefficient.

[0019] The matching process employs a combination of multi-path recall and fusion ranking. The multi-path recall executes the following strategies in parallel: Vector similarity recall: Using an approximate nearest neighbor search algorithm, the top K vehicle feature vectors with the highest cosine similarity to the target vehicle feature vector are retrieved from a pre-defined feature vector database, and the corresponding vehicle parameter set is extracted. Attribute-constrained recall: Based on key attributes such as vehicle type, region, and age of the target vehicle, filter out the set of vehicle parameters that meet exact matching or interval constraints from the inverted index; Knowledge graph semantic recall: Based on the vehicle knowledge graph, vehicle nodes that are semantically associated with the target vehicle attributes are obtained through graph neural networks, and their parameter sets are extracted.

[0020] The fusion ranking takes the features of the recall results (including similarity score, attribute matching degree, semantic association strength, etc.) as input, outputs a comprehensive relevance score, and selects the top M groups of vehicle parameters with the highest comprehensive relevance scores as the matching results to form the target vehicle parameter set.

[0021] Step S30: Detect whether the target vehicle parameter set is a complete parameter set; Specifically, based on the target vehicle's model and the valuation model used, the corresponding parameter standard matrix is ​​retrieved. This matrix defines all parameter types required for calculating the appraised value (e.g., depreciation rate, brand popularity rate, mileage wear rate, etc.). The target vehicle's parameter set is compared with the parameter standard matrix. If the parameter set contains all parameter types required by the standard matrix, it is considered complete; if any parameter type is missing (e.g., depreciation rate is present but brand popularity rate is missing), it is considered incomplete, and the missing parameter(s) is accurately recorded.

[0022] The rows of the preset parameter standard matrix correspond to each parameter type in the preset vehicle parameter type list, and the columns correspond to multiple dimension attributes, including parameter necessity level, parameter type identifier, confidence threshold lower limit, and constraint relationship between parameters.

[0023] In one embodiment, the target vehicle parameter matrix is ​​compared row by row with the parameter standard matrix, and the following detections are performed respectively: Parameter type integrity check: Based on the parameter necessity level in the parameter standard matrix, determine whether a corresponding row exists in the target vehicle parameter matrix. If a parameter type with a necessity level of "necessary" is missing, it is determined to be an incomplete parameter set, and the missing type is "parameter type missing".

[0024] Parameter value validity check: For each existing parameter, its confidence score is extracted and compared with the lower limit of the confidence threshold corresponding to that parameter in the parameter standard matrix. If the confidence score of any parameter is lower than its lower limit, it is determined that the parameter set needs to be completed, and the missing type is "insufficient parameter confidence".

[0025] Parameter constraint verification: Based on the parameter constraints in the parameter standard matrix (such as the negative correlation between depreciation rate and vehicle age, and the regional consistency between regional adjustment coefficient and brand popularity rate), the consistency of parameter values ​​is verified. If a parameter value is detected to violate the constraint, it is determined to be a set of conflicting parameters, and the missing type is "parameters conflict".

[0026] Step S40: If the target vehicle parameter set is not the complete parameter set, then the target supplementary parameters are generated based on the vehicle attribute information through the parameter evaluation big model, and the target vehicle value of the target vehicle is evaluated based on the target vehicle parameter set and the target supplementary parameters.

[0027] Specifically, when the parameter set is incomplete, the parameter evaluation model is triggered. This model has been fine-tuned using domain knowledge (e.g., "depreciation rates in different cities within the same province do not fluctuate by more than 3%)." The model generates a prompt for each missing parameter, for example: "Given the following data: 1-year-old vehicle, depreciation rate 98% in location A; 3-year-old vehicle, depreciation rate 90% in location B. Please calculate the depreciation rate of the same model in location A for a 3-year-old vehicle." Based on this prompt and the matched parameter set, the model performs mathematical reasoning to generate the missing supplementary parameter value (e.g., estimating a depreciation rate of 94%). After completing all missing parameters, the complete parameter set and the vehicle's new car price are substituted into the valuation formula to finally calculate the assessed value of the target vehicle.

[0028] This embodiment discloses a vehicle valuation method, apparatus, device, and storage medium based on a large-scale model. It receives vehicle attribute information of a target vehicle, generates a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large-scale vehicle attribute model, performs multi-dimensional matching on the target vehicle feature vector in a preset feature vector database to determine the target vehicle parameter set, checks whether the target vehicle parameter set is a complete parameter set, and if the target vehicle parameter set is not a complete parameter set, generates target supplementary parameters based on the vehicle attribute information using a large-scale parameter valuation model, and evaluates the target vehicle value based on the target vehicle parameter set and the target supplementary parameters. Through this method, this application uses the feature vector generated by the large-scale vehicle attribute model for multi-dimensional matching, intelligently searching for parameters of similar vehicles from existing data as a reference, reducing the number of parameters that need to be pre-maintained. When parameters are still incomplete, the large-scale parameter valuation model completes them based on existing information, valuing vehicles for which parameters have not been directly maintained. In the automotive finance field, this significantly reduces manual maintenance costs while improving the efficiency of the valuation system in assessing vehicle value.

[0029] Please see Figure 3 , Figure 3 This is a flowchart illustrating a second embodiment of a vehicle valuation method based on a large model provided in this application.

