Vehicle-mounted non-vehicle product output method, device, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence and the field of financial science and technology, and discloses a vehicle-mounted non-vehicle product output method, which comprises the following steps: acquiring voice data of a user, converting the voice data into a text, collecting insurance product purchase history, personal information and risk assessment data of the user, and analyzing and generating an original feature vector of the user; determining a similar user group by analyzing the similarity of user feature vectors, extracting potential features such as behavior pattern features, preference features and common features, and generating enhanced feature vectors in combination with key features such as prices and historical sales conditions of vehicle-mounted non-vehicle products; and based on the enhanced feature vector and the purchase history of the similar user, recommending the most suitable vehicle-mounted non-vehicle product. According to the method, by accurately analyzing the personalized demands and behavior modes of the users, the product recommendation accuracy can be improved, unnecessary recommendation is reduced, and the satisfaction degree of the users and the product promotion rate are improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, device, equipment and storage medium for outputting non-vehicle products attached to a vehicle. Background Art

[0002] In recent years, with the continuous growth of the number of cars in my country, the auto insurance market has gradually matured, and the types of auto insurance products and insurance demand have continued to increase. However, the current domestic auto insurance premiums have reached saturation, and the growth space of traditional auto insurance business has gradually shrunk. Insurance companies face many restrictions in the process of expanding the scale of traditional auto insurance business, and it is difficult to increase premium income by increasing auto insurance sales. At the same time, due to the increasingly fierce competition in the insurance market, auto insurance premiums have shown a downward trend, while the claims costs are relatively fixed, resulting in serious challenges to the profitability of insurance companies.

[0003] For consumers, the premium saturation of the auto insurance market has also brought about the problem of product homogeneity. Existing auto insurance products are similar in terms of coverage and premium structure. Consumers lack differentiation when choosing insurance products and cannot obtain tailor-made products according to their own personalized needs. This phenomenon makes it difficult for insurance companies to improve customer experience through product differentiation, and it is also impossible to effectively enhance customer brand loyalty and stickiness.

[0004] In this context, non-car insurance products are gradually becoming one of the breakthroughs for insurance companies to explore business innovation. Non-car insurance products are non-traditional insurance products attached to vehicles, such as accident insurance and extended warranty services. However, existing non-car insurance products are still relatively simple and fixed, and it is difficult to flexibly adjust them according to the personalized needs of each customer, and they cannot fully meet the diverse needs of different customers. This makes insurance companies face the challenge of product homogeneity similar to traditional auto insurance when promoting non-car insurance products, which limits the development of their potential market.

[0005] In addition, most existing non-auto insurance recommendation systems are based on simple rules or historical purchase data, which cannot effectively analyze users' actual needs and purchase preferences. This lack of an accurate recommendation mechanism has led to a low promotion rate for non-auto insurance products and low customer satisfaction, further affecting the revenue growth of insurance companies. Summary of the invention

[0006] The main purpose of the present invention is to provide a method, device, equipment and storage medium for outputting non-vehicle insurance products, aiming to solve the technical problems that the existing non-vehicle insurance analysis technology lacks an intelligent output mechanism based on user personalized needs and behavior patterns, cannot provide accurate product matching, and leads to low product promotion rate and customer satisfaction.

[0007] To achieve the above object, the present invention provides a method for outputting non-vehicle products attached to a vehicle, comprising:

[0008] Acquire voice data related to insurance products for each user, and convert the voice data into voice text;

[0009] Obtain each user's insurance product purchase history and collect each user's personal information and risk assessment data;

[0010] Analyze each user’s voice text, insurance product purchase history, personal information, and risk assessment data to form each user’s original feature vector;

[0011] Analyze the similarity between the original feature vector of the target user and the original feature vectors of other users to determine a similar user group of the target user;

[0012] Extracting potential features from the similar user group, wherein the potential features include behavior pattern features, preference features and common features;

[0013] Extracting key product features from the vehicle-attached non-vehicle product data, wherein the key product features include price and historical sales;

[0014] Combining the original feature vector, potential features and product key features to generate an enhanced feature vector;

[0015] Based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group, the vehicle-related non-vehicle product that best suits the target user is determined and output.

[0016] Furthermore, in order to achieve the above-mentioned purpose, the present invention provides a vehicle-mounted non-vehicle product output device, comprising:

[0017] A speech recognition module, used to obtain speech data related to insurance products from each user and convert the speech data into speech text;

[0018] A user data collection module is used to obtain each user's insurance product purchase history and collect each user's personal information and risk assessment data;

[0019] The original feature extraction module is used to analyze each user's voice text, insurance product purchase history, personal information and risk assessment data to form the original feature vector of each user;

[0020] A user group division module is used to analyze the similarity between the original feature vector of the target user and the original feature vectors of other users, and determine a similar user group of the target user;

[0021] A potential feature extraction module, used to extract potential features from the similar user group, wherein the potential features include behavior pattern features, preference features and common features;

[0022] A product feature extraction module is used to extract key product features from the vehicle-attached non-vehicle product data, wherein the key product features include price and historical sales;

[0023] A feature fusion module, used to combine the original feature vector, potential features and product key features to generate an enhanced feature vector;

[0024] The product output module is used to determine and output the most suitable non-vehicle accompanying product for the target user based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group.

[0025] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a memory, a processor, and a vehicle-related non-vehicle product output program stored in the memory and executable on the processor, and the vehicle-related non-vehicle product output program, when executed by the processor, implements the steps of the vehicle-related non-vehicle product output method as described above.

[0026] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a vehicle-related non-vehicle product output program is stored, and when the vehicle-related non-vehicle product output program is executed by a processor, the steps of the vehicle-related non-vehicle product output method as described above are implemented.

