Vehicle recommendation information processing method and system, computer equipment and storage medium

Abstract

The invention relates to the technical field of information processing, in particular to a vehicle recommendation information processing method and system, computer equipment and a storage medium. The method comprises the following steps: responding to a customer demand response triggering operation initiated by a vehicle merchant terminal, and collecting full-amount source data required for constructing a customer portrait and a vehicle source portrait; performing structured processing on the full-amount source data, and simultaneously extracting a customer feature vector and a vehicle source feature vector; the customer feature vectors and the vehicle source feature vectors are input into a recommendation calculation layer to calculate the demand integrating degree of each vehicle source feature vector and the customer feature vector, priority ranking is carried out, and a vehicle recommendation list containing vehicle source core information and demand integrating reasons is generated; and pushing the vehicle recommendation list to a corresponding vehicle merchant terminal, and recording subsequent interaction feedback data of the customer for the recommendation list for iterative optimization of the recommendation calculation layer. The matching degree of the vehicle source and the customer demand can be improved.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method, system, computer device, and storage medium for processing vehicle recommendation information. Background Technology

[0002] As the automotive consumer market matures, the used car industry continues to expand, and car dealers' business models have shifted from traditional offline to a fully integrated "online + offline" model. Especially with the rapid rise of Douyin (TikTok) live streaming, live streaming, with its strong real-time interactivity and wide reach, has become a key channel for used car dealers to acquire and convert customers. In live streaming customer acquisition scenarios, the core interaction between car dealers and customers is concentrated in private messaging—customers inquire about key information such as vehicle price, model, condition, and usage scenarios via private messages. Car dealers need to quickly match suitable vehicles based on these fragmented interactions, and the efficiency and accuracy of this process directly determine the success rate of transactions. However, there is a significant gap between the current used car industry's demand for "precise customer-cargo matching" and existing technological capabilities: customer needs are highly personalized, focusing not only on basic attributes such as price and model but also on in-depth information such as vehicle condition and maintenance records; while car dealers possess a vast number of vehicles, they lack an integrated solution for efficiently analyzing customer needs and comprehensively characterizing vehicle attributes. Traditional recommendation methods are insufficient to meet the real-time, precise matching needs of live streaming scenarios.

[0003] Furthermore, a similar patent, such as CN119988747A, discloses a service recommendation method, apparatus, terminal device, and storage medium, applicable to the field of data processing technology. This method includes: acquiring historical cockpit service usage information, current cockpit service usage information, and vehicle driving status information; mapping and enhancing the historical cockpit service usage information, current cockpit service usage information, and vehicle driving status information to generate individual driver behavior information; matching and calculating the individual driver behavior information and driver group type information to obtain matching group type information and matching degree information; and generating service recommendation information based on the individual driver behavior information, matching group type information, matching degree information, a preset service recommendation model, and preset prompt text information. This application can accurately capture the driver's infotainment service needs, effectively improving the accuracy and reliability of personalized service recommendations in intelligent cockpits. This invention improves the accuracy of personalized service recommendations in smart cockpits, but the solution focuses on cockpit service recommendations in driving scenarios and does not address the core needs of used car live-streaming private messaging scenarios. It lacks the integration and processing of used car-specific data sources such as private messaging interaction data and vehicle lifecycle data, does not reveal the hierarchical analysis mechanism of explicit and implicit customer needs, and cannot extract implicit demand parameters and convert them into computable feature vectors, resulting in a single matching dimension and insufficient accuracy. Summary of the Invention

[0004] To address the aforementioned technical problems in existing vehicle recommendation processes, this invention provides a method for processing vehicle recommendation information that comprehensively captures customer needs and vehicle attributes by integrating exclusive data sources such as private messaging interactions, the entire vehicle lifecycle, and customer historical data. This overcomes the limitations of traditional data sources, standardizes all data to ensure data quality and consistency, and extracts explicit and implicit customer needs in a layered manner, transforming them into computable feature vectors, thus breaking through the limitation of focusing only on explicit needs. The method calculates the degree of need matching through feature vector matching, generates a recommendation list with reasonable justifications based on priority, improving the accuracy and persuasiveness of recommendations. After push notifications, interactive feedback is recorded for iterative optimization, forming a closed loop. This overall upgrade of used car recommendations from "single-dimensional matching" to "full data support, considering both explicit and implicit needs" helps car dealers accurately connect with customer needs, thereby improving conversion efficiency. The method includes the following steps: In response to customer demand response triggering operations initiated by the car dealer terminal, the data collection layer calls a preset set of interfaces to collect all the source data required to build customer profiles and vehicle source profiles. The full source data includes private message interaction data between car dealers and customers, vehicle source data related to the entire life cycle of vehicles, and customer historical data related to past vehicle inquiries and transactions. The full source data is structured, the unstructured data is cleaned and format converted, and the structured data is normalized to a unified dimension to obtain standardized processed data. Based on the standardized processing data, the customer profile module and vehicle source profile module of the profile construction layer are used to extract multi-dimensional customer demand features and multi-dimensional vehicle attribute features, respectively. The multi-dimensional customer demand features include explicit demand parameters and implicit demand parameters, and dimension-matched customer feature vectors and vehicle source feature vectors are generated. The customer feature vector and the vehicle source feature vector are input into the recommendation calculation layer to calculate the demand fit between each vehicle source feature vector and the customer feature vector. All vehicle listings are prioritized based on the required matching degree, and a vehicle recommendation list containing core vehicle listing information and the reasons for the required matching degree is generated. The result output layer pushes the vehicle recommendation list to the corresponding car dealer terminal and records the customer's subsequent interaction feedback data on the recommendation list for iterative optimization of the recommendation calculation layer.

[0005] This invention responds to dealer terminal trigger operations by integrating used car-specific data sources such as private message interaction data, vehicle lifecycle data, and customer historical consultation and transaction data through a data collection layer. This comprehensively covers key information on customer needs and vehicle attributes, overcoming the limitations of traditional solutions in terms of data source scope and difficulty in meeting the core needs of the used car scenario. The entire dataset undergoes structured processing, cleaning and transforming unstructured data and normalizing structured data to ensure standardization and consistency, providing a high-quality data foundation for subsequent profile building. The profile building layer extracts explicit and implicit customer needs parameters, generating multi-dimensional customer feature vectors. Simultaneously, it mines multi-dimensional vehicle attribute features to form vehicle feature vectors, breaking through the limitations of traditional solutions that only focus on explicit needs and have a single matching dimension, achieving deep adaptation between needs and attributes. The recommendation calculation layer calculates the need fit through feature vector matching and generates a recommendation list containing core information and matching reasons according to priority, improving the persuasiveness of recommendations and ease of use for dealers. Customer interaction feedback data is recorded after push notifications to iteratively optimize the recommendation calculation layer, forming a closed-loop mechanism of "data collection-processing-profile-matching-feedback-optimization". The overall solution upgrades used car recommendations from "single-dimensional matching" to "full data support and in-depth mining of explicit and implicit needs," significantly improving recommendation accuracy, helping car dealers quickly connect with customers' real needs, and improving conversion efficiency and customer satisfaction.

[0006] Preferably, the step of collecting the full amount of source data required to construct customer profiles and vehicle source profiles by calling a preset set of interfaces through the data acquisition layer includes the following steps: The system calls the open interfaces of social platforms and instant messaging SDKs to collect private message interaction data in real time. The private message interaction data includes customer demand description text data and interaction time sequence correlation data. The system calls the vehicle condition detection system interface, the maintenance record query platform interface, and the vehicle source interface of the car dealer SaaS system to integrate and collect vehicle source data, which includes vehicle technical condition detection data, maintenance history data, and basic vehicle attribute data. Call the customer data interface of the car dealer SaaS system to collect customer historical data, which includes past inquiry records of car models, price feedback records and attribute data of historically sold car models. Based on the data source types of the private message interaction data, vehicle source data, and customer historical data, corresponding collection priorities and data update frequencies are configured for each type of data, and the collection and synchronization of all source data are completed according to the collection priorities and data update frequencies.

[0007] This invention addresses the core pain points of insufficient integration of dedicated data sources and lack of targeted data collection in the used car scenario through multi-interface integrated data collection and refined data collection management. It utilizes social platform interfaces and instant messaging SDKs to collect private message interaction data, integrates vehicle condition inspection and maintenance system interfaces to collect vehicle source data, and retrieves historical customer data from SaaS system interfaces. This comprehensively covers key information on customer needs and vehicle attributes, overcoming the limitations of traditional solutions in terms of data source scope and difficulty in meeting the core needs of used car live-streaming private messaging scenarios. By configuring collection priorities and update frequencies for different data types, it ensures that core data is collected first and synchronized in real time, avoiding problems of disordered data collection and insufficient timeliness. This "multi-interface integration + priority-based collection" system achieves comprehensive, efficient, and accurate collection of full-volume source data specific to used cars, providing complete and high-quality data support for subsequent customer and vehicle profile construction, and completely changing the current situation of weak data foundation and poor scenario adaptability of traditional recommendation solutions.

[0008] Preferably, the extraction of explicit requirement parameters based on standardized data processing includes the following steps: Standardized private message text data and explicit demand description fragments from historical consultation records are filtered out from the standardized data. Combined with the contextual information of the customer's current consultation session, a raw text set of explicit demands with context tags is obtained. The original text set of explicit requirements with context tags is dynamically segmented. Specifically, based on the new word discovery algorithm in the vehicle domain, the segmentation granularity, semantic dependency analysis and domain stop word filtering are adjusted in real time to retain the core semantic components of the requirements and context-related words, so as to obtain context-enhanced purified text fragments. The dynamic demand keyword dictionary that integrates customer behavior feedback is invoked to perform bidirectional semantic matching on the context-enhanced clean text fragment to obtain keyword matching results and context association weight set; Based on the keyword matching results and contextual association weight set, identify the core and derived demand dimensions explicitly stated by the customer, extract the fuzzy constraint descriptions corresponding to each demand dimension and perform semantic quantification to obtain a demand dimension-quantified constraint mapping table. Based on the aforementioned demand dimension-quantification constraint mapping table and combined with the vehicle configuration attribute association graph, demand value information and associated demand attributes are extracted from the context-enhanced clean text fragments to obtain a preliminary set of demand tuples with associated attributes. The domain rule engine, which integrates logical reasoning and scenario adaptation, is invoked to perform conflict detection, redundancy removal, and intelligent completion of missing related requirement dimensions on the preliminary requirement tuple set with associated attributes, so as to obtain a valid requirement tuple set after verification. The verified set of valid demand tuples is mapped to a preset dynamic explicit demand parameter system. A unique parameter identifier and scenario adaptation coefficient are assigned to the demand dimension of each tuple. The demand value is converted into a structured parameter value and associated with the parameter confidence. The set of confidence-enhanced structured demand key-value pairs is calculated based on keyword matching strength and context association weight. Based on the customer's historical explicit demand preference model, the set of confidence-enhanced structured demand key-value pairs is optimized to form explicit demand parameters containing parameter identifiers, structured parameter values, scenario adaptation coefficients, and confidence levels.

[0009] This invention precisely addresses the shortcomings of traditional solutions in accurately extracting explicit customer needs and lacking standardized processing mechanisms through refined text processing and structured parameter extraction throughout the entire process. It filters the original text of explicit needs and combines it with contextual information to ensure the relevance of the extracted needs to the specific context. Dynamic word segmentation and domain purification techniques are employed to retain core semantics and contextually relevant words, improving the accuracy of text processing. A dynamic keyword dictionary and domain rule engine are used to complete keyword matching, conflict detection, redundancy removal, and need completion, ensuring the validity and completeness of the need tuples. Valid need tuples are mapped to a preset parameter system to generate explicit need parameters containing parameter identifiers, structured values, adaptation coefficients, and confidence levels, achieving the structured and quantitative transformation of explicit needs. The entire process, from text processing to parameter generation, forms a complete standardized workflow, thoroughly solving the shortcomings of traditional solutions in extracting explicit needs in a vague manner and being difficult to transform into computable features. This provides accurate and structured core data support for constructing customer feature vectors.