[0030] like Figure 3 As shown, step S10 includes steps S101 to S103.

[0031] Step S101: Obtain the original vehicle information of the target vehicle, and convert the original vehicle information into vehicle attribute information according to the information type of the original vehicle information, wherein the vehicle attribute information includes structured information, text description information and / or image information; Specifically, the system receives vehicle assessment requests submitted by business personnel via mobile terminals or business systems. These requests include original vehicle information, including but not limited to: vehicle registration certificate, photos of the driver's license, on-site images of the vehicle's exterior and interior, vehicle condition description text entered by business personnel, and data returned from third-party maintenance record query interfaces.

[0032] After obtaining the raw vehicle information, it is categorized according to its information type and converted into standardized vehicle attribute information: For structured information, this mainly includes fields such as vehicle brand, series, specific model, model year, first registration date, current mileage, province and city of vehicle location, number of ownership transfers, and emission standards. This information usually exists in key-value pair format. The system directly extracts it as structured attribute fields and performs standardization processing. For example, "Brand and Model A" is standardized to a standard vehicle model code, "Place Name A" is standardized to a standard administrative division code, and mileage is standardized to a value in kilometers.

[0033] For text description information: This mainly includes supplementary descriptions of the vehicle's condition by business personnel, such as free text descriptions like "the left door has minor scratches, the interior is well maintained, it has been maintained at a 4S store throughout the entire process, and there are no major accident records," as well as repair history text descriptions obtained from the maintenance record interface. This text information is cleaned by removing irrelevant punctuation marks, unifying traditional and simplified Chinese characters, and performing word segmentation to form a text sequence that can be processed by the model.

[0034] For image information: mainly including photos of the vehicle's front, side, rear, interior, dashboard, engine compartment, etc., as well as photos of the vehicle registration certificate and driver's license. The images are uniformly scaled and normalized to adjust them to a fixed resolution acceptable to the model. For example, each image is uniformly scaled to 224 pixels by 224 pixels, and the color channels are standardized.

[0035] Step S102: Map the multimodal vehicle attribute information to the same feature space, and encode the structured information, text description information and / or image information through the vehicle attribute big model to generate a single-modal feature vector; Specifically, after data preprocessing, the three types of vehicle attribute information are input into the vehicle attribute big model. The vehicle attribute big model is a pre-trained multimodal deep learning model, obtained through joint training on large-scale unlabeled vehicle data using masked language modeling and contrastive learning. The vehicle attribute big model has the ability to map different modal information to the same feature space, making subsequent cross-modal fusion possible.

[0036] For encoding structured information: Each field in the structured information is treated as an independent feature dimension. For categorical fields, such as brand, vehicle model, and region, they are converted into dense vector representations through an embedding layer. Specifically, each category value corresponds to a fixed-dimensional embedding vector; for example, "Brand A" corresponds to a 128-dimensional vector, and "Model A" corresponds to another 128-dimensional vector. For numerical fields, such as mileage and vehicle age, they are normalized numerically and then concatenated with the categorical embedding vectors or mapped to the same dimension through a linear transformation. After concatenation or average pooling of the embedding vectors of all structured fields, a fixed-dimensional structured feature vector is generated, which serves as the encoded output for this modality.

[0037] For encoding the text description information: The preprocessed text sequence is input into the model's text encoder, which is based on the Transformer architecture and consists of multiple stacked self-attention layers. Each word in the text sequence is first converted into a word embedding vector, and positional encoding is superimposed to preserve the word's order information in the sequence. After multiple layers of self-attention computation, the model can capture the semantic relationships and contextual information between words. The vector of the first position (i.e., the classification token position) output by the text encoder, or the vectors of all positions, is averaged and pooled as the feature vector of the entire text description information. This feature vector captures semantic information in the text about vehicle condition, maintenance history, and added configurations.

[0038] For image information encoding: The preprocessed vehicle image is input into the model's visual encoder. The visual encoder typically employs a visual Transformer architecture, segmenting the image into multiple fixed-size image patches. Each image patch is linearly projected and converted into an image patch embedding vector, which is then superimposed with positional encoding. After these image patch embedding vectors undergo multi-layer self-attention computation, the model can learn visual features such as vehicle appearance, damage marks, and interior wear levels in the image. The classification token vector output by the visual encoder, or an average pooling of all image patch vectors, is taken as the feature vector of the image information.

[0039] Step S103: The single-modal feature vectors are fused using a cross-modal attention mechanism to generate the target vehicle feature vector.

[0040] Specifically, after generating three unimodal feature vectors, these feature vectors are fused through a cross-modal attention mechanism to generate the final target vehicle feature vector. The core idea of ​​the cross-modal attention mechanism is to enable information interaction between different modalities, so that the final fused feature vector can comprehensively reflect the overall information of the vehicle.

[0041] In a specific embodiment, the structured feature vector, text feature vector, and image feature vector are treated as three different modal inputs. A multi-head cross-attention mechanism is employed, using one modality as the query and another as the key and value, to calculate cross-modal attention weights. For example, using the text feature vector as the query and the image feature vector as the key and value, the attention of text to image is calculated, resulting in text-guided image enhancement features. Similarly, using the image feature vector as the query and the structured feature vector as the key and value, the attention of image to structured information is calculated, resulting in image-guided structured enhancement features. Through the calculation of multiple sets of cross-attention, the model can dynamically learn the correlation between different modalities: when the text description mentions "scratches on the left door," the model enhances the response of the corresponding door area in the image features; when the structured information shows "high mileage," the model enhances the feature representation of wear and tear in the text and image.