[0027] Beneficial effects: The present invention relates to the fields of artificial intelligence technology and financial technology, and discloses a method for outputting non-car products that are attached to a vehicle. The method obtains the user's voice data and converts it into text, collects the user's insurance product purchase history, personal information, and risk assessment data, analyzes and generates the user's original feature vector; determines similar user groups by analyzing the similarity of user feature vectors, extracts potential features such as behavioral pattern features, preference features, and common features, and generates enhanced feature vectors in combination with key features such as the price and historical sales of non-car products that are attached to the vehicle; and recommends the most suitable non-car products that are attached to the vehicle based on the enhanced feature vectors and the purchase history of similar users. The present invention can improve the accuracy of product recommendations, reduce unnecessary recommendations, and improve user satisfaction and product promotion rate by accurately analyzing user personalized needs and behavior patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0029] Figure 1 A schematic diagram of an application environment of a method for outputting a vehicle-mounted non-vehicle product in an embodiment of the present invention;

[0030] Figure 2 It is a flow chart of an embodiment of a method for outputting a vehicle-mounted non-vehicle product according to the present invention;

[0031] Figure 3 It is a functional module diagram of a preferred embodiment of the vehicle-mounted non-vehicle product output device of the present invention;

[0032] Figure 4 A schematic diagram of the structure of a computer device in one embodiment of the present invention;

[0033] Figure 5 FIG. 4 is another schematic diagram of the structure of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0034] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0035] The method for outputting non-vehicle products provided in the embodiment of the present invention can be applied in the following aspects: Figure 1 In an application environment, the user terminal communicates with the server terminal through a network. The server terminal can obtain the user's voice data through the user terminal and convert it into text, collect the user's insurance product purchase history, personal information and risk assessment data, analyze and generate the user's original feature vector; by analyzing the similarity of the user's feature vector, determine the similar user group, extract potential features such as behavior pattern features, preference features and common features, and combine key features such as the price and historical sales of the vehicle-mounted non-vehicle products to generate an enhanced feature vector; based on the enhanced feature vector and the purchase history of similar users, recommend the most suitable vehicle-mounted non-vehicle products. The present invention can improve the accuracy of product recommendations, reduce unnecessary recommendations, and improve user satisfaction and product promotion rate by accurately analyzing the user's personalized needs and behavior patterns. Among them, the user terminal can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0036] See also Figure 2 , Figure 2 This is a flow chart of an embodiment of a method for outputting non-vehicle products provided by the present invention. It should be noted that although a logical sequence is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0037] like Figure 2 As shown, the method for outputting non-vehicle products provided by the present invention comprises the following steps:

[0038] S10, acquiring voice data related to insurance products from each user, and converting the voice data into voice text;

[0039] In this embodiment, voice interaction is performed with the user through a specific communication platform or application. The user uses the platform or application to have conversations related to insurance products, and the system will collect the user's voice data in real time. This process can be performed through smart devices (such as smartphones, voice assistants, etc.). The platform needs to identify the user's identity and associate it with his account to ensure that the acquired voice data matches the correct user.

[0040] Using communication interfaces, such as WeChat, phone applications, etc., the voice stream sent by the user is obtained through API or SDK. The voice stream can be stored on the server in real time or offline, using an appropriate data format (such as WAV or MP3). A monitoring mechanism is set up on both the device and server sides to ensure the integrity and high-quality recording of the voice data, avoiding noise and audio loss. In addition, real-time transmission of voice data can be achieved through network protocols (such as HTTP or WebSocket).

[0041] After the collected voice data is acquired, it needs to be stored in the server for subsequent processing. The security and integrity of the data must be guaranteed during storage, especially for user data involving personal privacy, which should be encrypted for storage. At the same time, the data should be associated with other information of the user through a unique identifier to ensure that the corresponding user can be correctly identified during subsequent processing.

[0042] Voice data is managed through a cloud storage platform, combined with big data storage technologies (such as HBase, Hive, etc.) to ensure efficient access to massive voice data. At the same time, voice files are compressed during storage to save storage space, and encryption technologies such as AES are used to ensure that user privacy is not leaked. The unique identifier is generated by the user's login account or authentication system, and the OAuth authentication mechanism can be used to bind voice to user data.

[0043] After the voice data is collected, it is converted into text through ASR technology. ASR technology uses deep learning models (such as convolutional neural networks and recurrent neural networks) to process voice data, analyze the features in the voice waveform, and match it to the existing language model library, thereby converting the voice information into corresponding text information. For dialogues in the insurance field, it is particularly necessary to combine language models and terminology libraries in specific fields to improve recognition accuracy.

[0044] Use a third-party or self-built ASR system to process voice data. In the system, a model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) is used to convert speech to text. By setting up a specific language library in the insurance field, the ASR model is trained to identify key terms related to insurance to ensure the accuracy of text conversion. Distributed computing is used in the cloud to accelerate the processing of large-scale voice data, and voice post-processing technology is used to remove background noise and misrecognition of unclear parts of the voice.

[0045] The converted text may contain recognition errors or irregular formats (such as spelling errors, punctuation errors, etc.), so preprocessing operations are required. Preprocessing includes removing noise from the text, correcting errors in speech recognition (such as typos, mishearing), and standardizing text formats (such as sentence structure, punctuation, etc.).

[0046] Use natural language processing (NLP) technology to preprocess the text. Correct errors through language models (such as BERT, GPT), and perform semantic corrections in combination with proprietary terminology libraries in the insurance field. You can also use the rule engine to fix common typos, grammatical structures, and other problems, and add punctuation. During the processing, incorrectly identified sentences will be labeled by the algorithm for manual review to further improve the accuracy of preprocessing.

[0047] After the speech text processing is completed, it needs to be bound to the corresponding user through a unique identifier. Each user has an independent identifier in the system, which can be generated by the user's login account, device number, etc. After the text is associated with the user, it ensures that subsequent personalized recommendations can be made based on the correct user historical data.

[0048] The unique identifier can be generated by the user's unique information such as account number, mobile phone number, etc. The system records the correspondence between the user and the voice text through a relational database (such as MySQL) or a NoSQL database (such as MongoDB) to ensure that the user's preferences and needs can be accurately identified in the subsequent analysis and recommendation process.

[0049] Example description: A user asks through a smartphone app, "I need to add an extended warranty service for my new car." The system converts the voice content into text through voice recognition and extracts the key information "extended warranty service." This demand is then combined with the user's purchase history and personal information to generate personalized recommendations for non-car insurance. In the system's recommendation results, the user is given a variety of non-car insurance products suitable for his or her vehicle, including extended warranty, roadside assistance, and accident insurance services. In the end, the user chooses the extended warranty service and completes the insurance process.