[0010] Preferably, the process of identifying the core and derived needs dimensions explicitly stated by the customer based on the keyword matching results and contextual association weight set includes the following steps: Data deconstruction is performed on keyword matching results and context association weight sets to extract semantic attributes, context association objects, and association strength information of matched keywords, generating a keyword semantic-association feature dataset; Based on the keyword semantic-association feature dataset, a semantic similarity algorithm is used to associate and aggregate semantically similar keywords to generate multiple semantically associated keyword clusters. Each cluster corresponds to a set of potential associated demand expressions. Extract the intra-cluster semantic identifiers and inter-cluster semantic distance parameters of each semantically related keyword cluster. Combine the knowledge graph of vehicle domain requirements dimensions to perform dimension mapping and adaptation on each semantically related keyword cluster to obtain a preliminary dimension candidate set. Call the requirement dimension semantic verification model to perform bidirectional semantic verification between each dimension in the initial dimension candidate set and the keyword semantic-association feature dataset, and generate dimension semantic fit parameters and fit confidence basis; Based on the semantic fit parameters and fit confidence criteria, a dimension validity screening algorithm is used to remove invalid dimensions whose semantic fit does not meet the verification criteria, thus obtaining a set of valid requirement dimensions. For each dimension in the set of effective demand dimensions, the demand dimension logical association model is invoked to analyze the semantic implication relationship, causal inference relationship and scenario association relationship between dimensions, and generate a dimension logical association graph. Based on the dimensional logical association graph, identify dimensional nodes with semantic radiation effect in the graph, and determine the core requirement dimension and the dominant parameters of the dimension; Based on the dominant parameters of dimensions and the logical relationship graph of dimensions, we can mine the related dimensions formed by the semantic extension and logical deduction of the core requirement dimensions, and determine the derived requirement dimensions.

[0011] This invention comprehensively addresses the core pain points of traditional solutions, such as one-sided identification of demand dimensions and unclear distinction between core and derived demands, through semantic aggregation, dimensional mapping, and logical association analysis. It deconstructs keyword matching results and associated weight sets to extract semantic and association features, laying the foundation for dimensional identification. A semantic similarity algorithm is used to aggregate keyword clusters, combined with a domain knowledge graph for dimensional mapping, ensuring the accuracy of demand dimension identification. Through a demand dimension semantic verification model and validity screening algorithm, invalid dimensions are eliminated, resulting in a precise set of valid demand dimensions. The logical relationships between dimensions are analyzed to generate a dimensional logical association graph, accurately identifying core and derived demand dimensions with semantic radiation effects. This step overcomes the limitations of traditional solutions that simply extract keywords and lack in-depth dimensional analysis, achieving hierarchical identification and association mining of demand dimensions. This makes the structure of customers' explicit demands clearer and more comprehensive, providing a more accurate dimensional basis for subsequent demand quantification and vehicle source matching, significantly improving the targeting and accuracy of recommendation matching.

[0012] Preferably, the extraction of implicit requirement parameters based on standardized data processing includes the following steps: By filtering scenario description texts, behavioral tendency records, and emotional expression content from customer interaction data in standardized data processing, and combining them with the vehicle usage scenario classification system, a raw dataset of scenario-related implicit needs is generated. Cross-modal semantic parsing is performed on the original dataset of scene-related implicit needs. Textual semantic features and scene attribute features are integrated to extract information on the user, behavioral intention, scene constraints and sentiment, and generate a cross-modal demand semantic feature set. Input the cross-modal demand semantic feature set into the scene intent recognition model trained based on contrastive learning, mine the potential use scenarios not directly expressed in the text, and generate scene intent labels and scene matching criteria; Based on scene intent tags and scene matching criteria, the dynamic scene-demand dimension mapping library is invoked, and the core demand dimension corresponding to the scene is matched through semantic reasoning algorithm to generate a scene-demand dimension association list. By combining customer historical consultation behavior sequences and vehicle source preference feedback data, we analyze the potential preference tendencies of each dimension in the scenario-demand dimension association list, and generate demand dimension-preference tendency association parameters through sequence pattern mining algorithms. Based on the scenario-demand dimension association list and the demand dimension-preference tendency association parameters, an implicit demand semantic vector is constructed. The demand dimension features corresponding to the core scenario are strengthened through the scenario attention mechanism to generate a scenario-enhanced implicit demand vector. The scenario-enhanced implicit demand vector is input into the generative parameter transformation model, which converts the vector features into structured demand parameters and generates a candidate set of implicit demand parameters. The requirement parameter conflict resolution model is invoked, and the implicit requirement parameter candidate set is semantically consistent with the explicit requirement parameters. Conflicting parameters are eliminated and derived parameters are completed to generate implicit requirement parameters containing dimension identifiers, parameter values, and scenario adaptability.

[0013] This invention completely solves the core pain points of traditional solutions, namely, their inability to extract implicit customer needs and their limited matching dimensions, through cross-modal parsing, scene intent mining, and structured parameter transformation. It filters raw data on implicit needs, such as scene descriptions and behavioral tendencies, and combines this with a vehicle usage scenario classification system to ensure the scenario-specificity of demand mining. Cross-modal semantic parsing integrates text and scene features to extract multi-dimensional core information, overcoming the limitations of single-text analysis. A scene intent recognition model trained through contrastive learning is used to mine potential scenarios not directly expressed, and a dynamic mapping library and semantic reasoning are combined to match demand dimensions, achieving precise positioning of implicit needs. Analysis of customer historical behavior sequences generates preference-related parameters, and a scene attention mechanism is used to strengthen core features, constructing a scene-enhanced implicit demand vector to make implicit needs more identifiable. Through generative model transformation and conflict resolution, parameters that conflict with explicit needs are eliminated, and derived parameters are supplemented to generate standardized implicit demand parameters. The "scene mining - semantic parsing - vector construction - parameter transformation" system built in this step is the first to realize the structured and computable transformation of implicit needs in the used car scenario. It makes up for the shortcomings of traditional solutions that only focus on explicit needs, provides complete dimensional support for customer feature vector construction, and greatly improves the depth and accuracy of recommendation matching.

[0014] Preferably, the extraction of multi-dimensional vehicle attribute features based on the standardized processed data through the vehicle source profiling module of the profiling construction layer includes the following steps: From the standardized data subset related to vehicle sources, we separate vehicle basic attribute data, vehicle condition inspection data, maintenance history data and market circulation data to generate a multi-source original dataset of vehicle sources. Data association processing is performed on the original datasets of vehicles from multiple sources to match and integrate data from different sources using unique vehicle identifiers, generating an integrated dataset of vehicle sources with association identifiers; Based on the vehicle source integration dataset, basic vehicle attribute information is extracted and converted into structured basic attribute parameters through a vehicle configuration coding parsing algorithm, generating a set of basic vehicle attribute parameters. Feature mining is performed on the vehicle condition detection data in the vehicle source integration dataset to analyze the correlation between detection indicators and potential faults, and to generate vehicle health assessment parameters, which include dimensional parameters corresponding to the integrity of the vehicle body structure and the stability of the power system. The maintenance frequency, key component replacement cycle, and maintenance compliance information are extracted from historical maintenance data to generate maintenance feature parameters, which reflect the vehicle's usage and maintenance status. By combining market circulation data from the vehicle source integration dataset with market price data for the same vehicle model, vehicle residual value and market circulation activity information are calculated, and market attribute parameters are generated. By calling the vehicle attribute association graph, semantic association is performed on the vehicle source basic attribute parameter set, vehicle condition health assessment parameters, maintenance characteristic parameters and market attribute parameters, and potential association relationships between each parameter are explored to generate the vehicle source attribute association parameter set. By performing feature filtering and fusion on the set of vehicle source attribute associated parameters, redundant parameters are eliminated and the core attribute representation is strengthened to obtain multi-dimensional attribute features of vehicles.

[0015] This invention precisely addresses the shortcomings of traditional solutions in extracting vehicle attribute features in a one-sided manner and lacking in-depth correlation analysis by integrating multi-source vehicle data, extracting features across all dimensions, and mining correlations. It separates the original multi-source vehicle data and integrates them using unique identifiers to generate a vehicle dataset with correlation identifiers, ensuring data integrity and relevance, and avoiding the drawbacks of traditional data dispersion and difficulty in comprehensive analysis. Basic attributes are extracted and parsed into structured parameters; vehicle condition detection data is mined to generate health assessment parameters; maintenance information is extracted to generate feature parameters; and market attribute parameters such as residual value are calculated by combining market data, achieving full-dimensional coverage of vehicle attributes. By calling the vehicle attribute correlation graph, potential relationships between parameters are mined to generate an attribute correlation parameter set, overcoming the limitations of traditional solutions that extract features in isolation and lack correlation analysis. Redundant parameters are eliminated through feature filtering and fusion, strengthening the representation of core attributes and obtaining accurate and comprehensive multi-dimensional vehicle attribute features. The "data integration - multi-dimensional extraction - correlation mining - feature optimization" system built in this step fully covers the core attribute dimensions such as vehicle basics, vehicle condition, maintenance, and market, realizing in-depth mining and structured presentation of vehicle source features. It provides rich and accurate data support for the construction of vehicle source feature vectors, and completely changes the status quo of traditional solutions with single vehicle source feature dimensions and poor scenario adaptability.

[0016] Preferably, the step of inputting the customer feature vector and vehicle source feature vector into the recommendation calculation layer to calculate the demand fit between each vehicle source feature vector and the customer feature vector includes the following steps: Extract the core demand dimension identifiers and feature distribution features from the customer feature vector, and simultaneously extract the core attribute dimension identifiers and attribute distribution features from the vehicle source feature vector to generate a supply and demand dimension feature comparison set. Based on the supply and demand dimension feature comparison set, a mapping relationship between the customer demand dimension and the vehicle source attribute dimension is established through semantic association reasoning, generating a dimension mapping association table and mapping confidence parameters; Based on the dimension mapping association table, the customer feature vector and the vehicle source feature vector are dimension aligned, retaining the effective dimensions that are mapped to each other and eliminating the redundant dimensions that are not related, and generating supply and demand aligned vector pairs. Perform feature distribution difference analysis on the supply and demand alignment vector pairs, calculate the distribution fit parameters of customer demand features and vehicle source attribute features under each alignment dimension, and generate a dimension fit matrix. The requirement-attribute association strength model is invoked, and the dimension mapping confidence parameters and dimension fit matrix are combined to explore the potential association logic between dimensions and generate dimension association logic coefficients. Calculate the overall correlation tightness parameter of the supply and demand alignment vector pairs based on the dimensional fit matrix and dimensional correlation logical coefficient, and generate the vector correlation strength value. By combining the priority sequence of customer needs and the importance sequence of vehicle source attributes, the vector association strength value is modified for scenario adaptation, and scenario adaptation association parameters are generated. Based on the scenario adaptation and correlation parameters, the dimension fit matrix, dimension correlation logic coefficient and vector correlation strength value are integrated to generate the demand fit between each vehicle source feature vector and the customer feature vector.

[0017] This invention comprehensively addresses the core pain points of traditional solutions, such as single-dimensional demand matching and insufficient accuracy in calculating fit, through dimensional mapping, alignment analysis, and multi-factor fusion calculation. It extracts core dimensional identifiers and distribution features of customers and vehicle sources, establishing a supply-demand dimensional comparison set to lay the foundation for accurate matching. Semantic association reasoning is used to construct dimensional mapping relationships and generate confidence parameters, ensuring the rationality and reliability of dimensional matching. Dimensional alignment processing is performed on feature vectors to eliminate redundant dimensions, generating supply-demand aligned vector pairs and avoiding interference from irrelevant dimensions in the matching results. The distribution fit of each aligned dimension is calculated to generate a dimensional fit matrix, achieving refined evaluation of matching dimensions. Potential correlation logic between dimensions is mined, and combined with the priority of customer needs and the importance sequence of vehicle source attributes, the vector correlation strength value is adjusted for scenario adaptation to generate a comprehensive demand fit. The multi-level calculation system constructed in this step, consisting of "dimensional mapping, alignment analysis, dimensional fit, association enhancement, and scenario correction," breaks through the limitations of traditional solutions that rely on single-dimensional matching and lack comprehensive consideration. It achieves deep adaptation calculation of supply and demand characteristics in the used car scenario, making the demand fit more reflective of the degree of matching between customers' real needs and vehicle attributes, and providing accurate and objective quantitative basis for subsequent vehicle priority ranking.

[0018] As a preferred embodiment, the second technical solution of the present invention is a vehicle recommendation information processing system, used to execute the vehicle recommendation information processing method described above. This system includes a data acquisition layer, a profile construction layer, a recommendation calculation layer, and a result output layer, with each layer interacting through a data interface. The data acquisition layer collects full-volume source data through multi-source interfaces, including private message interaction data, vehicle source data, and customer historical data. It performs structured processing on the full-volume source data to obtain standardized processed data and stores it in a distributed database. The profile construction layer calls the standardized processed data from the distributed database and generates customer feature vectors and vehicle source feature vectors through explicit demand extraction, implicit demand mining, and vehicle source feature quantification. The recommendation calculation layer calculates the demand fit between each vehicle source feature vector and the customer feature vector based on the customer feature vector and the vehicle source feature vector. The result output layer generates a vehicle recommendation list containing reasons for fit based on the demand fit, pushes it to the car dealer terminal, records customer feedback data, and feeds it back to the recommendation calculation layer for iterative optimization.