[0042] After completing multiple sets of cross-modal attention calculations, all enhanced feature vectors are concatenated and then subjected to dimensionality reduction and mapping through a fully connected layer, ultimately outputting a dense vector of fixed dimensions, namely the target vehicle feature vector. The target vehicle feature vector integrates the precise attributes of structured information, the semantic connotation of textual description information, and the visual features of image information, and can comprehensively represent the overall state of the target vehicle.

[0043] Please see Figure 4 , Figure 4 This is a flowchart illustrating a third embodiment of a vehicle valuation method based on a large model provided in this application.

[0044] like Figure 4 As shown, step S20 includes steps S201 to S202.

[0045] Step S201: Calculate the similarity score between each preset feature vector in the preset feature vector database and the target vehicle feature vector, and sort each preset feature vector in descending order according to the similarity score to generate a candidate matching feature list; Specifically, a pre-built and continuously maintained feature vector database is used to store the feature vectors of historically evaluated vehicles and their corresponding complete parameter sets, providing a retrieval basis for subsequent similarity matching.

[0046] Each record in the database corresponds to a historically evaluated vehicle and contains the following core fields: First, a vehicle feature vector, which is generated by a large vehicle attribute model and has fixed dimensions; second, a set of vehicle parameters, including all valuation parameters used during the evaluation of the vehicle, specifically covering depreciation rate, brand popularity rate, mileage wear rate, regional adjustment coefficient, model scarcity coefficient, seasonal fluctuation coefficient, etc.; third, basic vehicle attribute information, such as model, region, and vehicle age, used to assist in retrieval; and fourth, the evaluation time and evaluation result of the historical vehicle, used for data traceability.

[0047] Each time a vehicle's valuation is successfully completed and manually verified, the system automatically inputs the vehicle's original attribute information into a large vehicle attribute model, generates a corresponding vehicle feature vector, and associates and stores this feature vector with the vehicle's complete parameter set. As the valuation business continues to expand, the number of records in the database continues to grow, and the combinations of vehicle models, regions, and ages covered are becoming increasingly diverse.

[0048] Once the target vehicle's feature vector is obtained, the matching process begins. The target vehicle's feature vector is used as the query vector, and similarity scores are calculated one by one with all feature vectors in the pre-set feature vector database.

[0049] Similarity calculation uses cosine similarity as a metric, which measures the degree of similarity in direction by calculating the cosine of the angle between two vectors. Specifically, it calculates the dot product of the target vehicle's feature vector and each feature vector in the database, and then divides it by the product of their magnitudes. The result ranges from 0 to 1. The closer the value is to 1, the more consistent the directions of the two vectors are, meaning the two vehicles are more similar in overall features.

[0050] The algorithm iterates through all feature vectors in the database, calculating a similarity score for each record. After all calculations are complete, all records are sorted in descending order of similarity score, generating a candidate matching feature list.

[0051] In practice, to improve processing efficiency, instead of calculating the similarity of all vectors in the database one by one, a pre-built approximate nearest neighbor index is used for retrieval. The index quickly locates the few records closest to the target vector, and only these records are used to perform accurate cosine similarity calculations, thereby significantly reducing computational overhead while ensuring accuracy.

[0052] Step S202: Select a preset number of preset feature vectors with similarity scores greater than a preset similarity score threshold from the candidate matching feature list as undetermined matching feature vectors, and extract the parameters corresponding to the undetermined matching feature vectors to determine the target vehicle parameter set.

[0053] Specifically, after generating the candidate matching feature list, the process enters the filtering stage, where the most reliable and relevant matches are selected as the potential matching feature vectors. The filtering employs a dual-condition constraint mechanism to ensure that the selected matches have sufficient similarity and reliability.

[0054] The first constraint is a similarity score threshold constraint. A preset similarity score threshold is set based on business experience. Starting from the head of the candidate matching feature list, only records with a similarity score greater than or equal to the threshold are retained. The purpose of this constraint is to eliminate matches with too low similarity to the target vehicle, avoiding the introduction of noisy data. For example, if a historical vehicle's similarity score to the target vehicle is only 0.6, it indicates that the two differ significantly in the feature space, and its parameter set should not be used as a reference.

[0055] The second constraint is the preset quantity constraint. Under the premise of meeting the similarity threshold, the number of selected matching items is further limited. Usually, the preset number of records with the highest ranking are selected. The purpose of this constraint is to control the size of the reference sample, ensuring that there is a sufficient number of reference data for subsequent parameter fusion, while avoiding the introduction of redundant information or reduction of processing efficiency due to too many samples.

[0056] After filtering out the undetermined matching feature vectors, the set of vehicle parameters corresponding to each undetermined matching feature vector is extracted and determined as the target vehicle parameter set.