[0050] Through intelligent voice recognition technology, natural language interaction between users and systems is realized, which improves the convenience and accuracy of data collection. At the same time, through the preprocessing of text data and the association of unique identifiers, the personalization and privacy protection of each user's data are ensured. In the insurance field, users can obtain personalized insurance product recommendations through simple voice interaction, thereby improving user experience and the promotion rate of insurance products.

[0051] S20, obtaining each user's insurance product purchase history, and collecting each user's personal information and risk assessment data;

[0052] In this embodiment, the insurance product purchase history includes detailed records of all insurance products that the user has ever purchased. These records not only cover non-car insurance products, but also other types of insurance, such as car insurance, accident insurance, medical insurance, etc. Each purchase record needs to include information such as the type of insurance, purchase date, policy term, premium amount, and related additional terms. This data is the basic data for analyzing user historical behavior and preferences.

[0053] The system accesses the user's insurance product purchase history through the insurance company's database or the API interface of the third-party insurance platform. When each user purchases insurance, the system automatically stores the policy information and collects the data with the user's consent. Use a relational database (such as MySQL or PostgreSQL) to save the insurance purchase history, and the data is associated with the user account through the user's unique identifier (such as user ID). The system ensures that each user's insurance history is updated in real time by regularly calling the purchase records in the database.

[0054] Personal information includes basic information of the user, such as name, gender, age, occupation, income level, residential address, contact information, etc. This information is very important for analyzing the user's insurance needs and risk assessment, especially for non-auto insurance. The user's occupation and age may affect the type of recommended products, such as whether extended warranty, roadside assistance or accident insurance is needed.

[0055] When a user registers on an insurance service platform or purchases an insurance product, the system collects the user's personal information through a registration form or third-party identity verification (such as through WeChat, Alipay, or mobile phone number). The data is transmitted via encryption (such as SSL encryption) and stored in an encrypted database (such as AES encryption). The user's personal information is associated with their insurance product purchase history through the user ID to ensure that a complete user profile can be obtained during the analysis process.

[0056] Risk assessment data is used to assess the risk level of users in specific scenarios. This data may include the user's driving record (such as whether there is a history of traffic violations or accidents), health status, and vehicle usage (such as vehicle mileage, maintenance records, etc.). Risk assessment data helps insurance companies develop personalized premiums and protection plans for users, and consider the user's specific needs in the recommendation of non-auto insurance.

[0057] Risk assessment data is obtained through integration with external data sources, such as traffic management departments and insurance companies' own historical claims data. The system calls these external data sources through APIs and compares them with the user's insurance account information to generate the user's risk assessment level. Big data analysis technologies (such as Hadoop or Spark) are used to process large amounts of risk assessment data such as driving records and health conditions to ensure the real-time and accuracy of the assessment.

[0058] Example description: The user registered on the insurance company platform and filled in personal information, including name, age, occupation and annual income. The system automatically obtained the user's purchased auto insurance and accident insurance history from his previous insurance purchase records. At the same time, the system also obtained the user's driving record in the past three years from the external traffic management department, showing that he had no major traffic violations. Based on the user's purchase history and good risk assessment data, the system recommended extended warranty services and accident insurance supplementary insurance to the user. These recommendations are automatically generated based on the user's personal information, purchase preferences and risk assessment, which greatly improves the personalization and accuracy of the recommendations.

[0059] By obtaining the user's insurance purchase history, personal information and risk assessment data, the system can analyze the user's needs and risk status more comprehensively and accurately. This provides basic data support for the subsequent generation of user portraits and recommendation of personalized non-auto insurance products, improving the accuracy of recommendations and product promotion rate. At the same time, combined with risk assessment data, the system can dynamically adjust product recommendation plans to provide users with insurance products that best suit their risk status, increase user satisfaction and reduce insurance companies' risks.

[0060] S30, analyzing each user's voice text, insurance product purchase history, personal information, and risk assessment data to form an original feature vector for each user;

[0061] In this embodiment, the user's voice text data contains the natural language interaction content related to the user and the insurance product, such as the user's expression of demand for a specific insurance product. By analyzing these voice texts, the system can extract the user's potential demand and intention for the vehicle-mounted non-vehicle insurance product. This analysis process includes natural language processing (NLP) and semantic analysis technology to ensure that the user's real needs can be understood.

[0062] The system uses natural language processing (NLP) technology to perform semantic analysis on the text converted from the user's voice. Using syntactic analysis and named entity recognition (NER) technology, the system can identify keywords related to insurance products (such as "extended warranty", "accident insurance", etc.) from the text. After these keywords are extracted and classified, they form the user's insurance demand feature data. Combined with the proprietary language model in the insurance field, the system can effectively capture the potential needs expressed by users.

[0063] The user's insurance product purchase history includes records of all insurance products that the user has purchased in the past, which reflect the user's historical behavior and preferences. For example, whether the user tends to purchase extended warranties or additional accident insurance. The system extracts the user's insurance purchase preference characteristics by analyzing these purchase records.

[0064] The system extracts relevant insurance product information from the user's historical purchase records. Through statistical analysis of data such as purchase frequency, product type, and purchase amount, the system can construct the user's purchase preference characteristics. Use data mining algorithms (such as K-means clustering) to classify the user's purchase history data and identify the user's preference for vehicle-related and non-vehicle insurance products. For example, the system can identify whether the user is more inclined to purchase products such as extended warranty or accident insurance based on purchase behavior.

[0065] Personal information (such as the user's age, gender, occupation, income, etc.) is an important basic data for forming a user profile. This information can affect the user's demand for insurance products. For example, young car owners may be more inclined to purchase extended vehicle warranties, while high-income users may pay more attention to high-end services such as roadside assistance. By analyzing personal information, the system can form the user's economic status and demographic characteristics.

[0066] The system generates the user's demographic and economic characteristics by performing statistical analysis on the user's personal information. Using statistical methods such as logistic regression analysis, the system can identify the relationship between specific demographic characteristics and the demand for insurance products. For example, age and occupation can be used to predict the intensity of user demand for non-vehicle insurance products (such as accident insurance and rescue services). In addition, the system can analyze the correlation between the user's income level and insurance purchasing behavior through machine learning models (such as decision trees) to derive economic characteristics.