[0019] This invention, through a layered architecture design and end-to-end collaborative linkage, completely solves the core pain points of loose recommendation system architecture, inefficient data flow, and poor scenario adaptability in the used car live-streaming private messaging scenario. The data acquisition layer integrates multiple source interfaces to collect full-volume source data specific to used cars, which is then structured and stored in a distributed database, ensuring data comprehensiveness, standardization, and storage security, thus overcoming the limitations of traditional solutions in terms of limited data sources and disordered data processing. The profile building layer calls standardized data, extracting explicit needs, mining implicit needs, and quantifying vehicle source features to generate complete-dimensional customer and vehicle source feature vectors. This achieves layered analysis and structured transformation of explicit and implicit needs, breaking through the limitations of traditional solutions that only focus on explicit needs and have a single matching dimension. The recommendation calculation layer accurately calculates the demand fit based on feature vectors, and the result output layer generates a recommendation list with reasonable justifications and pushes it to the car dealer's terminal. Simultaneously, customer feedback is recorded for iterative optimization, forming a closed-loop system of "collection-processing-profile-matching-feedback-optimization". Each layer interacts efficiently through data interfaces, building a professional and systematic processing architecture adapted to the used car scenario. This completely changes the traditional recommendation system architecture's generalization and poor scenario adaptability, achieving efficient operation and accurate adaptation of the entire recommendation process, and significantly improving the efficiency and conversion rate of car dealers in meeting customer needs.

[0020] As a preferred embodiment, the third technical solution of the present invention is a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the vehicle recommendation information processing method described above.

[0021] This invention precisely addresses the shortcomings of traditional solutions, such as lack of stable hardware and inefficiency in handling large-scale data scenarios, by linking hardware support and program execution with computer equipment. The computer's memory provides ample storage space for the vehicle recommendation information processing method, stably storing various key information including full-source data, standardized processed data, feature vectors, recommendation models, and customer feedback data, avoiding the drawbacks of scattered data storage and inefficient access. When the processor executes the computer program, it efficiently schedules the entire process, including data acquisition, structured processing, profile construction, fit calculation, and result output. With hardware computing power, it achieves rapid integration of multi-source data, efficient computation of complex algorithms, and real-time matching of large-scale feature vectors, overcoming the limitations of traditional solutions that rely on general-purpose hardware, have slow processing speeds, and struggle to handle the high-concurrency data processing demands of the used car scenario. This step deeply integrates the vehicle recommendation information processing method with the computer equipment, ensuring the stable implementation and efficient execution of the recommendation logic. It provides reliable hardware and software collaboration support for real-time and accurate recommendations in the used car live-streaming private messaging scenario, completely changing the low efficiency and poor stability of traditional recommendation solutions.

[0022] As a preferred embodiment, the fourth technical solution of the present invention is: a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the vehicle recommendation information processing method described above.

[0023] This invention comprehensively addresses the core pain points of traditional solutions—limited application and difficulty in cross-platform reuse—by utilizing computer-readable storage media for program storage and cross-device adaptation. The computer-readable storage media stably stores the computer program corresponding to the vehicle recommendation information processing method, ensuring program integrity and portability, and avoiding the drawbacks of traditional solutions where programs rely on specific hardware and are difficult to copy and distribute. The computer program in the storage medium can be read and executed by different types of processors, enabling cross-platform deployment and reuse of the recommendation method on various computer devices. Whether it's car dealer terminal equipment, cloud servers, or other data processing devices, the corresponding recommendation system can be quickly deployed through this storage medium, significantly reducing the cost and barriers to system promotion and application. This step provides a flexible and convenient storage and dissemination carrier for the vehicle recommendation information processing method, ensuring the wide adaptability and large-scale application of the recommendation logic. It completely changes the current situation of narrow application scope and high promotion difficulty of traditional recommendation solutions, providing strong support for the popularization of personalized recommendation technology in the used car industry, allowing more car dealers to easily enjoy the efficiency improvements brought by accurate recommendations.

[0024] It has the following beneficial effects: (1) By constructing a modular, full-link microservice architecture, the core pain points of traditional frameworks, such as limited compatibility and insufficient component collaboration, have been completely resolved, laying a solid foundation for multi-language development and large-scale deployment. The architecture covers five core components, including service providers, consumers, and a registry center. Each component has a clear responsibility and works in synergy, breaking away from the single dependency of Java and Spring ecosystems and supporting flexible access to services developed in different languages, thus overcoming the barriers of insufficient cross-language compatibility in traditional frameworks. The service registry center supports multi-copy storage, ensuring the reliability and high availability of service information; the service runtime container standardizes the management of service loading and lifecycle, reducing the complexity of service deployment and maintenance; the service monitoring console enables real-time control of runtime status, solving the drawbacks of scattered monitoring and difficult problem localization in traditional architectures. This architecture design not only ensures the stability and scalability of the microservice system, but also provides a unified support carrier for subsequent service registration and discovery, data transmission optimization, and containerization integration, perfectly adapting to the core needs of enterprise-level multi-team collaboration, multi-language development, and large-scale deployment.

[0025] (2) By optimizing service registration, multi-replica storage, dynamic health status synchronization, and local caching, the problems of unreliable service registration, delayed status awareness, and low call efficiency in traditional frameworks are precisely solved. Service providers automatically register core service information upon startup, eliminating the need for manual configuration and reducing the risk of human error. The service registry adopts a multi-replica distributed storage and ephemeral node mechanism, ensuring that service information is not lost and automatically deregistering metadata when a service goes offline, avoiding invalid service residues. The heartbeat mechanism allows the registry to perceive the health status of services in real time and promptly pushes changes to consumers, solving the drawback of untimely service status updates in traditional frameworks. Service consumers prioritize using the cache to initiate calls through local caching and incremental update mechanisms, significantly reducing the frequency of interaction with the registry, reducing network overhead and call latency. At the same time, the cache is actively pulled for updates when it expires, balancing call efficiency and information timeliness. The end-to-end service registration and status synchronization mechanism built in this step improves the high availability and real-time performance of service discovery, providing a reliable guarantee for the stable operation of large-scale microservice clusters.

[0026] (3) By using a custom serialization protocol and flexible encoding / decoding adaptation, the core pain points of traditional frameworks, such as poor serialization compatibility, low transmission efficiency, and difficulty in adapting to complex scenarios, are comprehensively solved. The custom serialization protocol, designed based on enterprise-level requirements, supports the transmission of complex data types and version compatibility, breaking the limitations of traditional serialization schemes that are only suitable for a single scenario and prone to version iteration conflicts. It provides multiple serialization schemes and corresponding codecs, and combined with a unified management module, it enables dynamic selection and can be flexibly adapted according to service configuration or calling scenario to meet the performance and compatibility requirements of different data transmissions. Serialization encoding converts objects into compact binary data, reducing network transmission volume and bandwidth consumption. Deserialization ensures accurate data restoration and avoids cross-service data parsing anomalies. Combined with a load balancing strategy, requests are distributed reasonably to avoid single-point service overload and improve the stability and efficiency of the overall call chain. The efficient, flexible, and compatible serialization and transmission system built in this step not only enhances the framework's cross-language adaptability but also significantly improves the data transmission efficiency between microservices, perfectly supporting the high-frequency and complex data interaction needs in multi-language development scenarios.

[0027] (4) Through deep integration with the container orchestration platform, elastic scheduling, automated operation and maintenance, and visual monitoring of microservice resources were achieved, completely overcoming the shortcomings of traditional frameworks, such as lack of dynamic adaptation capabilities and complex operation and maintenance of large-scale deployments. Leveraging the native service discovery and configuration management functions of the container orchestration platform, service registration and discovery were automated and configuration was centrally managed, reducing manual intervention costs and avoiding distributed configuration inconsistencies. Utilizing the platform's powerful resource scheduling capabilities, the number of instances was dynamically adjusted based on service call load, automatically scaling up during peak periods to ensure service availability and scaling down during off-peak periods to save resources, solving the problems of rigid resource allocation and low utilization in traditional frameworks. A unified service management interface and monitoring tools integrated the container platform and the framework's own operational metrics, allowing real-time visualization of key information such as the number of service instances, health status, and call frequency, supporting abnormal alarms and rapid problem localization, significantly reducing the operational difficulty of large-scale microservice clusters. This step achieved seamless integration of microservices and the container ecosystem, improving system scalability and operational efficiency, and providing strong support for enterprise-level large-scale, high-concurrency deployment scenarios, significantly enhancing the practicality and feasibility of the microservice architecture. Attached Figure Description

[0028] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the steps of the vehicle recommendation information processing method of the present invention; Figure 2 This is a schematic diagram of the vehicle recommendation information processing system of the present invention; Figure 2 The layers are labeled as follows: 10, Data Acquisition Layer; 20, Profile Building Layer; 30, Recommendation Calculation Layer; 40, Result Output Layer. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0030] To achieve the above objectives, please refer to Figure 1 Embodiment 1 of the present invention provides a method for processing vehicle recommendation information, including the following steps: S01: In response to the customer demand response trigger operation initiated by the car dealer terminal, the data collection layer 10 calls the preset interface set to collect the full source data required to build the customer profile and vehicle source profile. The full source data includes private message interaction data between the car dealer and the customer, vehicle source data related to the entire life cycle of the vehicle, and customer historical data related to the customer's past vehicle inquiries and transactions. In this embodiment of the invention, in the used car trading platform, the car dealer selects the consultation session of customer "ID8975" in the terminal backend, clicks the "Generate Recommendation" button to initiate a customer demand response trigger operation, and the system starts the data acquisition layer 10 after receiving the trigger signal. The system calls a pre-defined set of interfaces: Through the Douyin Open Platform API (request frequency set to real-time, response timeout 3 seconds) and WeChat Merchant Edition SDK (using a long connection protocol), it collects private message interaction data between the car dealer and the customer. This includes the customer's message sent at 09:15 on 2024-12-05: "Budget under 150,000 RMB, looking to buy a 7-seater MPV, mainly for company reception, preferably less than 2 years old with complete maintenance records," and the addition at 09:20: "Not white body, preference for joint venture brands," as well as interaction sequence data (two messages 5 minutes apart, customer did not withdraw or modify content); it calls the self-developed vehicle condition detection system HTTP interface (returning data in JSON format) to obtain vehicle condition data for all 7-seater MPVs on the platform, such as the 2022 Buick GL8: body structure inspection score 94, engine condition score 92, no major accident records; and adjusts... Using the SOAP interface of a third-party maintenance query platform (API key authentication required), we collect vehicle maintenance history, such as the Buick GL8 having been maintained 4 times in the past 2 years, all at authorized 4S stores, including tire and oil changes; we call the vehicle source management interface of the car dealer SaaS system (using RESTful style) to obtain basic vehicle source data, such as the Buick GL8 2022 model priced at 142,000 yuan, registration date in November 2022, and mileage of 28,000 kilometers; we call the customer data interface of the car dealer SaaS system to collect the customer's historical data: in September 2024, the customer inquired about the Honda Odyssey 2021 model, stating that the 165,000 yuan price was "beyond budget," there were no transaction records in 2023, and the customer purchased a Volkswagen Passat for company use in 2022. We complete the full source data collection, and all data is accompanied by a collection timestamp and data source identifier.