[0057] In a specific embodiment, the complete set of parameters stored in each pending matching record is read, including all valuation parameters such as depreciation rate, brand popularity rate, mileage wear rate, regional adjustment coefficient, model scarcity coefficient, and seasonal fluctuation coefficient. Since there are multiple pending matching records, multiple parameter sets will be obtained. These parameter sets together constitute the candidate pool of the target vehicle parameter set.

[0058] The candidate pool is initially integrated by combining multiple parameter sets. Each record is assigned a weight coefficient based on its similarity score; the higher the similarity score, the larger the weight coefficient. For each parameter type, a weighted average of all candidate records for that parameter is calculated as its initial value. For example, for the depreciation rate parameter, the depreciation rate values ​​of each candidate record are multiplied by their corresponding weight coefficients, summed, and then divided by the sum of the weight coefficients to obtain the weighted average depreciation rate.

[0059] In addition to calculating the weighted average, this embodiment also calculates a confidence score for each parameter. The confidence score takes into account the following factors: first, the number of pending matching records; the more records, the higher the confidence; second, the concentration of similarity scores; the closer the similarity scores of each record, the higher the confidence; and third, the consistency of the values ​​of each record on this parameter; the more concentrated the values, the higher the confidence.

[0060] If, during the matching process, there are no pending matching feature vectors after filtering due to insufficient records in the database or low similarity, the matching is deemed to have failed, and the parameter evaluation large model completion process is directly triggered to generate a complete parameter set.

[0061] based on Figure 2 In the illustrated embodiment, step S30 includes: Obtain the standard parameter matrix; A target vehicle parameter matrix is ​​constructed based on the target vehicle parameter set, and the target vehicle parameter set is used to detect whether it is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix.

[0062] Specifically, the parameter standard matrix is ​​retrieved and a standardized matrix is ​​constructed by combining all the valuation parameters required by the valuation model in the appendix. The matrix clearly includes all the parameter types necessary for assessing the value of a vehicle, such as depreciation rate, brand popularity rate, and mileage wear rate. At the same time, the standard position and effective value range of each parameter are marked, as well as the parameter dimension specifications adapted to different models and regions. Moreover, this matrix is ​​consistent with the parameter standards of the used car valuation coefficient library manually maintained by financial institutions, providing a unified judgment basis for parameter integrity detection.

[0063] Following the parameter order and dimension specifications of the parameter standard matrix, all parameters in the target vehicle parameter set are structured and matrix-constructed. The specific values ​​of each type of parameter are precisely filled into the matrix positions with the same dimension and order as the parameter standard matrix. If a certain type of parameter is missing in the target vehicle parameter set, the corresponding matrix position is marked as empty. If a parameter value exceeds the normal range, it is filled with the actual value and marked as an anomaly. Finally, a target vehicle parameter matrix is ​​formed that completely corresponds to the dimension and order of the parameter standard matrix.

[0064] The constructed target vehicle parameter matrix is ​​compared row by row and column by column with the parameter standard matrix in all dimensions and positions. This process checks whether the parameter type, matrix dimension, and parameter position of the target vehicle parameter matrix are completely consistent with the parameter standard matrix, and confirms whether there are any missing parameter types, inconsistent matrix dimensions, or disordered parameter positions. If any of these conditions are found, the target vehicle parameter set is directly determined to be an incomplete parameter set, and the missing parameter types, inconsistent dimensions, or disordered position information are marked simultaneously.

[0065] If, after comparing the dimensions and positional order, the parameter type, dimension, and positional order of the target vehicle parameter matrix are consistent with those of the parameter standard matrix, then the validity of the parameter values ​​at each position in the target vehicle parameter matrix is ​​further verified. The actual value of each parameter is compared with the corresponding valid value range marked in the parameter standard matrix to check for any invalid parameter values ​​that exceed the valid value range. If there are invalid parameters and there are no other reasonable values ​​to replace them, the target vehicle parameter set is also determined to be an incomplete parameter set, and the type and value of the invalid parameters are marked. If the dimension and positional order of the target vehicle parameter matrix completely match those of the parameter standard matrix, and all parameter values ​​are within the corresponding valid value range, then the target vehicle parameter set is determined to be a complete parameter set.

[0066] In a specific embodiment, a target vehicle parameter matrix is ​​constructed based on the target vehicle parameter set, and the target vehicle parameter set is checked to determine whether it is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix, including: Perform dimensional matching and positional comparison between the target vehicle parameter matrix and the parameter standard matrix; If the parameter dimension and position of the target vehicle parameter matrix are completely consistent with the parameter dimension and position of the parameter standard matrix, then the target vehicle parameter set is determined to be the complete parameter set. If the target vehicle parameter matrix has missing dimensions, missing positions, and / or invalid values, then the target vehicle parameter set is determined not to be the complete parameter set.

[0067] Specifically, the normalized target vehicle parameter matrix is ​​compared and checked one by one with the parameter standard matrix in all dimensions and positions. First, dimension matching is performed to check whether the number of parameter dimensions and the parameter types corresponding to each dimension in the target vehicle parameter matrix are completely consistent with the parameter standard matrix, and to confirm whether there are any missing parameter dimensions or inconsistent dimension types. Then, position comparison is performed. According to the fixed position order of the parameter standard matrix, the arrangement position of each parameter in the target vehicle parameter matrix is ​​checked row by row and column by column to confirm whether there are any disordered parameter positions or missing key positions.