[0067] The user's risk assessment data includes driving records, health conditions, vehicle usage, etc., which can reflect the user's risk level in driving and daily life. Analyzing this data helps predict the user's demand for insurance products. For example, users with higher risks may be more inclined to purchase accident insurance or high-risk supplementary insurance, while users with lower risks may pay more attention to daily maintenance and extended warranty.

[0068] The system analyzes the user's risk status by obtaining information such as the user's driving record and health assessment data. Using a scoring model (such as the FICO scoring model or a custom risk scoring algorithm), the system quantifies each user's risk data into a risk level. By comparing the risk data of other users, the system can identify the difference between high-risk users and low-risk users. Then, based on the user's risk level, the system recommends corresponding insurance products to them. For example, high-risk users may be recommended to purchase additional non-auto insurance riders.

[0069] The features extracted from voice text, insurance product purchase history, personal information and risk assessment data are integrated to form the user's original feature vector. This feature vector includes data in multiple dimensions, each of which represents a user's specific behavior, preference or characteristic in a certain field. For example, demand in voice text, product type in purchase history, age and income in personal information, driving record in risk assessment, etc. The original feature vector is an important input for the subsequent personalized recommendation model.

[0070] Through feature engineering, the system integrates various user data and constructs a multi-dimensional feature vector. The dimension of each feature vector corresponds to an important feature of the user, such as insurance demand, economic status, risk level, etc. Through dimensionality reduction techniques such as principal component analysis (PCA), the feature vector can be optimized so that the subsequent collaborative filtering algorithm can perform efficient calculations. Finally, the user's original feature vector is stored in the database and serves as the basis for subsequent similar user matching and product recommendations.

[0071] Example: On the insurance company platform, the user expressed his need for an extended warranty for the vehicle through voice. At the same time, the system obtained the user's purchase history and found that he had purchased auto insurance and accident insurance many times. The system also combined the user's personal information (such as age and income level) and driving record, and finally generated the user's original feature vector. Through this feature vector, the system analyzed the user's high demand for extended warranty and accident insurance, and recommended suitable non-auto insurance products for the user.

[0072] By comprehensively analyzing the user's voice text, insurance product purchase history, personal information and risk assessment data, the system can construct a complete and personalized user original feature vector. This vector reflects the user's real needs and behavioral preferences, and can be effectively used for subsequent vehicle and non-vehicle insurance product recommendations. It improves the accuracy of user portraits, provides a basis for personalized recommendations, and also improves the promotion rate of insurance products and user satisfaction.

[0073] S40, analyzing the similarity between the original feature vector of the target user and the original feature vectors of other users, and determining a similar user group of the target user;

[0074] In this embodiment, the original feature vector of the target user is generated by analyzing multiple dimensions of data such as his voice text, insurance product purchase history, personal information, risk assessment data, etc. This feature vector contains information such as the user's insurance needs, purchase preferences, economic status, and risk level. Through this vector, the characteristics and behavior patterns of the user's needs for non-auto insurance products can be fully described.

[0075] The system first obtains the original feature vector of the target user from the database to ensure that it contains key information in multiple dimensions. The vector is standardized using feature engineering technology to ensure that data of different dimensions are compared on the same scale. This process can use methods such as Z-score standardization or Min-Max standardization.

[0076] In order to calculate the similarity of the target user, it is necessary to obtain the original feature vectors of all other users in the system. The original feature vector of each user is also composed of multi-dimensional data, reflecting their behavior patterns and preferences in the demand for non-auto insurance products.

[0077] The system efficiently obtains feature vectors of other users through distributed databases (such as Hadoop or Elasticsearch). In order to optimize computing efficiency, feature vectors can be stored in memory databases (such as Redis) in advance for quick access. The acquired data needs to be aligned with the feature vector of the target user to ensure consistent dimensions during comparison.

[0078] Comparing the target user's feature vector with the feature vectors of other users one by one is a key step in calculating similarity. By calculating the differences between the target user and other users in different dimensions, the system can determine the similarity between these users in terms of insurance needs, purchasing behavior, economic conditions, etc. The degree of similarity reflects whether the behavior patterns and needs of the target user and other users are similar.

[0079] Use a similarity calculation algorithm to compare one by one. Common similarity calculation methods include cosine similarity, Euclidean distance, and Manhattan distance. In cosine similarity calculation, the system treats each feature vector as a vector in a high-dimensional space and compares the angle between the two vectors. The smaller the angle, the higher the similarity.

[0080] The similarity score calculated each time indicates the degree of similarity between the target user and other users, and the score range is usually between 0 and 1 (1 means complete similarity, 0 means complete dissimilarity). The similarity score is the basis for determining which users have similar needs to the target user. These similar users will constitute the similar user group of the target user.

[0081] The similarity scores are obtained through the similarity calculation algorithm mentioned above, and the system stores these scores in the user feature database. In order to quickly retrieve users similar to the target user, the inverted index technology can be used to index the user's similarity scores and store them in the database for subsequent query and analysis. The system can also adjust the similarity calculation based on the weight of specific dimensions (such as the insurance demand feature may have a greater weight than the purchase history feature) to reflect more accurate user similarity.

[0082] The system sets a similarity threshold and selects other users whose similarity scores with the target user exceed the threshold as similar users. The similarity threshold is set by the system according to business needs. For example, it can be set to 0.7, which means that only users with a similarity score of more than 70% will be considered similar users. This similar user group is used for subsequent recommendations of non-auto insurance products.

[0083] The system uses SQL queries or distributed query engines (such as Apache Hive, Presto) to filter out users whose similarity scores are greater than the threshold in the user similarity database. In actual operation, fuzzy queries or sorted queries can be used to optimize the screening efficiency to ensure that similar users can be quickly identified. The system can also dynamically adjust the similarity threshold to ensure that the most appropriate products are recommended in different scenarios.