[0031] S02: Perform structured processing on the full source data, clean and convert the unstructured data, and normalize the structured data to a unified dimension to obtain standardized processed data. In this embodiment of the invention, structured processing is performed on the collected full source data. Unstructured data processing: Private message text is processed using an NLP word segmentation algorithm based on a vehicle domain dictionary (the dictionary contains professional terms such as "7-seater MPV", "company reception", "joint venture brand" etc.), removing meaningless function words such as "want" and "hope", filtering the negative word "don't" in "don't want white body" and retaining the core constraint "body color ≠ white", and converting it into a standardized text string of "7-seater MPV company reception vehicle within budget of 150,000, maintenance records of vehicle age less than 2 years old, body color ≠ white, joint venture brand preferred"; the descriptive text "no obvious scratches on the body, slight dent on the left door has been repaired" in the vehicle condition inspection report is processed by an entity recognition algorithm to extract the structured field "body scratches: none, door dent: yes (repaired)". Structured data processing: Using a unit normalization algorithm, all vehicle prices are uniformly converted to "ten thousand yuan" (retaining one decimal place, e.g., 142,000 yuan is converted to 142,000 yuan), vehicle age is uniformly converted to "years" (accurate to 0.1 years, e.g., from November 2022 to December 2024, the vehicle age is calculated as 2.1 years), and mileage is uniformly converted to "ten thousand kilometers" (retaining one decimal place, e.g., 28,000 kilometers is converted to 28,000 kilometers). Using an extreme value normalization algorithm, the vehicle condition detection score (out of 100) is converted to a value in the range of 0-1 (94 points is calculated as 94 / 100=0.94, 92 points is calculated as 0.92), resulting in standardized processed data containing structured text and normalized values. All data fields are labeled with data type (text, numeric, date) and integrity indicators (complete / missing / abnormal).

[0032] S03: Based on the standardized processing data, the customer profile module and vehicle source profile module of the profile construction layer 20 are used to extract multi-dimensional customer demand features and multi-dimensional vehicle attribute features respectively. The multi-dimensional customer demand features include explicit demand parameters and implicit demand parameters, and dimension-matched customer feature vectors and vehicle source feature vectors are generated. In this embodiment of the invention, standardized processing data is loaded through the customer profile module of the profile construction layer 20. The explicit demand extraction rule engine (with built-in mapping rules such as "budget-price", "vehicle model-vehicle type", and "vehicle age-usage duration") extracts explicit demand parameters from private messages and historical consultation records: budget range 120,000-150,000 (derived based on the customer's preference for "within 150,000" and rejection of a 165,000 quote), vehicle type 7-seater MPV, vehicle age ≤ 2 years, body color ≠ white, brand type joint venture, and usage scenario company reception. The implicit demand mining model, finely tuned based on BERT (training data includes 500,000 customer consultation-demand corresponding samples), analyzes the "company reception" scenario to mine implicit demand parameters: interior material ≥ leather (reception scenario requires a sense of quality), air conditioning system with zone control (meets the comfort needs of multiple passengers), safety configuration including side airbags (safety requirements for business scenarios), and trunk volume ≥ 500L (reception storage needs). The explicit and implicit demand parameters are quantified into a 50-dimensional customer feature vector. For example, a budget of 120,000-150,000 corresponds to a feature value of 0.6 (0-1 range, the higher the value, the higher the budget), a 7-seater MPV corresponds to a feature value of 1.0 (unique hot coding of vehicle type dimension), a vehicle age of ≤2 years corresponds to a feature value of 0.9, and a joint venture brand corresponds to a feature value of 0.85. The remaining dimensions are filled with default values ​​according to the rules. The vehicle source profiling module loads standardized processed data and extracts multi-dimensional attribute features of the vehicle. Taking the 2022 Buick GL8 as an example, the basic attributes are: price 142,000 yuan, model 7-seater MPV, vehicle age 2.1 years, body color black, brand joint venture, interior leather, air conditioning dual-zone control, side airbag configuration, trunk volume 521L; quality attributes (vehicle condition score 0.93, maintenance completeness 0.96 (4 maintenance services / average of 3.5 maintenance services for vehicles of the same age), no major accidents corresponding to 0.98); and market attributes (average transaction price of the same model in the past 3 months 145,000 yuan, registration duration 8 days). These attribute features are quantified into a 50-dimensional vehicle source feature vector, which corresponds one-to-one with the dimensions of the customer feature vector. For example, a price of 142,000 yuan corresponds to a feature value of 0.58, a 7-seater MPV corresponds to 1.0, a vehicle age of 2.1 years corresponds to 0.88, and a joint venture brand corresponds to 0.9, ensuring dimension matching.

[0033] S04: Input the customer feature vector and vehicle source feature vector into the recommendation calculation layer 30 to calculate the demand fit between each vehicle source feature vector and the customer feature vector; In this embodiment of the invention, the customer feature vector and all 7-seater MPV vehicle source feature vectors are input into the recommendation calculation layer 30. First, a semantic association reasoning model (trained based on a vehicle domain knowledge graph) is used to establish a mapping relationship between the customer demand dimension and the vehicle source attribute dimension. For example, the customer's "budget 120,000-150,000" maps to the vehicle source "price" dimension, "7-seater MPV" maps to the "model" dimension, and "interior leather" maps to the "interior material" dimension. A dimension mapping association table is generated, and the confidence parameters of each mapping are labeled (model mapping confidence 1.0, price mapping confidence 0.95, interior material mapping confidence 0.9). Dimension alignment is performed on the customer and vehicle source vectors, retaining 28 valid dimensions with mapping relationships and removing unrelated dimensions (such as vehicle manufacturer codes), generating supply and demand alignment vector pairs. The Euclidean distance algorithm is used to calculate the fit parameter for each alignment dimension. For example, the fit between the customer's budget feature value of 0.6 and the Buick GL8 price feature value of 0.58 is 1-|0.6-0.58|=0.98; the fit between the customer's vehicle age feature value of 0.9 and the Buick GL8 vehicle age feature value of 0.88 is 0.98; and the fit between the customer's interior material feature value of 0.8 and the Buick GL8 interior material feature value of 1.0 is 0.92. This generates a dimension fit matrix. The requirement-attribute association strength model is invoked (trained based on historical transaction data, outputting the association weights between dimensions). For example, the association strength of "usage scenario-air conditioning configuration" is 0.85, and the association strength of "budget-price" is 0.9. Combining the dimension mapping confidence parameter and the fit matrix, the vector association strength value is calculated by weighted summation: Buick GL8 association strength value = (price fit 0.98 × association strength 0.9 × confidence 0.95) + (model fit 1.0 × association strength 1.0 × confidence 1.0) + (vehicle age fit 0.98 × association strength 0.88 × confidence 0.92) + ... + (interior fit 0.92 × association strength 0.8 × confidence 0.9), and finally 0.932 is obtained. By combining customer demand priorities (budget > model > vehicle age > brand > interior) and the importance of vehicle attributes (vehicle condition > price > configuration > vehicle age), the correlation strength value is adjusted for scenario adaptation. For example, in the scenario of "company reception", the weight of air conditioning configuration and safety configuration is increased by 10%, and the demand fit of Buick GL8 is 0.945 after adjustment; similarly, the fit of 2021 Honda Odyssey is calculated to be 0.87 and the fit of 2022 Toyota Cena is 0.91, so as to obtain the demand fit of all vehicle sources.

[0034] S05: Prioritize all vehicle sources based on the required matching degree, and generate a vehicle recommendation list containing core vehicle source information and reasons for the required matching degree; In this embodiment of the invention, all 7-seater MPV vehicles are sorted from high to low according to their suitability for demand. The sorting results are Buick GL8 2022 (0.945), Toyota Cena 2022 (0.91), Honda Odyssey 2021 (0.87), and Kia Carnival 2022 (0.82). The top 3 are selected to generate a vehicle recommendation list. Each vehicle listing includes key information: Buick GL8 2022 – Model name, price 142,000 RMB (3,000 RMB lower than the average price of similar models), 2.1 years old, 28,000 km mileage, black exterior, leather interior, dual-zone climate control, side airbags, condition score 0.93, 4 maintenance records (all 4S dealerships); Toyota Cinna 2022 – Price 138,000 RMB, 2.3 years old, 31,000 km mileage, beige exterior, fabric interior, single-zone climate control, side airbags, condition score 0.91, 3 maintenance records; Honda Odyssey 2021 – Price 148,000 RMB, 3.2 years old, 45,000 km mileage, black exterior, leather interior, dual-zone climate control, side airbags, condition score 0.89, 5 maintenance records. Each vehicle listing comes with a reasonable explanation for its suitability, such as the Buick GL8: "Matches your budget of 150,000 (quoted price of 142,000), the 7-seater MPV model meets your needs, the 2.1-year age is close to the expected 2 years, it is a joint venture brand, leather interior, and dual-zone air conditioning to meet the needs of company reception, the vehicle condition score is 0.93 and the maintenance record is complete," ensuring that the reasons are directly related to the customer's needs.

[0035] S06: The vehicle recommendation list is pushed to the corresponding car dealer terminal through the result output layer 40, and the customer's subsequent interaction feedback data on the recommendation list is recorded for iterative optimization of the recommendation calculation layer 30.

[0036] In this embodiment of the invention, the result output layer 40 pushes the recommendation list in two ways: First, it calls the private message interface of the car dealer's SaaS system to push the recommendation list to the customer conversation interface of the car dealer's terminal in the form of a structured message. The message includes links to car source images, a summary of core information and the reasons for the agreement, and supports the car dealer to forward it to the customer with one click; Second, it calls the push interface of the car dealer's mobile APP (using the APNs protocol, iOS) and the FCM protocol (Android) to send a notification that "the recommendation list has been generated" to the car dealer's mobile APP. Clicking the notification allows the user to view the complete list. Simultaneously, the feedback recording module is activated to record subsequent customer interactions using event tracking technology: On December 5, 2024, at 10:00 AM, the customer clicked to view the detailed vehicle condition report for the Buick GL8; at 10:08 AM, they inquired about a 5,000 yuan discount; and at 10:15 AM, they indicated they were considering this model. Recorded data is stored in the format of "timestamp-customer ID-vehicle ID-interaction type-interaction content," such as "2024-12-05 10:00-ID8975-GL82022-click to view-vehicle condition report" and "2024-12-05 10:08-ID8975-GL82022-inquiry-5,000 yuan discount." Every day at 3:00 AM, the system automatically synchronizes the previous day's feedback data to the training data pool of recommendation calculation layer 30 for monthly recommendation model iterations (using gradient descent to optimize model parameters). For example, based on the customer's focus on "discounts," subsequent models will add a "price elasticity" feature dimension to improve recommendation accuracy.

[0037] Furthermore, the step of collecting the full amount of source data required to construct customer profiles and vehicle source profiles by calling a preset set of interfaces through the data acquisition layer 10 includes the following steps: The system calls the open interfaces of social platforms and instant messaging SDKs to collect private message interaction data in real time. The private message interaction data includes customer demand description text data and interaction time sequence correlation data. The system calls the vehicle condition detection system interface, the maintenance record query platform interface, and the vehicle source interface of the car dealer SaaS system to integrate and collect vehicle source data, which includes vehicle technical condition detection data, maintenance history data, and basic vehicle attribute data. Call the customer data interface of the car dealer SaaS system to collect customer historical data, which includes past inquiry records of car models, price feedback records and attribute data of historically sold car models. Based on the data source types of the private message interaction data, vehicle source data, and customer historical data, corresponding collection priorities and data update frequencies are configured for each type of data, and the collection and synchronization of all source data are completed according to the collection priorities and data update frequencies.

[0038] In this embodiment of the invention, the data acquisition layer 10 calls a preset set of interfaces to collect all the source data required to construct customer profiles and vehicle profiles for the used car transaction scenario. It calls the Douyin Open Platform interface and the WeChat instant messaging SDK to collect real-time private message interaction data between car dealers and customers. This private message interaction data includes text data describing customer needs (such as "I want an SUV around 100,000 yuan, spacious enough to take elderly people and children, and no more than 3 years old" or "Budget within 150,000 yuan, preferring Japanese brands, with complete maintenance records") and interaction time-series related data (such as a customer sending initial requirements at 10:30 on 2024-12-01, adding "I don't want turbocharged models" at 10:35, and asking "Are there any suitable hybrid models?" at 10:40). The collection frequency is set to real-time reception with a delay of no more than 1 second. The system integrates and collects vehicle source data by calling the interfaces of its self-developed vehicle condition detection system, third-party maintenance record query platform, and vehicle source interface of the car dealer SaaS system. The vehicle condition detection system interface provides vehicle technical condition detection data (such as a body structure score of 92 points, an engine condition score of 88 points, and 1 fault code clearing record). The maintenance record query platform interface provides maintenance history data (5 maintenance services in the past 3 years, replacement of air conditioning filter and brake pads, and no major repair records). The vehicle source interface of the car dealer SaaS system provides basic vehicle source attribute data (model is Toyota RAV4 2022, registration date is October 2021, mileage is 32,000 kilometers, and price is 148,000 yuan). The collection frequency of the three types of data is set to be synchronized once per hour. The system utilizes the customer data interface of the car dealer SaaS system to collect historical customer data, including: past inquiries about car models (e.g., Honda CR-V 2020 and Nissan X-Trail 2021 in October 2024), price feedback records (expression that the 160,000 RMB price quote for the CR-V was too high, while there was no objection to the 150,000 RMB price quote for the X-Trail), and historical transaction data for car models (e.g., a Toyota Corolla 1.6L automatic sold in 2022, a family car). The data collection frequency is set to full synchronization daily at 2 AM. Data collection priorities are configured based on data source type: private message interaction data is the highest priority (directly reflecting real-time needs), vehicle source data is the medium priority (affecting recommendation accuracy), and customer historical data is the normal priority (assisting in demand judgment). Data update frequencies are configured as follows: private message interaction data is updated in real-time, vehicle source data is updated hourly, and customer historical data is updated daily. This configuration ensures the timely and complete collection and synchronization of all source data.