[0068] If, upon testing, the number and type of parameters in the target vehicle parameter matrix perfectly match the parameter standard matrix, and the order of all parameters is completely consistent with the fixed order of the parameter standard matrix, without any missing dimensions, disordered order, or missing order issues, and all parameter values ​​within the matrix are free of invalid markers (i.e., all values ​​are within the valid value range specified by the parameter standard matrix), then the target vehicle parameter set is directly determined to be a complete parameter set and can be directly substituted into the valuation model for vehicle value assessment.

[0069] If, upon inspection, the target vehicle parameter matrix exhibits any of the following anomalies: insufficient number of parameter dimensions, missing dimensions whose type does not conform to the standard, disordered parameter arrangement, missing key positions with blank or no values, or invalid value markers at any parameter position, or invalid values ​​exceeding the valid value range, then it is directly determined that the target vehicle parameter set is not a complete parameter set. Simultaneously, all anomalies are categorized and marked, clearly indicating the missing parameter dimensions, misaligned / missing parameter positions, invalid parameter types, and specific values, forming a detailed list of parameter anomalies.

[0070] based on Figure 2 In the illustrated embodiment, step S40 includes: The lightweight inference model generates basic vehicle evaluation parameters based on the vehicle attribute information and calculates the confidence score of the basic vehicle evaluation parameters. Specifically, the lightweight inference model inputs all vehicle attribute information (including structured information, text descriptions, and image information) of the target vehicle. For missing or invalid parameter types in the target vehicle parameter set, it quickly generates corresponding basic vehicle evaluation parameters, covering all parameter types that need to be supplemented, such as depreciation rate, brand popularity rate, and mileage wear rate. Based on the completeness of the vehicle attribute information upon which the parameter inference is based and its matching degree with historical parameter samples, the lightweight inference model calculates a corresponding confidence score for each generated basic vehicle evaluation parameter.

[0071] The vehicle basic evaluation parameters whose confidence scores are greater than or equal to a preset confidence threshold are used as lightweight inference supplementary parameters. Specifically, a confidence threshold is pre-set according to risk control requirements. The confidence threshold is calibrated by the accuracy data of historical parameter inference to ensure that the selected parameter values ​​meet the usage standards of the valuation model. The confidence scores of all vehicle basic assessment parameters are compared one by one with the pre-set confidence threshold. Vehicle basic assessment parameters with confidence scores greater than or equal to the threshold are selected and retained as lightweight inference supplementary parameters.

[0072] Based on preset parameter constraints, the deep inference model generates supplementary deep inference parameters according to the vehicle basic evaluation parameters, vehicle attribute information, and confidence scores where the confidence score is less than the preset confidence score. Specifically, the system retrieves preset parameter constraints, including the effective value range of each parameter, regional parameter fluctuation limits, and rules governing the correlation between vehicle age and parameters. It inputs the vehicle's basic assessment parameters with confidence scores below a preset confidence threshold, the target vehicle's full set of attribute information, and the corresponding confidence scores for these parameters into the deep inference model. The deep inference model, combined with the preset parameter constraints, performs deep mathematical reasoning and correction on the low-confidence vehicle basic assessment parameters. Simultaneously, relying on the correlation characteristics of massive historical valuation parameters and combining detailed dimensions of vehicle attribute information, it performs precise calculations, ultimately generating supplementary deep inference parameters that meet the parameter constraints and have higher inference reliability. These parameters are used to supplement missing / invalid parameter types with low confidence.

[0073] The lightweight inference supplementary parameters and the deep inference supplementary parameters are determined as the target supplementary parameters.

[0074] Specifically, the selected lightweight inference supplementary parameters are integrated with the deep inference supplementary parameters generated by the deep inference model. Following the parameter type and order requirements of the parameter standard matrix, the two types of supplementary parameters are structured to ensure that the integrated supplementary parameters are free of duplication and omissions, and completely match any missing or invalid parameter types in the target vehicle parameter set. After integration, the complete parameter set containing both lightweight and deep inference supplementary parameters is determined as the final target supplementary parameters.

[0075] In a specific embodiment, step S40 further includes: A parameter identifier matrix is ​​generated based on the target vehicle parameter set and the target supplementary parameters, and the baseline value of the target vehicle is obtained. Specifically, the effective parameters and supplementary parameters from the target vehicle parameter set are integrated. Following the fixed order and parameter type of the parameter standard matrix, a unique parameter identifier is assigned to each parameter. This identifier includes the parameter type, parameter source (target vehicle parameter set / target supplementary parameter), and parameter validity level. Based on the mapping relationship between this identifier and the corresponding parameter, a structured parameter identifier matrix is ​​constructed. The benchmark value corresponding to the target vehicle is retrieved from the value evaluation coefficient library. This benchmark value is the official guide price or fair market ex-factory price of a new car of the same model as the target vehicle, serving as the core calculation basis for vehicle value assessment and consistent with the new car price determination rules in the valuation model in the appendix.