[0084] Similar user groups are other user groups that have similar behaviors and needs as the target user through screening. This user group reflects the common characteristics of these users in their needs for non-auto insurance products. Similar user groups provide a basis for subsequent product recommendation models. By analyzing the purchase history and preferences of similar users, the most suitable non-auto insurance products can be recommended to the target user.

[0085] The system stores all similar users that pass the threshold screening in the similar user group database and associates them with the feature vector of the target user. Similar user group data can be stored in NoSQL databases (such as MongoDB and Cassandra) to support query and analysis of large-scale user groups. The system passes the data of similar user groups to the subsequent recommendation algorithm module as an important input for product recommendation.

[0086] Example: User A expressed his need for extended warranty for his vehicle through voice. The system compared the original feature vector of user A with other users and calculated that the similarity between users B, C, and D and A was more than 0.8. The system grouped B, C, and D into a similar user group for user A, and by analyzing the purchase history of this group, it was found that B, C, and D had all purchased extended warranty and accident insurance. Based on the purchase records of these similar users, the system recommended similar non-vehicle insurance products to user A, including extended warranty and accident insurance. In the end, user A chose the extended warranty plan recommended by the system and successfully purchased insurance.

[0087] By analyzing the original feature vectors of the target user and other users, the system can effectively identify user groups with similar needs. The construction of similar user groups not only improves the accuracy of the recommendation system, but also improves the matching degree of non-auto insurance products. Similar user groups can reflect the common characteristics between users, so that the system can provide more personalized and accurate insurance product recommendations for target users based on the purchase history of similar users, thereby increasing the promotion rate of insurance products.

[0088] S50, extracting potential features from the similar user group, wherein the potential features include behavior pattern features, preference features, and common features;

[0089] In this embodiment, the behavior pattern feature refers to the common behavior patterns of similar user groups in the purchase of insurance products. These patterns may involve purchase frequency, purchase duration, renewal habits, etc., and can reflect the behavior patterns of users when purchasing non-auto insurance products. By analyzing these behavior patterns, the system can predict the actions that the target user may take, such as renewal or purchase of additional insurance products.

[0090] The system uses data mining technology to analyze the insurance purchase history of similar user groups and extract common behavior patterns. For example, it uses time series analysis technology (such as the ARIMA model) to identify users' purchase frequency and periodic behavior, such as purchasing accident insurance or extended warranty insurance every year or at regular intervals. At the same time, the system can use association rule algorithms (such as the Apriori algorithm) to discover commonalities in behavior patterns, such as users who frequently purchase a certain type of insurance product often also purchase other related products.

[0091] Preference characteristics reflect the preferences of similar user groups for non-car insurance products. These preferences may include the user's preference for specific insurance products (such as extended warranty, accident insurance, rescue services, etc.), as well as the preference for product prices and service content. The analysis of preference characteristics helps the system recommend the insurance products that best meet the needs of target users.

[0092] The system uses clustering algorithms (such as K-means clustering or hierarchical clustering) to classify the purchase history data of similar user groups and extract the user's purchase preferences. For example, the system can find that a certain type of user is more inclined to buy a higher-priced all-inclusive insurance, while another type of user is more inclined to choose a basic insurance product with a high cost-effectiveness. By analyzing preference data such as purchase amount, product type, and additional services, the system can accurately capture the user's preference characteristics.

[0093] Common features refer to the common characteristics of similar user groups in terms of personal information, economic status, risk assessment, etc. These features help the system further refine the classification of user groups and provide a more accurate basis for product recommendations. Common features may include age group, occupation category, vehicle type, etc.

[0094] The system uses principal component analysis (PCA) or feature selection algorithms to identify common characteristics of similar user groups in terms of personal information and economic conditions. For example, the system may find that the common characteristics of a certain type of user are that they are between 30 and 40 years old, have a high income level, have rich driving experience, and tend to purchase high-end extended warranty services. These common characteristics can serve as an important reference for recommending non-auto insurance products to ensure that the recommended products can meet the needs of target users.

[0095] After extracting behavioral pattern features, preference features, and common features from similar user groups, the system needs to standardize these features. Standardization is to ensure that features of different dimensions can be compared and calculated within the same range to avoid model bias due to differences in feature scales.

[0096] The extracted features are standardized using Z-score standardization or Min-Max standardization techniques. The standardized feature values ​​will be compressed into a unified range, such as between 0 and 1, to ensure that the behavior pattern features, preference features, and common features can be reasonably weighed in the recommendation algorithm and comprehensively calculated in subsequent steps.

[0097] After standardization, the system integrates the extracted behavioral pattern features, preference features, and common features into a potential feature vector. This vector represents the comprehensive characteristics of similar user groups and reflects the common behaviors and preferences of similar user groups in terms of insurance product needs. These feature vectors provide a basis for subsequent recommendation algorithms.

[0098] The system combines different types of potential features into multi-dimensional feature vectors through feature fusion technology. Each dimension of the feature vector corresponds to a standardized potential feature (such as purchase frequency, product preference, etc.), and these features are stored in the database for subsequent use. In order to optimize storage and query efficiency, vector databases (such as Faiss) can be used to manage these feature vectors to support real-time query and calculation of large-scale data.

[0099] By extracting behavioral pattern features, preference features, and common features from similar user groups, the system can capture the key differences and commonalities in users' insurance needs. This multi-dimensional feature extraction method can effectively improve the accuracy of product recommendations, enabling the system to provide target users with recommendations for non-auto insurance products that best match their behaviors and preferences.

[0100] S60, extracting key product features from the vehicle-attached non-vehicle product data, where the key product features include price and historical sales status;

[0101] In this embodiment, the non-vehicle insurance mainly refers to the insurance products that are provided during the use of the vehicle and are different from traditional vehicle insurance (such as compulsory traffic insurance, vehicle damage insurance, third-party liability insurance, etc.). This type of insurance is not directly aimed at compensating for damage to the vehicle itself, but provides protection for scenarios such as vehicle users, driving environment and other additional services. The main types of non-vehicle insurance include:

[0102] Accident Insurance: Driver and Passenger Accident Insurance: Provides protection for vehicle drivers and passengers in the event of personal injury or death in a traffic accident, and is an additional personal insurance for vehicle owners and passengers.