[0039] Furthermore, the extraction of explicit requirement parameters based on standardized data processing includes the following steps: Standardized private message text data and explicit demand description fragments from historical consultation records are filtered out from the standardized data. Combined with the contextual information of the customer's current consultation session, a raw text set of explicit demands with context tags is obtained. The original text set of explicit requirements with context tags is dynamically segmented. Specifically, based on the new word discovery algorithm in the vehicle domain, the segmentation granularity, semantic dependency analysis and domain stop word filtering are adjusted in real time to retain the core semantic components of the requirements and context-related words, so as to obtain context-enhanced purified text fragments. The dynamic demand keyword dictionary that integrates customer behavior feedback is invoked to perform bidirectional semantic matching on the context-enhanced clean text fragment to obtain keyword matching results and context association weight set; Based on the keyword matching results and contextual association weight set, identify the core and derived demand dimensions explicitly stated by the customer, extract the fuzzy constraint descriptions corresponding to each demand dimension and perform semantic quantification to obtain a demand dimension-quantified constraint mapping table. Based on the aforementioned demand dimension-quantification constraint mapping table and combined with the vehicle configuration attribute association graph, demand value information and associated demand attributes are extracted from the context-enhanced clean text fragments to obtain a preliminary set of demand tuples with associated attributes. The domain rule engine, which integrates logical reasoning and scenario adaptation, is invoked to perform conflict detection, redundancy removal, and intelligent completion of missing related requirement dimensions on the preliminary requirement tuple set with associated attributes, so as to obtain a valid requirement tuple set after verification. The verified set of valid demand tuples is mapped to a preset dynamic explicit demand parameter system. A unique parameter identifier and scenario adaptation coefficient are assigned to the demand dimension of each tuple. The demand value is converted into a structured parameter value and associated with the parameter confidence. The set of confidence-enhanced structured demand key-value pairs is calculated based on keyword matching strength and context association weight. Based on the customer's historical explicit demand preference model, the set of confidence-enhanced structured demand key-value pairs is optimized to form explicit demand parameters containing parameter identifiers, structured parameter values, scenario adaptation coefficients, and confidence levels.

[0040] In this embodiment of the invention, by filtering standardized private message text data ("SUV around 100,000 RMB with large space suitable for taking the elderly and children, and the car age is no more than 3 years", "Don't want turbocharged models", "Are there any suitable hybrid models") from standardized processed data and explicit demand description fragments in historical consultation records ("Consulting about the 2020 Honda CR-V", "The price of 160,000 RMB is too high"), and combining the contextual information of the customer's current consultation session (initial demand is a family SUV, supplemented by rejection of turbocharged engines, and subsequent interest in hybrid models), a set of explicit demand original texts with contextual tags is obtained (tagged "family scenario - rejection of turbocharged engines - interest in hybrid models"). The text set was dynamically segmented. Based on a new word discovery algorithm in the vehicle domain (by analyzing vehicle source data and customer consultation texts over the past year to update the new word library in areas such as "hybrid models" and "turbocharged"), the segmentation granularity was adjusted in real time. "SUV around 100,000 yuan" was segmented into "around 100,000 yuan" and "SUV", and "don't want turbocharged models" was segmented into "don't want", "turbocharged", and "model". Semantic dependency analysis was used to identify the causal relationship between "bringing elderly people and children" and "large space" and the relationship was preserved. Stop words in areas such as "of" and "whether" were filtered out, and the core semantic components of the demand and contextual words were preserved to obtain context-enhanced purified text fragments ("around 100,000 yuan", "SUV", "large space", "family use", "vehicle age not exceeding 3 years", "don't want turbocharged", "hybrid models"). The system invokes a dynamic demand keyword dictionary that integrates customer behavior feedback (including keywords for demand dimensions such as "price range", "vehicle model", "vehicle age", and "power type", and associates them with customer history clicks and transaction feedback weights) to perform bidirectional semantic matching on the purified text fragments: "around 100,000" matches the keyword "price range" (matching degree 95%), "SUV" matches the keyword "vehicle model" (matching degree 100%), "vehicle age not exceeding 3 years" matches the keyword "vehicle age" (matching degree 98%), and "turbocharged" and "hybrid vehicle" match the keyword "power type" (matching degree 96%). The keyword matching results and context association weight set are obtained ("family use" and "large space" are associated with a weight of 0.9, and "don't want turbocharged" and "hybrid vehicle" are associated with a weight of 0.85). Based on the matching results and weight set, the core demand dimensions (price range, vehicle type, vehicle age, power type) and derived demand dimensions (space) are identified. The fuzzy constraint descriptions corresponding to each dimension are extracted and semantically quantified: the price range "around 100,000" is quantified as "80,000-120,000", the vehicle age "no more than 3 years" is quantified as "≤3 years", the power type "no turbocharged" and "hybrid vehicle" are quantified as "hybrid" and "naturally aspirated", and the space "large" is quantified as "rear legroom ≥80cm", thus obtaining a demand dimension-quantified constraint mapping table.Based on this mapping table and combined with the vehicle configuration attribute association graph ("SUV" is associated with "space" and "family" attributes, "hybrid" is associated with "power type" attribute), demand value information and associated demand attributes are extracted from the purified text fragments: price range "80,000-120,000" is associated with "family budget" attribute, vehicle type "SUV" is associated with "space demand" attribute, vehicle age "≤3 years" is associated with "vehicle condition" attribute, and power type "hybrid / naturally aspirated" is associated with "driving experience" attribute, resulting in a preliminary set of demand tuples with associated attributes ((price range, 80,000-120,000, family budget), (vehicle type, SUV, space demand), (vehicle age, ≤3 years, vehicle condition), (power type, hybrid / naturally aspirated, driving experience), (space, ≥80cm, family scenario)). The domain rule engine, which integrates logical reasoning and scenario adaptation, is invoked to perform conflict detection (no conflict), redundancy removal (no redundancy), and intelligent completion of missing related requirement dimensions on the initial requirement tuple set (based on the family SUV scenario, the requirement dimension of "mileage ≤ 50,000 km" is completed), resulting in a validated set of valid requirement tuples. This set of valid tuples is mapped to a preset dynamic explicit requirement parameter system, assigning each tuple a unique parameter identifier (price range: P001, vehicle model: V002, vehicle age: Y003, power type: P004, space: S005, mileage: M006) and a scenario adaptation coefficient (family scenario adaptation coefficient 0.95). The requirement values ​​are then converted into structured parameter values ​​(80,000-120,000 RMB, SUV, ≤ 3 years, hybrid / naturally aspirated, ≥ 80cm, ≤ 50,000 km) and associated with parameter confidence scores (calculated based on keyword matching strength and context weight, with a price range confidence score of 92%, vehicle model confidence score of 98%, and power type confidence score of 90%), resulting in a set of confidence-enhanced structured requirement key-value pairs. Based on the customer's historical explicit demand preference model (combined with the customer's past consultation records of Japanese brands), the parameters of this set are optimized, and the parameter "Brand Preference: Japanese" is added. This is integrated to form explicit demand parameters containing parameter labels, structured parameter values, scenario adaptation coefficients and confidence levels (such as (P001, 80,000-120,000, 0.95, 92%), (V002, SUV, 0.95, 98%), (Y003, ≤3 years, 0.95, 95%), (P004, Hybrid / Naturally Aspirated, 0.95, 90%), (S005, ≥80cm, 0.95, 88%), (M006, ≤50,000 km, 0.95, 85%), (B007, Japanese, 0.95, 80%).

[0041] Furthermore, the process of identifying the core and derived needs explicitly stated by the customer based on the keyword matching results and contextual weight set includes the following steps: Data deconstruction is performed on keyword matching results and context association weight sets to extract semantic attributes, context association objects, and association strength information of matched keywords, generating a keyword semantic-association feature dataset; Based on the keyword semantic-association feature dataset, a semantic similarity algorithm is used to associate and aggregate semantically similar keywords to generate multiple semantically associated keyword clusters. Each cluster corresponds to a set of potential associated demand expressions. Extract the intra-cluster semantic identifiers and inter-cluster semantic distance parameters of each semantically related keyword cluster. Combine the knowledge graph of vehicle domain requirements dimensions to perform dimension mapping and adaptation on each semantically related keyword cluster to obtain a preliminary dimension candidate set. Call the requirement dimension semantic verification model to perform bidirectional semantic verification between each dimension in the initial dimension candidate set and the keyword semantic-association feature dataset, and generate dimension semantic fit parameters and fit confidence basis; Based on the semantic fit parameters and fit confidence criteria, a dimension validity screening algorithm is used to remove invalid dimensions whose semantic fit does not meet the verification criteria, thus obtaining a set of valid requirement dimensions. For each dimension in the set of effective demand dimensions, the demand dimension logical association model is invoked to analyze the semantic implication relationship, causal inference relationship and scenario association relationship between dimensions, and generate a dimension logical association graph. Based on the dimensional logical association graph, identify dimensional nodes with semantic radiation effect in the graph, and determine the core requirement dimension and the dominant parameters of the dimension; Based on the dominant parameters of dimensions and the logical relationship graph of dimensions, we can mine the related dimensions formed by the semantic extension and logical deduction of the core requirement dimensions, and determine the derived requirement dimensions.

[0042] In this embodiment of the invention, targeting the needs of customers in the used car transaction scenario for "SUVs around 100,000 yuan, spacious enough to carry elderly people and children, no more than 3 years old, no turbocharged, and interested in hybrid models", the keyword matching results (price range 95%, model 100%, vehicle age 98%, power type 96%) and context association weight set (family use - spacious 0.9, no turbocharged - hybrid 0.85) are deconstructed to extract the semantic attributes of the matching keywords ("around 100,000 yuan" is a price attribute, "SUV" is a model attribute, "3 years" is a time attribute, and "hybrid" is a power attribute), context association objects ("family use" is associated with "spacious", "turbocharged" is associated with "hybrid") and association strength information (0.9, 0.85), generating a keyword semantic-association feature dataset. Based on this dataset, the cosine semantic similarity algorithm (with a similarity threshold of 0.8) is used to associate and aggregate semantically similar keywords: "around 100,000" and "80,000-120,000" are aggregated into a price cluster, "SUV" is aggregated into a vehicle type cluster, "vehicle age not exceeding 3 years" is aggregated into a vehicle age cluster, "turbocharged", "hybrid" and "naturally aspirated" are aggregated into a power cluster, and "large space" and "rear legroom ≥ 80cm" are aggregated into a space cluster, generating 5 semantically associated keyword clusters, each cluster corresponding to a set of demand expressions. Extract the intra-cluster semantic identifiers (price, vehicle model, vehicle age, power, space) and inter-cluster semantic distance parameters (price-vehicle distance 0.6, vehicle model-space distance 0.3, power-vehicle distance 0.5) of each cluster. Combine this with the vehicle domain demand dimension knowledge graph (including 20 dimensions such as price, vehicle model, vehicle age, power, space, brand, mileage, etc. and their relationships) to perform dimension mapping adaptation for each cluster: price cluster maps to the "price range" dimension, vehicle model cluster maps to the "vehicle type" dimension, vehicle age cluster maps to the "vehicle age range" dimension, power cluster maps to the "power type" dimension, and space cluster maps to the "space size" dimension to obtain a preliminary set of dimension candidates. A semantic verification model for demand dimensions (trained based on 500,000 historical customer demand data points) was invoked to perform bidirectional verification between each dimension in the initial candidate set and the keyword semantic-association feature dataset: the "price range" dimension showed a 94% semantic fit with "around 100,000" (confidence basis: the customer explicitly mentioned the budget), the "vehicle type" dimension showed a 99% fit with "SUV" (confidence basis: the vehicle type was directly specified), and the "space size" dimension showed a 92% fit with "carrying elderly people and children" (confidence basis: scenario association). Dimension semantic fit parameters and fit confidence basis were generated. Based on the fit parameters (with a threshold of 85%), a dimension validity screening algorithm was used to remove invalid dimensions such as "interior material" that had no matching keywords, resulting in a set of valid demand dimensions (price range, vehicle type, vehicle age range, power type, space size).The logical association model of demand dimensions is used to analyze the relationships between dimensions: vehicle type (SUV) implies space size (requiring large space), which is a semantic implication relationship; family scenario (carrying elderly and children) derives space requirement as a causal relationship; power type and vehicle type are related by scenario, generating a logical association graph of dimensions. Based on the graph, dimension nodes with semantic radiation effect are identified (vehicle type SUV radiates space and power dimensions, price range radiates vehicle dimension), determining the core demand dimensions as price range, vehicle type, and power type, with the dimension dominance parameters (SUV dominance 0.9, price 0.85, power 0.8). Based on the dominance parameters and the graph, related dimensions extended from the core dimensions are mined: "space size" is extended from vehicle type SUV, "fuel consumption level" from power type, and "configuration level" from price range, determining the derived demand dimensions as space size, fuel consumption level, and configuration level.