[0076] The fusion weights of each vehicle parameter in the target vehicle parameter set and the target supplementary parameters are determined based on the parameter identification matrix. Specifically, based on the vehicle valuation risk control rules, the core basis for parameter weight allocation is set. Parameters whose source is the target vehicle parameter set (matching its own maintenance coefficient library) are assigned a higher basic weight; parameters whose source is supplementary parameters are assigned a corresponding basic weight according to their inferred reliability level. Then, using the parameter identification matrix as the core basis, the source and validity level information of each parameter in the matrix are extracted. Combined with the preset weight allocation rules, the corresponding fusion weight is calculated and assigned to each valid parameter in the target vehicle parameter set and each complete parameter in the target supplementary parameters. The fusion weight allocation of all parameters is consistent with the core influence logic of the valuation parameters in the appendix. Core parameters such as depreciation rate, mileage wear rate, and brand popularity rate are assigned higher fusion weights.

[0077] The value of the target vehicle is evaluated based on the baseline value, the target vehicle parameter set, the target supplementary parameters, and the fusion weight.

[0078] Specifically, the effective parameter values ​​in the target vehicle parameter set and the complete parameter values ​​in the target supplementary parameters are combined with their corresponding fusion weights to obtain the weighted values ​​of each parameter. Following the core computational logic of the vehicle valuation model, the weighted values ​​of all parameters are multiplied together to obtain the comprehensive parameter coefficients of the target vehicle. The baseline value of the target vehicle is then calculated using these comprehensive parameter coefficients to obtain the preliminary assessed value of the target vehicle. This preliminary assessed value undergoes risk control adaptation verification. Once the verification passes, this value is determined as the final assessed value of the target vehicle and output.

[0079] Please see Figure 5 , Figure 5 This application provides a schematic block diagram of a large-model-based vehicle valuation device, which is used to execute the aforementioned large-model-based vehicle valuation method. The large-model-based vehicle valuation device can be configured on a server.

[0080] like Figure 5 As shown, the vehicle valuation device 400 based on a large model includes: The vehicle feature vector generation module 410 is used to receive vehicle attribute information of the target vehicle and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information through a vehicle attribute big model. The vehicle parameter set determination module 420 is used to perform multi-dimensional matching in a preset feature vector database based on the target vehicle feature vector to determine the target vehicle parameter set of the target vehicle. The complete parameter set detection module 430 is used to detect whether the target vehicle parameter set is a complete parameter set. The vehicle value assessment module 440 is used to generate target supplementary parameters based on the vehicle attribute information through a parameter assessment model if the target vehicle parameter set is not the complete parameter set, and to assess the target vehicle value of the target vehicle based on the target vehicle parameter set and the target supplementary parameters.

[0081] Furthermore, the vehicle feature vector generation module 410 includes: The vehicle attribute information conversion unit is used to acquire the original vehicle information of the target vehicle and convert the original vehicle information into vehicle attribute information according to the information type of the original vehicle information, wherein the vehicle attribute information includes structured information, text description information and / or image information; A single-modal feature vector generation unit is used to map the multimodal vehicle attribute information to the same feature space, and to encode the structured information, the text description information and / or the image information through the vehicle attribute big model to generate a single-modal feature vector; The target vehicle feature vector generation unit is used to fuse the single-modal feature vectors through a cross-modal attention mechanism to generate the target vehicle feature vector.

[0082] Furthermore, the vehicle parameter set determination module 420 includes: The candidate matching feature list generation unit is used to calculate the similarity score between each preset feature vector in the preset feature vector database and the target vehicle feature vector, and to sort each preset feature vector in descending order according to the similarity score to generate a candidate matching feature list; The target vehicle parameter set determination unit is used to select a preset number of preset feature vectors with similarity scores greater than a preset similarity score threshold from the candidate matching feature list as undetermined matching feature vectors, and extract the parameters corresponding to the undetermined matching feature vectors to determine the target vehicle parameter set.

[0083] Furthermore, the complete parameter set detection module 430 includes: The parameter standard matrix acquisition unit is used to acquire the parameter standard matrix; The complete parameter set detection unit is used to construct a target vehicle parameter matrix based on the target vehicle parameter set, and to detect whether the target vehicle parameter set is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix.

[0084] Furthermore, the complete parameter set detection unit includes: The matching and comparison subunit is used to perform dimensional matching and positional comparison between the target vehicle parameter matrix and the parameter standard matrix; The first determining subunit is used to determine that the target vehicle parameter set is the complete parameter set if the parameter dimension and position of the target vehicle parameter matrix are completely consistent with the parameter dimension and position of the parameter standard matrix. The second determining subunit is used to determine that the target vehicle parameter set is not the complete parameter set if the target vehicle parameter matrix has missing dimensions, missing positions, and / or invalid values.

[0085] Furthermore, the vehicle valuation module 440 includes: The confidence score calculation unit is used to generate basic vehicle evaluation parameters based on the vehicle attribute information through the lightweight inference model, and to calculate the confidence score of the basic vehicle evaluation parameters. The lightweight inference supplementary parameter determination unit is used to use the vehicle basic evaluation parameters whose confidence scores are greater than or equal to a preset confidence threshold as lightweight inference supplementary parameters. The deep inference supplementary parameter generation unit is used to generate deep inference supplementary parameters based on preset parameter constraints, through the deep inference model, according to the vehicle basic evaluation parameters with confidence scores less than the preset confidence, the vehicle attribute information, and the confidence scores; The target supplementary parameter determination unit is used to determine the lightweight inference supplementary parameter and the deep inference supplementary parameter as the target supplementary parameter.