[0103] Extended warranty insurance: Continues to provide repair coverage after the original warranty expires, covering mechanical failures, electrical problems, etc., helping car owners reduce repair costs.

[0104] Emergency Rescue Insurance: Provides towing and emergency rescue services for vehicles due to breakdowns, accidents, etc. on the road, such as towing services, tire changes, and fuel delivery.

[0105] Vehicle theft and robbery insurance: protects against losses caused by theft or robbery of the vehicle, compensates for the losses after the vehicle is stolen or pays part of the value.

[0106] Liability insurance for people outside the vehicle: protects against damage caused to pedestrians or property outside the vehicle while the vehicle is in motion, and provides additional compensation liability.

[0107] Isolated breakage insurance: Covers situations where the car window glass is broken alone without involving damage to other vehicle parts. It is usually not included in traditional vehicle damage insurance.

[0108] Tire insurance: provides repair or replacement costs for wheels or tires damaged due to accidents, making up for the shortcomings of ordinary vehicle damage insurance.

[0109] Price features refer to the price information of vehicle-attached and non-vehicle insurance products, including the basic price, discount price, and possible additional fees of the product. Price is a key factor in user decision-making and affects the user's behavior in choosing different insurance products. Analyzing the price features of vehicle-attached and non-vehicle products can provide users with the most cost-effective insurance product recommendations.

[0110] The system accesses the insurance product database to extract the price information of each non-vehicle product. The price data of each product includes the standard price, historical discount information, and the price of additional services. The system uses SQL queries to extract these price data from the database and uses price as a key matching factor when matching the user feature vector with the product feature vector. In addition, the system can convert the price data to a range between 0 and 1 through data preprocessing using Min-Max normalization so that it can be integrated and calculated with other features.

[0111] Historical sales data includes the sales records and popularity of non-vehicle products in the past period of time. These data can reveal the popularity of a certain type of product among different user groups and the market performance of specific products. By analyzing historical sales data, the system can better predict which products may be favored by users in the future and make accurate product recommendations based on user feature vectors.

[0112] The system extracts the sales records of non-vehicle products by querying the historical sales database or the insurance company's CRM system. These records include sales quantity, sales time, and sales trends in a specific period. Using time series analysis, the system can analyze the sales of different products in different time periods, and combine user feature vectors to match non-vehicle products purchased by similar users. In addition, the system can generate a popularity ranking of products through a sales ranking algorithm so that they can be recommended to target users first.

[0113] Key product features include comprehensive data on price and historical sales. By extracting these two key features, the system can evaluate the comprehensive performance of non-vehicle insurance products and match them with the user's feature vector to recommend the most suitable product for the user. For example, a product with a lower price and high historical sales volume may be more suitable for users who value cost-effectiveness, while historical sales data of high-end products is suitable for high-income users.

[0114] The system uses feature fusion technology to integrate the extracted prices and historical sales to generate a comprehensive feature vector for the product. The price and sales data of each product are added to the vector as specific dimensions. By matching with the user's enhanced feature vector (such as user preferences, economic conditions, etc.), the system can accurately recommend products that meet the user's needs. The calculation of feature fusion can be achieved through machine learning models (such as random forests or gradient boosting trees), which can combine various feature weights to generate personalized product recommendations.

[0115] After extracting the price and historical sales, these features need to be standardized to ensure that features of different dimensions can be compared and calculated on the same scale. Standardized features can reduce the impact of different feature scales, making the recommendation algorithm more accurate and efficient.

[0116] The system uses Z-score normalization or Min-Max normalization technology to convert price and historical sales data into normalized values. For example, price data is mapped to a range of 0 to 1, and historical sales are also normalized by relative sales. The normalized data will be integrated into the product's feature vector and matched with the user's feature vector to ensure the accuracy of the recommendation.

[0117] Example: A user expressed through the insurance platform that he needed to choose an extended warranty service for his newly purchased vehicle. The system extracted the price information of the relevant extended warranty products from the vehicle-related non-vehicle product data and found that the price range was between 500 and 1,500 yuan. By analyzing historical sales data, the system found that a certain extended warranty product priced at 1,000 yuan had the highest sales volume in the past six months. Combining the user's financial situation and insurance preferences, the system finally recommended this most cost-effective extended warranty service to the user and successfully facilitated the sale.

[0118] By extracting key features such as price and historical sales from the data of non-vehicle products, the system can make an accurate assessment of the product's cost-effectiveness and market performance. Combined with the user's feature vector, these key features enable the recommendation system to provide users with product recommendations that best meet their needs.

[0119] S70, combining the original feature vector, potential features and product key features to generate an enhanced feature vector;

[0120] In this embodiment, the original feature vector includes features extracted from the user's voice text, purchase history, personal information and risk assessment data, representing the user's personalized needs and behaviors. The latent features are behavioral pattern features, preference features and common features extracted from similar user groups, reflecting the commonality of similar user groups in insurance needs and purchasing behaviors. The goal of this step is to merge the original feature vector of the target user with the latent feature vector of the similar user group to generate a more comprehensive user feature representation. The system merges the original feature vector and the latent feature vector in the feature space through feature splicing technology. Specifically, the original feature vector of each user and the latent features extracted from the similar user group are regarded as different parts of the high-dimensional vector, and the dimensions of these vectors are directly spliced ​​together to form a larger and more comprehensive feature vector. For example, the original feature vector of the user may contain 5 dimensions (purchase history, economic status, risk level, etc.), and the latent feature vector contains 3 dimensions (purchase preferences, behavioral patterns, common features), which are merged to form an 8-dimensional vector. This vector serves as the basis for the subsequent generation of enhanced feature vectors.

[0121] Key product features include information such as the price and historical sales of non-auto insurance products. These product features are used to describe the core characteristics of each insurance product and directly affect the user's selection behavior. Combining these key features with the user's original features and potential features can help the system better match the most suitable product for the user.