[0043] Furthermore, the extraction of implicit requirement parameters based on standardized data processing includes the following steps: By filtering scenario description texts, behavioral tendency records, and emotional expression content from customer interaction data in standardized data processing, and combining them with the vehicle usage scenario classification system, a raw dataset of scenario-related implicit needs is generated. Cross-modal semantic parsing is performed on the original dataset of scene-related implicit needs. Textual semantic features and scene attribute features are integrated to extract information on the user, behavioral intention, scene constraints and sentiment, and generate a cross-modal demand semantic feature set. Input the cross-modal demand semantic feature set into the scene intent recognition model trained based on contrastive learning, mine the potential use scenarios not directly expressed in the text, and generate scene intent labels and scene matching criteria; Based on scene intent tags and scene matching criteria, the dynamic scene-demand dimension mapping library is invoked, and the core demand dimension corresponding to the scene is matched through semantic reasoning algorithm to generate a scene-demand dimension association list. By combining customer historical consultation behavior sequences and vehicle source preference feedback data, we analyze the potential preference tendencies of each dimension in the scenario-demand dimension association list, and generate demand dimension-preference tendency association parameters through sequence pattern mining algorithms. Based on the scenario-demand dimension association list and the demand dimension-preference tendency association parameters, an implicit demand semantic vector is constructed. The demand dimension features corresponding to the core scenario are strengthened through the scenario attention mechanism to generate a scenario-enhanced implicit demand vector. The scenario-enhanced implicit demand vector is input into the generative parameter transformation model, which converts the vector features into structured demand parameters and generates a candidate set of implicit demand parameters. The requirement parameter conflict resolution model is invoked, and the implicit requirement parameter candidate set is semantically consistent with the explicit requirement parameters. Conflicting parameters are eliminated and derived parameters are completed to generate implicit requirement parameters containing dimension identifiers, parameter values, and scenario adaptability.

[0044] In this embodiment of the invention, by filtering scenario description text ("traveling with elderly people and children", "family use"), behavioral tendency records (inquiries about Honda CR-V and Nissan X-Trail, rejection of high quotes) and emotional expression content (positive tone when inquiring about hybrid models) from standardized processed data, and combining this with a vehicle usage scenario classification system (10 scenarios and features including family commuting, long-distance self-driving, and business reception), a scenario-related implicit demand raw dataset (family use scenario label, 7-seat preference, low fuel consumption focus) is generated. Cross-modal semantic parsing (integrating text NLP and scenario feature recognition) is performed on this dataset to extract information on the user (multi-person family use), behavioral intent (safe and comfortable travel), scenario constraints (large number of passengers), and emotional tendency (preference for energy saving), generating a cross-modal demand semantic feature set (user: 3-5 people, intent: safety and comfort, constraints: ≥5 seats, tendency: low fuel consumption). The feature set is input into a scene intent recognition model trained based on contrastive learning (training data contains 1 million scene-intent correspondence data) to mine potential use cases: "traveling with elderly and children" is associated with the "family long-distance travel" scenario (scenario matching criteria: elderly and children need space and comfort when traveling, matching degree 91%), generating the scene intent tag "family long-distance travel" and matching criteria. Based on this tag, a dynamic scene-demand dimension mapping library is called (storing the association rules between scenes and dimensions, such as family long-distance travel → space, safety, fuel consumption), and the core demand dimensions corresponding to the scene are matched through a positive semantic reasoning algorithm: "space size", "safety configuration", "fuel consumption level", "number of seats", generating a scene-demand dimension association list. Combining the customer's historical consultation behavior sequence (inquiries about CR-V and X-Trail in October 2024) and vehicle source preference feedback (rejection of high quotes of 160,000 RMB), we analyzed the potential preference tendencies in each dimension of the list: Space dimension favors "large space" (based on inquiries about SUVs), Safety dimension favors "high safety configuration" (based on family use scenarios), and Fuel consumption favors "low fuel consumption" (based on interest in hybrid vehicles). Using the FP-Growth sequence pattern mining algorithm (minimum support 0.7), we generated correlation parameters between demand dimensions and preference tendencies (Space - Large 0.88, Safety - High 0.82, Fuel Consumption - Low 0.9). Based on the correlated list and parameters, we constructed an implicit demand semantic vector (Space 0.88, Safety 0.82, Fuel Consumption 0.9, Seats 0.75). Through a scenario attention mechanism (family long-distance scenario weight 0.9), we strengthened the demand dimension features corresponding to core scenarios, generating scenario-enhanced implicit demand vectors (Space 0.92, Safety 0.89, Fuel Consumption 0.95, Seats 0.81). The vector is input into a generative parameter transformation model (trained based on the Transformer architecture) and converted into structured requirement parameters: space "rear legroom ≥ 85cm", safety "equipped with ESP + 6 airbags", fuel consumption "fuel consumption ≤ 6L per 100km", and seating "5 seats or more", generating a candidate set of implicit requirement parameters.The demand parameter conflict resolution model is invoked, and the explicit demand parameters (price 80,000-120,000 yuan, vehicle age ≤3 years) are used to verify that "fuel consumption ≤6L per 100km" does not conflict with hybrid power and "5 seats and above" is compatible with SUV models. "6 seats" is removed (most SUVs under 120,000 yuan are 5 seats) and the "tire pressure monitoring" safety configuration parameters are added. Implicit demand parameters are generated with dimension labels (safety: S001, fuel consumption: F002, seats: Se003), parameter values ​​(ESP+6 airbags, ≤6L, ≥5 seats), and scenario adaptability (0.89, 0.95, 0.81).

[0045] Furthermore, the extraction of multi-dimensional vehicle attribute features based on the standardized processed data through the vehicle source profile module of profile construction layer 20 includes the following steps: From the standardized data subset related to vehicle sources, we separate vehicle basic attribute data, vehicle condition inspection data, maintenance history data and market circulation data to generate a multi-source original dataset of vehicle sources. Data association processing is performed on the original datasets of vehicles from multiple sources to match and integrate data from different sources using unique vehicle identifiers, generating an integrated dataset of vehicle sources with association identifiers; Based on the vehicle source integration dataset, basic vehicle attribute information is extracted and converted into structured basic attribute parameters through a vehicle configuration coding parsing algorithm, generating a set of basic vehicle attribute parameters. Feature mining is performed on the vehicle condition detection data in the vehicle source integration dataset to analyze the correlation between detection indicators and potential faults, and to generate vehicle health assessment parameters, which include dimensional parameters corresponding to the integrity of the vehicle body structure and the stability of the power system. The maintenance frequency, key component replacement cycle, and maintenance compliance information are extracted from historical maintenance data to generate maintenance feature parameters, which reflect the vehicle's usage and maintenance status. By combining market circulation data from the vehicle source integration dataset with market price data for the same vehicle model, vehicle residual value and market circulation activity information are calculated, and market attribute parameters are generated. By calling the vehicle attribute association graph, semantic association is performed on the vehicle source basic attribute parameter set, vehicle condition health assessment parameters, maintenance characteristic parameters and market attribute parameters, and potential association relationships between each parameter are explored to generate the vehicle source attribute association parameter set. By performing feature filtering and fusion on the set of vehicle source attribute associated parameters, redundant parameters are eliminated and the core attribute representation is strengthened to obtain multi-dimensional attribute features of vehicles.

[0046] In this embodiment of the invention, for a 2022 Toyota RAV4 in a used car transaction scenario, the following data is separated from the standardized data subset related to the vehicle: basic vehicle attribute data (model: Toyota RAV4 2022 2.5L hybrid two-wheel drive version, registration date: October 2021, body color: white, wheelbase: 2690mm, seat material: fabric, guide price: 225,800 yuan, asking price: 148,000 yuan) and vehicle condition inspection data (body structure score: 92 points, engine condition score: 88 points). The system analyzes and integrates the following data: transmission condition score (90 points), brake system test (normal), steering system test (normal), fault code record (1 cleared air conditioning fault code), maintenance history data (first maintenance date: October 2022, maintenance frequency in the past 3 years: 5 times, replaced parts: air conditioning filter 1 time, brake pads 1 time, major repair record: none), and market circulation data (sales volume of the same model in the past 6 months: 120 units, average transaction price of the same configuration model: 152,000 yuan, current vehicle listing duration: 15 days). This generates a multi-source vehicle source dataset. The dataset is then processed by association, using the vehicle's unique identifier "LNMBE84C8M6001234" as the matching keyword. The "model" and "registration date" from the basic attribute data are matched and integrated with the "body structure score" from the vehicle condition test data, the "maintenance frequency" from the maintenance data, and the "average transaction price" from the market circulation data. This ensures consistent data association across different sources for the same vehicle source, generating an integrated vehicle source dataset with association identifiers. Based on this integrated dataset, basic vehicle attribute information is extracted. A vehicle configuration coding parsing algorithm (parses the code for "2.5L hybrid two-wheel drive version," mapping it to "displacement 2.5L," "power type hybrid," and "drive mode two-wheel drive") is used to convert this information into structured basic attribute parameters (displacement: 2.5L, power type: hybrid, drive mode: two-wheel drive, wheelbase: 2690mm, seat material: fabric, price: 148,000 RMB), generating a set of basic vehicle attribute parameters. Feature mining is performed on the vehicle condition detection data in the integrated dataset, and a fault correlation analysis algorithm is used to analyze the relationship between detection indicators and potential faults: a body structure score of 92 corresponds to a "body structure integrity" dimension parameter of 0.92 (out of 1.0, higher scores indicate better integrity); an engine condition score of 88, a transmission score of 90, and no major fault records correspond to a "powertrain stability" dimension parameter of 0.89; and normal braking / steering systems correspond to a "handling system safety" parameter of 0.95, generating vehicle health assessment parameters. Based on historical maintenance data, the following information was extracted: maintenance frequency (an average of 1.67 maintenance times per year, higher than the average of 1.2 times for vehicles of the same age), replacement cycle of key components (air conditioning filter every 2 years / time, brake pads every 3 years / time, in line with the original manufacturer's recommended cycle), and maintenance compliance (all 5 maintenance services were completed at authorized 4S stores, with a compliance rate of 100%). Maintenance characteristic parameters were then generated (maintenance frequency coefficient 1.39, component replacement compliance rate 1.0, and maintenance institution compliance rate 1.0).Combining market circulation data and market conditions data for the same model (residual value of the same configuration model in the past 6 months: 75% in the first year, 70% in the second year, and 65% in the third year), the residual value of the current vehicle (registered in October 2021, residual value of 65% in the third year, current price of 148,000 yuan / guide price of 225,800 yuan = 65.5%, residual value parameter 0.655) and market circulation activity (120 transactions of the same model in the past 6 months, 28 inquiries within 15 days of listing, activity parameter 0.82) are used to generate market attribute parameters. The system invokes a vehicle attribute association graph (including association rules between basic attributes, vehicle condition, maintenance, and market attributes, such as the association between hybrid power and low fuel consumption and high residual value). It semantically associates the vehicle source basic attribute parameter set, vehicle condition health assessment parameters, maintenance characteristic parameters, and market attribute parameters to uncover potential associations between "hybrid power type" and "low fuel consumption" and "high residual value," as well as positive associations between "maintenance compliance" and "powertrain stability," generating a vehicle source attribute association parameter set. This parameter set undergoes feature filtering and fusion. An analysis of variance algorithm is used to remove redundant parameters such as "body color" (with a variance of 0.12 on customer demand, below the threshold of 0.2), strengthening the representation of core attributes such as "power type," "vehicle condition health," and "residual value," resulting in multi-dimensional vehicle attribute features (power type: hybrid 0.9, price: 148,000 0.85, vehicle condition health: 0.91, maintenance characteristics: 0.96, residual value: 0.655, space: wheelbase 2690mm corresponding to 0.88).