[0086] Furthermore, the vehicle valuation module 440 includes: The parameter identifier matrix generation unit is used to generate a parameter identifier matrix based on the target vehicle parameter set and the target supplementary parameters, and to obtain the benchmark value of the target vehicle; The fusion weight determination unit is used to determine the fusion weight of each vehicle parameter in the target vehicle parameter set and the target supplementary parameter according to the parameter identification matrix; The target vehicle value assessment unit is used to assess the value of the target vehicle based on the benchmark value, the target vehicle parameter set, the target supplementary parameters, and the fusion weight.

[0087] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and modules can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] The aforementioned apparatus can be implemented as a computer program, which can be used in, for example... Figure 6 It runs on the computer device shown.

[0089] Please see Figure 6 , Figure 6 This is a schematic block diagram illustrating the structure of a computer device according to an embodiment of this application. The computer device may be a server.

[0090] See Figure 6 The computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0091] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any vehicle valuation method based on a large model.

[0092] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0093] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program enables the processor to perform any vehicle valuation method based on a large model.

[0094] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0095] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0096] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: Receive vehicle attribute information of the target vehicle, and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large vehicle attribute model; Based on the target vehicle feature vector, multi-dimensional matching is performed in a preset feature vector database to determine the target vehicle parameter set of the target vehicle. Detect whether the target vehicle parameter set is a complete parameter set; If the target vehicle parameter set is not the complete parameter set, then the target supplementary parameters are generated based on the vehicle attribute information by the parameter evaluation big model, and the target vehicle value of the target vehicle is evaluated based on the target vehicle parameter set and the target supplementary parameters.

[0097] In one embodiment, vehicle attribute information of a target vehicle is received, and a target vehicle feature vector corresponding to the target vehicle is generated based on the vehicle attribute information using a large vehicle attribute model, for the purpose of: Obtain the original vehicle information of the target vehicle, and convert the original vehicle information into vehicle attribute information according to the information type of the original vehicle information, wherein the vehicle attribute information includes structured information, text description information and / or image information; The multimodal vehicle attribute information is mapped to the same feature space, and the structured information, text description information and / or image information are encoded through the vehicle attribute big model to generate a single-modal feature vector; The target vehicle feature vector is generated by fusing the single-modal feature vectors through a cross-modal attention mechanism.

[0098] In one embodiment, multi-dimensional matching is performed on the target vehicle feature vector in a preset feature vector database to determine the target vehicle parameter set, which is used to achieve: Calculate the similarity score between each preset feature vector in the preset feature vector database and the target vehicle feature vector, and sort each preset feature vector in descending order according to the similarity score to generate a candidate matching feature list; From the candidate matching feature list, a preset number of preset feature vectors with similarity scores greater than a preset similarity score threshold are selected as undetermined matching feature vectors, and the parameters corresponding to the undetermined matching feature vectors are extracted to determine the target vehicle parameter set.

[0099] In one embodiment, detecting whether the target vehicle parameter set is a complete parameter set is used to achieve: Obtain the standard parameter matrix; A target vehicle parameter matrix is ​​constructed based on the target vehicle parameter set, and the target vehicle parameter set is used to detect whether it is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix.

[0100] In one embodiment, a target vehicle parameter matrix is ​​constructed based on the target vehicle parameter set, and the target vehicle parameter set is checked to determine whether it is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix, for the purpose of: Perform dimensional matching and positional comparison between the target vehicle parameter matrix and the parameter standard matrix; If the parameter dimension and position of the target vehicle parameter matrix are completely consistent with the parameter dimension and position of the parameter standard matrix, then the target vehicle parameter set is determined to be the complete parameter set. If the target vehicle parameter matrix has missing dimensions, missing positions, and / or invalid values, then the target vehicle parameter set is determined not to be the complete parameter set.

[0101] In one embodiment, a target supplementary parameter is generated by evaluating a large model based on the vehicle attribute information, to achieve the following: The lightweight inference model generates basic vehicle evaluation parameters based on the vehicle attribute information and calculates the confidence score of the basic vehicle evaluation parameters. The vehicle basic evaluation parameters whose confidence scores are greater than or equal to a preset confidence threshold are used as lightweight inference supplementary parameters. Based on preset parameter constraints, the deep inference model generates supplementary deep inference parameters according to the vehicle basic evaluation parameters, vehicle attribute information, and confidence scores where the confidence score is less than the preset confidence score. The lightweight inference supplementary parameters and the deep inference supplementary parameters are determined as the target supplementary parameters.

[0102] In one embodiment, the target vehicle value of the target vehicle is evaluated based on the target vehicle parameter set and the target supplementary parameters, for the purpose of: A parameter identifier matrix is ​​generated based on the target vehicle parameter set and the target supplementary parameters, and the baseline value of the target vehicle is obtained. The fusion weights of each vehicle parameter in the target vehicle parameter set and the target supplementary parameters are determined based on the parameter identification matrix. The value of the target vehicle is evaluated based on the baseline value, the target vehicle parameter set, the target supplementary parameters, and the fusion weight.

[0103] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the vehicle value assessment methods based on a large model provided in the embodiments of this application.