[0122] Use feature weighted fusion technology to fuse product key features with user feature vectors. First, standardize the feature vectors such as price and sales data of each insurance product to facilitate weighted calculation with data such as economic status and purchasing preferences in the user feature vector. For example, different feature weights can be set to ensure that prices have a greater impact on users who are more sensitive to economic conditions, while appropriately reducing the weight of prices for high-income users. The fusion process can be achieved through linear weighting or feature fusion models based on machine learning, such as multi-layer perceptron (MLP).

[0123] The enhanced feature vector is a high-dimensional vector that combines the original features of the user, the potential features of similar user groups, and the key features of the product. It represents the user's all-round needs and behavioral characteristics. This vector can better describe the user's potential demand for non-auto insurance products, and combines user preferences, behavioral patterns, and product market performance. The enhanced feature vector provides strong support for the final product recommendation algorithm.

[0124] Generate enhanced feature vectors through feature fusion models. The model can use a multi-layer perceptron (MLP) based on deep learning or a feature weighted fusion model to ensure that the original features, latent features, and product key features are reasonably weighted in the same vector space. The enhanced vector after feature fusion has stronger expressive power and can reflect the multi-dimensional needs of users and the characteristics of products. For example, the feature vector of a user may show that he tends to buy economical extended warranty insurance, while the characteristics of similar user groups show that such products have good market performance. The system can predict the product that the user is most likely to choose by enhancing the feature vector.

[0125] After generating the enhanced feature vector, the system needs to standardize and optimize the vector to ensure that features of different dimensions can be calculated on the same scale and avoid introducing deviations due to differences in feature values. The optimized enhanced feature vector can be used more efficiently in subsequent recommendation algorithms.

[0126] Use Z-score standardization or Min-Max standardization to compress all dimensions of the enhanced feature vector to the same range (such as between 0 and 1) for subsequent recommendation algorithms. During this process, the system adjusts the feature vector through optimization algorithms (such as gradient descent) to ensure that the final enhanced feature vector can accurately reflect user needs and product characteristics.

[0127] For example, the potential features of users and non-vehicle products accompanying the vehicle are extracted through a deep learning model, and these potential features are combined with the user's original feature vector through matrix decomposition to generate an enhanced feature vector.

[0128] User enhanced feature vector generation formula:

[0129] in, It is the potential features extracted by the deep learning model, representing the user's deep needs or behavior patterns; u is the user's original feature vector; α is a weight parameter used to adjust the ratio of potential features to original features.

[0130] Product enhancement feature vector generation formula:

[0131] in, is the enhanced feature vector of product i; is the potential feature vector extracted from the data of non-vehicle product i through deep learning, reflecting the deep characteristics of the product; v i is the original feature vector of the non-car product i, which contains the basic information of the product, such as price, sales history, etc.; β is the parameter that controls the merging weight between the latent features and the original features.

[0132] The user's enhanced feature vector The enhanced feature vector with the product Fusion is performed to generate a comprehensive feature vector. Through the fusion operation, the system can calculate the matching degree between the user and the product. The formula is as follows:

[0133]

[0134] in, It is the comprehensive feature vector of user u and product i, indicating their matching degree. is the enhanced feature vector of user u, is the enhanced feature vector of product i.

[0135] In order to optimize the feature vectors of users and products, the system uses a loss function to minimize the difference between the predicted value and the actual user behavior. The loss function formula is as follows:

[0136]

[0137] Among them, L is the loss function, which measures the actual behavior r of user u on product i ui and predictive behavior The difference between ui It is the user’s actual behavior towards the product (e.g., whether to purchase or not); It is the predicted value obtained by fusing the user and product feature vectors; is a regularization term used to prevent overfitting.

[0138] The system minimizes this loss function and adjusts the feature vectors of users and products to make the recommendation results more accurate.

[0139] By combining the original features of users, the potential features of similar user groups, and the key features of products, the system generates a more comprehensive and accurate enhanced feature vector. This enhanced feature vector can fully reflect the needs, preferences, and behavior patterns of users, while taking into account the market performance and price factors of the product, thereby improving the accuracy of the recommendation system.

[0140] S80, based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group, determine and output the most suitable vehicle-related and non-vehicle products for the target user,

[0141] In this embodiment, the enhanced feature vector is a comprehensive representation of the original features of the target user, the potential features of similar users, and the key features of the insurance product. By using the enhanced feature vector, the system can score the on-board non-car insurance products and evaluate the degree to which each product matches the needs of the target user. The score is calculated comprehensively based on multiple dimensions such as user needs, preferences, and market performance.

[0142] Using matching algorithms (such as dot product, cosine similarity, etc.), the user's enhanced feature vector is compared with the feature vector of each non-car product, and the matching score of each product is calculated. For example, the system calculates the final score of each insurance product through linear weighting according to the weight distribution of user preferences (such as price, coverage, service content, etc.). The higher the score, the more suitable the product is for the user. This process can be accelerated by matrix decomposition algorithm for batch calculation.

[0143] The purchase history of similar user groups reflects which vehicle and non-vehicle insurance products are more popular among groups similar to the target users. The historical purchase data of these products can provide important references for recommendations, especially those products that have been purchased multiple times, which usually means that the market performance is good and meets user needs.

[0144] The system extracts data such as the purchase frequency, purchase time, and purchase amount of non-vehicle insurance products from the purchase records of similar user groups through database query or historical record retrieval. Using aggregate query, the system can identify the insurance products with the highest purchase frequency and the best feedback among similar user groups. These products will be given priority in being recommended to target users. Through data analysis techniques (such as K-means clustering or Apriori association rule algorithm), the system can identify common purchase behaviors between similar users and extract high-frequency purchase products.

[0145] By analyzing the purchasing behavior of similar users, the system can identify which insurance products are most popular among similar user groups. For example, insurance products with a higher purchase frequency indicate that these products are in greater demand among this user group. In addition, some products may have a strong correlation, and when users buy one product, they often choose another related product.

[0146] Use association analysis (such as the Apriori algorithm) to analyze the purchase history of similar users and identify frequently purchased product categories and their related products. For example, the system can find that user A has similar purchasing habits with similar user groups B and C, and these users usually purchase extended warranties and additional roadside assistance services. Therefore, the system will give priority to recommending these products to target users. The analysis process can also be combined with factors such as the user's economic conditions and purchase duration to further refine the recommendations.