[0047] Furthermore, the step of inputting the customer feature vector and vehicle source feature vector into the recommendation calculation layer 30 to calculate the demand fit between each vehicle source feature vector and the customer feature vector includes the following steps: Extract the core demand dimension identifiers and feature distribution features from the customer feature vector, and simultaneously extract the core attribute dimension identifiers and attribute distribution features from the vehicle source feature vector to generate a supply and demand dimension feature comparison set. Based on the supply and demand dimension feature comparison set, a mapping relationship between the customer demand dimension and the vehicle source attribute dimension is established through semantic association reasoning, generating a dimension mapping association table and mapping confidence parameters; Based on the dimension mapping association table, the customer feature vector and the vehicle source feature vector are dimension aligned, retaining the effective dimensions that are mapped to each other and eliminating the redundant dimensions that are not related, and generating supply and demand aligned vector pairs. Perform feature distribution difference analysis on the supply and demand alignment vector pairs, calculate the distribution fit parameters of customer demand features and vehicle source attribute features under each alignment dimension, and generate a dimension fit matrix. The requirement-attribute association strength model is invoked, and the dimension mapping confidence parameters and dimension fit matrix are combined to explore the potential association logic between dimensions and generate dimension association logic coefficients. Calculate the overall correlation tightness parameter of the supply and demand alignment vector pairs based on the dimensional fit matrix and dimensional correlation logical coefficient, and generate the vector correlation strength value. By combining the priority sequence of customer needs and the importance sequence of vehicle source attributes, the vector association strength value is modified for scenario adaptation, and scenario adaptation association parameters are generated. Based on the scenario adaptation and correlation parameters, the dimension fit matrix, dimension correlation logic coefficient and vector correlation strength value are integrated to generate the demand fit between each vehicle source feature vector and the customer feature vector.

[0048] In this embodiment of the invention, by extracting the core demand dimension identifiers (price, model, power, vehicle age, space, safety, fuel consumption) and feature distribution characteristics (price concentrated in the 80,000-120,000 range, power preference for hybrid) and feature distribution characteristics (price concentrated in the 80,000-120,000 range, power preference for hybrid) from the customer feature vector (generated based on the aforementioned requirements, core dimensions: price range 80,000-120,000 0.9, vehicle type SUV 1.0, power type hybrid 0.95, vehicle age ≤ 3 years 0.88, space size ≥ 80cm 0.92, safety configuration 0.89, fuel consumption ≤ 6L 0.95) from the customer feature vector, the invention simultaneously extracts... Take the core attribute dimension identifiers (price, model, power, vehicle age, space, safety, fuel consumption, residual value) and attribute distribution features (price 148,000, hybrid power, 3-year vehicle age) from the Toyota RAV4 vehicle source feature vector (core dimensions: price 148,000 0.7, vehicle type SUV 1.0, hybrid power 0.9, vehicle age 3 years 0.85, space 2690mm 0.88, safety configuration 0.92, fuel consumption 5.9L 0.96, residual value 0.655) to generate a supply and demand dimension feature comparison set. Based on this reference set, a mapping relationship between customer demand dimension and vehicle source attribute dimension is established through bidirectional semantic association reasoning: customer "price range 80,000-120,000" maps to vehicle source "price 148,000", "SUV" maps to "SUV", "hybrid" maps to "hybrid", "vehicle age ≤ 3 years" maps to "vehicle age 3 years", "space ≥ 80cm" maps to "space 2690mm", "safety configuration" maps to "safety configuration", and "fuel consumption ≤ 6L" maps to "fuel consumption 5.9L". A dimension mapping association table and mapping confidence parameters are generated (vehicle type 1.0, power 0.98, fuel consumption 0.97, vehicle age 0.95, space 0.93, safety 0.92, price 0.85). Based on the dimension mapping association table, the customer feature vector and the vehicle source feature vector are aligned in dimensions. Seven effective dimensions that are mutually mapped are retained: price, model, power, vehicle age, space, safety, and fuel consumption. Redundant dimensions such as the vehicle source "residual value" (which has no corresponding customer demand dimension) are removed. Supply and demand alignment vector pairs are generated (customer vector: [0.9,1.0,0.95,0.88,0.92,0.89,0.95]; vehicle source vector: [0.7,1.0,0.9,0.85,0.88,0.92,0.96]). A feature distribution difference analysis was performed on the supply and demand alignment vector pairs. Euclidean distance was used to calculate the distribution fit parameters under each alignment dimension: vehicle model dimension customer 1.0 / vehicle source 1.0 → fit 1.0; power dimension 0.95 / 0.9 → 0.947; fuel consumption 0.95 / 0.96 → 0.989; vehicle age 0.88 / 0.85 → 0.966; space 0.92 / 0.88 → 0.956; safety 0.89 / 0.92 → 0.967; price 0.9 / 0.7 → 0.778, generating a dimension fit matrix [0.778, 1.0, 0.947, 0.966, 0.956, 0.967, 0.989].The demand-attribute association strength model (trained based on 500,000 historical transaction data points, outputting inter-dimensional association strength coefficients) is invoked. Combined with the dimension mapping confidence parameter and the dimension fit matrix, potential association logic between dimensions such as "power-fuel consumption" (association strength 0.92) and "space-vehicle type" (0.88) is mined, generating dimension association logic coefficients [0.85, 1.0, 0.92, 0.88, 0.88, 0.9, 0.92]. Based on the dimension fit matrix and dimension association logic coefficients, a weighted summation algorithm is used to calculate the overall association tightness parameter of the supply-demand alignment vector pairs: (0.778×0.85+1.0×1.0+0.947×0.92+0.966×0.88+0.956×0.88+0.967×0.9+0.989×0.92) / 7=0.912, generating a vector association strength value of 0.912. Combining the customer demand priority sequence (price > model > power > fuel consumption > vehicle age > space > safety) and the vehicle source attribute importance sequence (model > power > vehicle condition > price > fuel consumption > space > safety), a scenario adaptation correction algorithm is used to correct the vector association strength value: the price dimension has a low fit (0.778) but a high customer priority, so the correction coefficient is 0.98; the model and power dimensions have high fit and high priority, so the correction coefficient is 1.02; the final corrected scenario adaptation association parameter is 0.912 × (0.98 + 1.02 + 1.0 + 1.0 + 0.99 + 0.99 + 0.99) / 7 = 0.915. Based on this parameter, the dimension fit matrix, dimension association logical coefficient, and vector association strength value are integrated to generate a demand fit of 0.915 (out of 1.0, the higher the score, the higher the fit).

[0049] Furthermore, Embodiment 2 of the present invention also provides a vehicle recommendation information processing system, such as... Figure 2As shown, a vehicle recommendation information processing system is used to perform the vehicle recommendation information processing method described above. This system includes a data acquisition layer 10, a profile construction layer 20, a recommendation calculation layer 30, and a result output layer 40. Each layer interacts through a data interface. The data acquisition layer 10 collects full-source data through a multi-source interface. This full-source data includes private message interaction data, vehicle source data, and customer historical data. The full-source data is structured to obtain standardized processed data, which is then stored in a distributed database. The profile construction layer 20 calls the standardized processed data from the distributed database and generates customer feature vectors and vehicle source feature vectors through explicit demand extraction, implicit demand mining, and vehicle source feature quantification. The recommendation calculation layer 30 calculates the demand fit between each vehicle source feature vector and the customer feature vector based on the customer feature vector and the vehicle source feature vector. The result output layer 40 generates a vehicle recommendation list containing reasons for fit based on the demand fit, pushes it to the car dealer terminal, records customer feedback data, and feeds it back to the recommendation calculation layer 30 for iterative optimization.

[0050] In the embodiments of the present invention, the processing system for vehicle recommendation information is applied to a used car trading platform, which includes a data collection layer 10, a portrait construction layer 20, a recommendation calculation layer 30, and a result output layer 40. Each layer realizes real-time interaction through RESTful data interfaces. The data collection layer 10 deploys a multi-source interface integration module, which calls the Douyin open platform interface and the WeChat instant messaging SDK to collect private message interaction data (such as customer demand texts like "SUV around 100,000 yuan, hybrid, for family use" and interaction time sequences), calls the self-developed vehicle condition detection system interface, the third-party maintenance query platform interface, and the vehicle source interface of the car dealer SaaS system to collect vehicle source data (such as the basic attributes, vehicle condition detection, and maintenance records of Toyota RAV4), and calls the customer data interface of the car dealer SaaS system to collect customer historical data (such as customer's past inquiries about CR-V and records of rejecting high quotes); performs structured processing on the collected full-volume source data: converts private message texts into parsable strings through NLP, and removes invalid characters such as "de" and "ne"; normalizes the vehicle condition detection scores into numerical values in the range of 0-1; uniformly converts prices into the unit of "ten thousand yuan"; after processing, obtains standardized processed data, stores it in the HBase distributed database, and adopts the row key format of "scenario identifier - data period - vehicle source ID / customer ID" to ensure data retrieval efficiency. The portrait construction layer 20 deploys a customer portrait module and a vehicle source portrait module, and calls the standardized processed data in the distributed database through the JDBC interface: the customer portrait module extracts explicit demands (obtains parameters such as price of 80,000 - 120,000 yuan and vehicle type SUV) from the private message interaction data, mines implicit demands (extracts demands such as "long-distance family travel" and "high safety configuration" based on the fine-tuned BERT model), combines customer historical data to fuse preferences (supplements Japanese brand preferences), and quantifies them into a 50-dimensional customer feature vector (such as price 0.9, vehicle type 1.0, power 0.95); the vehicle source portrait module quantifies the basic attributes of vehicle source data (power hybrid 0.9, price 148,000 yuan 0.7), quantifies the quality attributes (vehicle condition health 0.91), and integrates them into a 50-dimensional vehicle source feature vector (matched with the customer vector dimension), and the two types of vectors are respectively stored in the Redis cache and the vehicle source feature library. The recommendation calculation layer 30 deploys a DNN model service and a matching calculation module, and loads the trained DNN model (the input layer receives a 100-dimensional concatenated vector, each of the 3 hidden layers has 128 neurons, and the output layer outputs a matching degree of 0-1); when a customer sends a new demand, it calls the customer feature vector in Redis and the vehicle source feature vector in the vehicle source feature library in real time, and inputs them into the DNN model to calculate the demand fit degree (such as the fit degree of Toyota RAV4 is 0.915, and the fit degree of Honda CR-V 2020 model is 0.89); if the customer supplements the demand of "no fabric seats", it triggers the portrait construction layer 20 to update the customer feature vector (supplements the seat material parameter of 0.1) again, and recalculates the fit degree.The output layer 40 deploys a sorting output module, a multi-terminal push module, and a log recording module: The sorting output module sorts vehicles from high to low according to their suitability for demand, selects the top 5 vehicles to generate a recommendation list, including basic vehicle information (model, price, mileage), quality information (vehicle condition rating 0.91, maintenance completeness 0.96) and the rationale for suitability ("Matches your hybrid needs, fuel consumption of 5.9L per 100km meets the ≤6L requirement, and the 3-year age is close to your ≤3-year expectation"); The multi-terminal push module pushes the recommendation list to the car dealer's SaaS system private message interface (for car dealers to copy and send) and the car dealer's mobile APP (real-time message notification) via API interface; The log recording module records the recommendation list and subsequent customer feedback (such as clicking on Toyota RAV4, inquiring about interior materials), and stores it in the log database in the format of "recommendation time-customer ID-vehicle ID-feedback result". Every day at midnight, the log data is synchronized to the training dataset of the recommendation calculation layer 30 for DNN model iterative optimization (retrained once a month to ensure matching accuracy ≥85%).

[0051] Furthermore, Embodiment 3 of the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the vehicle recommendation information processing method described above.

[0052] Furthermore, Embodiment 4 of the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the vehicle recommendation information processing method described above.