[0104] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0105] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0106] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions 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 valuation method based on a large model, characterized in that, include: Receive vehicle attribute information of the target vehicle, and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large vehicle attribute model; Based on the target vehicle feature vector, multi-dimensional matching is performed in a preset feature vector database to determine the target vehicle parameter set; Detect whether the target vehicle parameter set is a complete parameter set; If the target vehicle parameter set is not the complete parameter set, then the target supplementary parameters are generated based on the vehicle attribute information through the parameter evaluation big model, and the target vehicle value of the target vehicle is evaluated based on the target vehicle parameter set and the target supplementary parameters.

2. The vehicle valuation method based on a large model according to claim 1, characterized in that, The step of receiving the vehicle attribute information of the target vehicle and generating a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information using a large vehicle attribute model includes: Obtain the original vehicle information of the target vehicle, and convert the original vehicle information into vehicle attribute information according to the information type of the original vehicle information, wherein the vehicle attribute information includes structured information, text description information and / or image information; The multimodal vehicle attribute information is mapped to the same feature space, and the structured information, text description information and / or image information are encoded through the vehicle attribute big model to generate a single-modal feature vector; The target vehicle feature vector is generated by fusing the single-modal feature vectors through a cross-modal attention mechanism.

3. The vehicle valuation method based on a large model according to claim 1, characterized in that, The step of determining the target vehicle parameter set by performing multi-dimensional matching based on the target vehicle feature vector in a preset feature vector database includes: Calculate the similarity score between each preset feature vector in the preset feature vector database and the target vehicle feature vector, and sort each preset feature vector in descending order according to the similarity score to generate a candidate matching feature list; From the candidate matching feature list, a preset number of preset feature vectors with similarity scores greater than a preset similarity score threshold are selected as undetermined matching feature vectors, and the parameters corresponding to the undetermined matching feature vectors are extracted to determine the target vehicle parameter set.

4. The vehicle valuation method based on a large model according to claim 1, characterized in that, The detection of whether the target vehicle parameter set is a complete parameter set includes: Obtain the standard parameter matrix; A target vehicle parameter matrix is ​​constructed based on the target vehicle parameter set, and the target vehicle parameter set is used to detect whether it is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix.

5. The vehicle valuation method based on a large model according to claim 4, characterized in that, The step of constructing a target vehicle parameter matrix based on the target vehicle parameter set, and detecting whether the target vehicle parameter set is the complete parameter set based on the parameter standard matrix and the target vehicle parameter matrix, includes: Perform dimensional matching and positional comparison between the target vehicle parameter matrix and the parameter standard matrix; If the parameter dimension and position of the target vehicle parameter matrix are completely consistent with the parameter dimension and position of the parameter standard matrix, then the target vehicle parameter set is determined to be the complete parameter set. If the target vehicle parameter matrix has missing dimensions, missing positions, and / or invalid values, then the target vehicle parameter set is determined not to be the complete parameter set.

6. The vehicle valuation method based on a large model according to claim 1, characterized in that, The parameter evaluation model includes a lightweight inference model and a deep inference model. The generation of target supplementary parameters based on the vehicle attribute information using the parameter evaluation model includes: The lightweight inference model generates basic vehicle evaluation parameters based on the vehicle attribute information and calculates the confidence score of the basic vehicle evaluation parameters. The vehicle basic evaluation parameters whose confidence scores are greater than or equal to a preset confidence threshold are used as lightweight inference supplementary parameters. Based on preset parameter constraints, the deep inference model generates supplementary deep inference parameters according to the vehicle basic evaluation parameters, vehicle attribute information, and confidence scores where the confidence score is less than the preset confidence score. The lightweight inference supplementary parameters and the deep inference supplementary parameters are determined as the target supplementary parameters.

7. The vehicle valuation method based on a large model according to claim 6, characterized in that, The process of evaluating the target vehicle value based on the target vehicle parameter set and the target supplementary parameters includes: A parameter identifier matrix is ​​generated based on the target vehicle parameter set and the target supplementary parameters, and the baseline value of the target vehicle is obtained. The fusion weights of each vehicle parameter in the target vehicle parameter set and the target supplementary parameters are determined based on the parameter identification matrix. The value of the target vehicle is evaluated based on the baseline value, the target vehicle parameter set, the target supplementary parameters, and the fusion weight.

8. A vehicle valuation device based on a large model, characterized in that, include: The vehicle feature vector generation module is used to receive vehicle attribute information of the target vehicle and generate a target vehicle feature vector corresponding to the target vehicle based on the vehicle attribute information through a vehicle attribute big model. The vehicle parameter set determination module is used to perform multi-dimensional matching in a preset feature vector database based on the target vehicle feature vector to determine the target vehicle parameter set of the target vehicle. The complete parameter set detection module is used to detect whether the target vehicle parameter set is a complete parameter set; The vehicle value assessment module is used to generate target supplementary parameters based on the vehicle attribute information through a parameter assessment model if the target vehicle parameter set is not the complete parameter set, and to assess the target vehicle value of the target vehicle based on the target vehicle parameter set and the target supplementary parameters.

9. A computer device, characterized in that, The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the vehicle valuation method based on a large 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, causes the processor to implement the vehicle valuation method based on a large model as described in any one of claims 1 to 7.

Images