[0053] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0054] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A processing method of vehicle recommendation information, characterized by, Includes the following steps: In response to customer demand response triggering operations initiated by the car dealer terminal, the data collection layer calls a preset set of interfaces to collect all the source data required to build customer profiles and vehicle source profiles. The full source data includes private message interaction data between car dealers and customers, vehicle source data related to the entire life cycle of vehicles, and customer historical data related to past vehicle inquiries and transactions. The full source data is structured, the unstructured data is cleaned and format converted, and the structured data is normalized to a unified dimension to obtain standardized processed data. Based on the standardized processing data, the customer profile module and vehicle source profile module of the profile construction layer are used to extract multi-dimensional customer demand features and multi-dimensional vehicle attribute features, respectively. The multi-dimensional customer demand features include explicit demand parameters and implicit demand parameters, and dimension-matched customer feature vectors and vehicle source feature vectors are generated. The customer feature vector and the vehicle source feature vector are input into the recommendation calculation layer to calculate the demand fit between each vehicle source feature vector and the customer feature vector. All vehicle listings are prioritized based on the required matching degree, and a vehicle recommendation list containing core vehicle listing information and the reasons for the required matching degree is generated. The result output layer pushes the vehicle recommendation list to the corresponding car dealer terminal and records the customer's subsequent interaction feedback data on the recommendation list for iterative optimization of the recommendation calculation layer.

2. The processing method of vehicle recommendation information according to claim 1, characterized in that, The process of collecting full source data required to construct customer profiles and vehicle source profiles by calling a preset set of interfaces through the data acquisition layer includes the following steps: The system calls the open interfaces of social platforms and instant messaging SDKs to collect private message interaction data in real time. The private message interaction data includes customer demand description text data and interaction time sequence correlation data. The system calls the vehicle condition detection system interface, the maintenance record query platform interface, and the vehicle source interface of the car dealer SaaS system to integrate and collect vehicle source data, which includes vehicle technical condition detection data, maintenance history data, and basic vehicle attribute data. Call the customer data interface of the car dealer SaaS system to collect customer historical data, which includes past inquiry records of car models, price feedback records and attribute data of historically sold car models. Based on the data source types of the private message interaction data, vehicle source data, and customer historical data, corresponding collection priorities and data update frequencies are configured for each type of data, and the collection and synchronization of all source data are completed according to the collection priorities and data update frequencies.

3. The processing method of vehicle recommendation information according to claim 1, characterized in that, The extraction of explicit requirement parameters based on standardized data processing includes the following steps: Standardized private message text data and explicit demand description fragments from historical consultation records are filtered out from the standardized data. Combined with the contextual information of the customer's current consultation session, a raw text set of explicit demands with context tags is obtained. The original text set of explicit requirements with context tags is dynamically segmented. Specifically, based on the new word discovery algorithm in the vehicle domain, the segmentation granularity, semantic dependency analysis and domain stop word filtering are adjusted in real time to retain the core semantic components of the requirements and context-related words, so as to obtain context-enhanced purified text fragments. The dynamic demand keyword dictionary that integrates customer behavior feedback is invoked to perform bidirectional semantic matching on the context-enhanced clean text fragment to obtain keyword matching results and context association weight set; Based on the keyword matching results and contextual association weight set, identify the core and derived demand dimensions explicitly stated by the customer, extract the fuzzy constraint descriptions corresponding to each demand dimension and perform semantic quantification to obtain a demand dimension-quantified constraint mapping table. Based on the aforementioned demand dimension-quantification constraint mapping table and combined with the vehicle configuration attribute association graph, demand value information and associated demand attributes are extracted from the context-enhanced clean text fragments to obtain a preliminary set of demand tuples with associated attributes. The domain rule engine, which integrates logical reasoning and scenario adaptation, is invoked to perform conflict detection, redundancy removal, and intelligent completion of missing related requirement dimensions on the preliminary requirement tuple set with associated attributes, so as to obtain a valid requirement tuple set after verification. The verified set of valid demand tuples is mapped to a preset dynamic explicit demand parameter system. A unique parameter identifier and scenario adaptation coefficient are assigned to the demand dimension of each tuple. The demand value is converted into a structured parameter value and associated with the parameter confidence. The set of confidence-enhanced structured demand key-value pairs is calculated based on keyword matching strength and context association weight. Based on the customer's historical explicit demand preference model, the set of confidence-enhanced structured demand key-value pairs is optimized to form explicit demand parameters containing parameter identifiers, structured parameter values, scenario adaptation coefficients, and confidence levels.

4. The processing method of vehicle recommendation information according to claim 3, characterized in that, The process of identifying the core and derivative needs explicitly stated by the customer based on the keyword matching results and contextual weight set includes the following steps: Data deconstruction is performed on keyword matching results and context association weight sets to extract semantic attributes, context association objects, and association strength information of matched keywords, generating a keyword semantic-association feature dataset; Based on the keyword semantic-association feature dataset, a semantic similarity algorithm is used to associate and aggregate semantically similar keywords to generate multiple semantically associated keyword clusters. Each cluster corresponds to a set of potential associated demand expressions. Extract the intra-cluster semantic identifiers and inter-cluster semantic distance parameters of each semantically related keyword cluster. Combine the knowledge graph of vehicle domain requirements dimensions to perform dimension mapping and adaptation on each semantically related keyword cluster to obtain a preliminary dimension candidate set. Call the requirement dimension semantic verification model to perform bidirectional semantic verification between each dimension in the initial dimension candidate set and the keyword semantic-association feature dataset, and generate dimension semantic fit parameters and fit confidence basis; Based on the semantic fit parameters and fit confidence criteria, a dimension validity screening algorithm is used to remove invalid dimensions whose semantic fit does not meet the verification criteria, thus obtaining a set of valid requirement dimensions. For each dimension in the set of effective demand dimensions, the demand dimension logical association model is invoked to analyze the semantic implication relationship, causal inference relationship and scenario association relationship between dimensions, and generate a dimension logical association graph. Based on the dimensional logical association graph, identify dimensional nodes with semantic radiation effect in the graph, and determine the core requirement dimension and the dominant parameters of the dimension; Based on the dominant parameters of dimensions and the logical relationship graph of dimensions, we can mine the related dimensions formed by the semantic extension and logical deduction of the core requirement dimensions, and determine the derived requirement dimensions.

5. The processing method of vehicle recommendation information according to claim 3, characterized in that, The extraction of implicit requirement parameters based on standardized data processing includes the following steps: By filtering scenario description texts, behavioral tendency records, and emotional expression content from customer interaction data in standardized data processing, and combining them with the vehicle usage scenario classification system, a raw dataset of scenario-related implicit needs is generated. Cross-modal semantic parsing is performed on the original dataset of scene-related implicit needs. Textual semantic features and scene attribute features are integrated to extract information on the user, behavioral intention, scene constraints and sentiment, and generate a cross-modal demand semantic feature set. Input the cross-modal demand semantic feature set into the scene intent recognition model trained based on contrastive learning, mine the potential use scenarios not directly expressed in the text, and generate scene intent labels and scene matching criteria; Based on scene intent tags and scene matching criteria, the dynamic scene-demand dimension mapping library is invoked, and the core demand dimension corresponding to the scene is matched through semantic reasoning algorithm to generate a scene-demand dimension association list. By combining customer historical consultation behavior sequences and vehicle source preference feedback data, we analyze the potential preference tendencies of each dimension in the scenario-demand dimension association list, and generate demand dimension-preference tendency association parameters through sequence pattern mining algorithms. Based on the scenario-demand dimension association list and the demand dimension-preference tendency association parameters, an implicit demand semantic vector is constructed. The demand dimension features corresponding to the core scenario are strengthened through the scenario attention mechanism to generate a scenario-enhanced implicit demand vector. The scenario-enhanced implicit demand vector is input into the generative parameter transformation model, which converts the vector features into structured demand parameters and generates a candidate set of implicit demand parameters. The requirement parameter conflict resolution model is invoked, and the implicit requirement parameter candidate set is semantically consistent with the explicit requirement parameters. Conflicting parameters are eliminated and derived parameters are completed to generate implicit requirement parameters containing dimension identifiers, parameter values, and scenario adaptability.

6. The processing method of vehicle recommendation information according to claim 1, characterized in that, The process of extracting multi-dimensional vehicle attribute features from the vehicle source profiling module of the profiling construction layer based on the standardized processed data includes the following steps: From the standardized data subset related to vehicle sources, we separate vehicle basic attribute data, vehicle condition inspection data, maintenance history data and market circulation data to generate a multi-source original dataset of vehicle sources. Data association processing is performed on the original datasets of vehicles from multiple sources to match and integrate data from different sources using unique vehicle identifiers, generating an integrated dataset of vehicle sources with association identifiers; Based on the vehicle source integration dataset, basic vehicle attribute information is extracted and converted into structured basic attribute parameters through a vehicle configuration coding parsing algorithm, generating a set of basic vehicle attribute parameters. Feature mining is performed on the vehicle condition detection data in the vehicle source integration dataset to analyze the correlation between detection indicators and potential faults, and to generate vehicle health assessment parameters, which include dimensional parameters corresponding to the integrity of the vehicle body structure and the stability of the power system. The maintenance frequency, key component replacement cycle, and maintenance compliance information are extracted from historical maintenance data to generate maintenance feature parameters, which reflect the vehicle's usage and maintenance status. By combining market circulation data from the vehicle source integration dataset with market price data for the same vehicle model, vehicle residual value and market circulation activity information are calculated, and market attribute parameters are generated. By calling the vehicle attribute association graph, semantic association is performed on the vehicle source basic attribute parameter set, vehicle condition health assessment parameters, maintenance characteristic parameters and market attribute parameters, and potential association relationships between each parameter are explored to generate the vehicle source attribute association parameter set. By performing feature filtering and fusion on the set of vehicle source attribute associated parameters, redundant parameters are eliminated and the core attribute representation is strengthened to obtain multi-dimensional attribute features of vehicles.

7. The processing method of vehicle recommendation information according to claim 1, characterized in that, The step of inputting the customer feature vector and vehicle source feature vector into the recommendation calculation layer to calculate the demand fit between each vehicle source feature vector and the customer feature vector includes the following steps: Extract the core demand dimension identifiers and feature distribution features from the customer feature vector, and simultaneously extract the core attribute dimension identifiers and attribute distribution features from the vehicle source feature vector to generate a supply and demand dimension feature comparison set. Based on the supply and demand dimension feature comparison set, a mapping relationship between the customer demand dimension and the vehicle source attribute dimension is established through semantic association reasoning, generating a dimension mapping association table and mapping confidence parameters; Based on the dimension mapping association table, the customer feature vector and the vehicle source feature vector are dimension aligned, retaining the effective dimensions that are mapped to each other and eliminating the redundant dimensions that are not related, and generating supply and demand aligned vector pairs. Perform feature distribution difference analysis on the supply and demand alignment vector pairs, calculate the distribution fit parameters of customer demand features and vehicle source attribute features under each alignment dimension, and generate a dimension fit matrix. The requirement-attribute association strength model is invoked, and the dimension mapping confidence parameters and dimension fit matrix are combined to explore the potential association logic between dimensions and generate dimension association logic coefficients. Calculate the overall correlation tightness parameter of the supply and demand alignment vector pairs based on the dimensional fit matrix and dimensional correlation logical coefficient, and generate the vector correlation strength value. By combining the priority sequence of customer needs and the importance sequence of vehicle source attributes, the vector association strength value is modified for scenario adaptation, and scenario adaptation association parameters are generated. Based on the scenario adaptation and correlation parameters, the dimension fit matrix, dimension correlation logic coefficient and vector correlation strength value are integrated to generate the demand fit between each vehicle source feature vector and the customer feature vector.

8. A system for processing vehicle recommendation information, characterized in that, The system for processing vehicle recommendation information as described in claim 1 includes a data acquisition layer, a profile building layer, a recommendation calculation layer, and a result output layer. Each layer interacts through a data interface. The data acquisition layer collects full-source data through a multi-source interface. This full-source data includes private message interaction data, vehicle source data, and customer historical data. The full-source data is then structured to obtain standardized processed data, which is stored in a distributed database. The profile building layer calls the standardized processed data from the distributed database and generates customer feature vectors and vehicle source feature vectors through explicit demand extraction, implicit demand mining, and vehicle source feature quantification. The recommendation calculation layer calculates the demand fit between each vehicle source feature vector and the customer feature vector based on the customer feature vector and the vehicle source feature vector. The result output layer generates a vehicle recommendation list containing reasons for fit based on the demand fit, pushes it to the car dealer's terminal, records customer feedback data, and feeds it back to the recommendation calculation layer for iterative optimization.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for processing vehicle recommendation information as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for processing vehicle recommendation information as described in any one of claims 1-7.

Images