[0147] The system combines the matching score with the purchase history of similar users to select the most suitable on-board and off-board insurance products for the target users. The matching score reflects the fit between the product and the user's needs, while the purchase history of similar users provides market validation for the product. Combining these two dimensions, the system can recommend products more accurately.

[0148] The matching score and the purchase history of similar users are calculated using weighted average or machine learning algorithms (such as random forest or gradient boosting tree). For example, the system can give higher weights to the products with the highest purchase frequency among similar user groups, and adjust the recommendation priority according to the matching score, and finally filter out a group of the most suitable products for users to choose. This process ensures that the recommended products can meet user needs to the greatest extent through model optimization.

[0149] The system will display the selected vehicle and non-vehicle insurance products to the target users. The output product list includes not only the recommended product names, but also relevant price information, coverage content, historical sales, etc., so that users can clearly understand the advantages and disadvantages of each product and make a more informed choice.

[0150] Through the front-end display interface (such as mobile applications, web pages or intelligent voice assistants), the system will display the final recommended products to users. The product list is sorted by priority, and the products that best meet user needs will be at the top. At the same time, users can view detailed information about each product through the interface, such as price, coverage, purchase history, etc. In addition, the system can further adjust the product recommendation results based on user feedback. Data visualization technology (such as D3.js or ECharts) can be used to display product comparison charts, which is convenient for users to intuitively compare various insurance products.

[0151] Example description: User A expressed on the insurance platform that he needed to purchase non-vehicle insurance for his vehicle. The system first calculated the matching score with each extended warranty product based on the enhanced feature vector of user A (including his economic status, risk assessment, purchase preferences, etc.). At the same time, the system extracted the most frequently purchased non-vehicle products from user groups similar to user A, and found that most users also chose roadside assistance services when purchasing extended warranties. Finally, the system comprehensively matched the matching score and purchase history and recommended a combination product that included extended warranty and roadside assistance. User A viewed the list of recommended products, and finally selected the combination and completed the insurance.

[0152] By combining the user's enhanced feature vector and the purchase history of similar users, the system can more accurately recommend the target user the vehicle and non-vehicle insurance products that best meet their needs. This recommendation method based on personalized needs and market verification not only improves the accuracy of the recommendation, but also improves user satisfaction and increases the product promotion rate. At the same time, the system can continuously optimize the recommendation results through the feedback mechanism, making the recommendation of insurance products more flexible and intelligent.

[0153] The present invention relates to the fields of artificial intelligence technology and financial technology, and discloses a method for outputting non-car products that are attached to a vehicle. The method obtains user voice data and converts it into text, collects the user's insurance product purchase history, personal information and risk assessment data, analyzes and generates the user's original feature vector; determines similar user groups by analyzing the similarity of user feature vectors, extracts potential features such as behavior pattern features, preference features and common features, and generates enhanced feature vectors in combination with key features such as the price and historical sales of non-car products that are attached to the vehicle; and recommends the most suitable non-car products that are attached to the vehicle based on the enhanced feature vectors and the purchase history of similar users. The present invention can improve the accuracy of product recommendations, reduce unnecessary recommendations, and improve user satisfaction and product promotion rate by accurately analyzing user personalized needs and behavior patterns.

[0154] In one embodiment, the above S10 includes:

[0155] S101, performing voice interaction with the user related to the insurance product through the communication platform and collecting the user's voice data;

[0156] S102, using an automatic speech recognition module to convert the speech data into speech text;

[0157] S103, performing a preprocessing operation on the speech text, wherein the preprocessing operation includes removing noise, correcting speech recognition errors, and standardizing text format;

[0158] S104: Associating the pre-processed voice text with the corresponding user through the user's unique identifier.

[0159] In this embodiment, the user is interacted with through a communication platform (such as a telephone, mobile application, or intelligent voice assistant) to collect voice data related to the user and the insurance product. These voice data may include the user's insurance needs, consulting questions, feedback, etc. The system needs to be able to obtain high-quality audio data from the conversation and ensure that it is relevant to the insurance product. The system establishes real-time voice communication with a communication platform (such as a mobile application, telephone system, or smart speaker) through an integrated voice interface (API). Voice data is collected and stored on the server in real time to ensure that high-quality audio is captured. By using audio coding technology (such as PCM or AAC), the voice data is compressed and efficiently transmitted to the cloud while retaining the clarity and integrity of the audio.

[0160] The automatic speech recognition (ASR) module is responsible for converting the collected user voice data into readable text. The accuracy of speech recognition depends on the training of the ASR model, especially in the insurance field, to ensure that the model can correctly identify industry terms and keywords related to insurance products. The system uses deep learning models (such as LSTM, Transformer, or Conformer-based speech recognition models) to process speech data and convert it into text. During the training process, the ASR model needs to load the insurance industry's proprietary vocabulary and context understanding module to improve the recognition accuracy of insurance-related terms. The system achieves efficient speech-to-text function by integrating commercial ASR services (such as Google Speech-to-Text, AmazonTranscribe) or custom speech recognition engines.

[0161] The text generated by speech recognition may contain noise, recognition errors, or irregular formats, so the text needs to be preprocessed. Preprocessing includes removing excess noise and pauses, correcting common recognition errors, and standardizing the text format to make it suitable for subsequent processing steps and analysis. The system cleans and standardizes the text through natural language processing (NLP) technology. Using regular expressions and semantic analysis tools, the system can remove irrelevant content (such as noise markers, meaningless words), and correct common errors (such as "auto insurance" is recognized as "insurance") through specific rules in the insurance field. At the same time, the text format will be standardized into a unified grammatical structure to ensure that it is suitable for subsequent feature extraction and analysis.

[0162] Each user has a unique identifier (such as a user ID or mobile phone number), and the system needs to ensure that the pre-processed voice text is associated with the user. This step ensures that each user's insurance needs are correctly archived and that the user's profile and needs can be accurately matched in subsequent analysis. Using a user authentication system (such as OAuth or JWT token), the system can generate a unique identifier for each user. When a user interacts with the system by voice, the system automatically associates the identifier with the user's voice text. The pre-processed text will be stored in a relational database or NoSQL database (such as MongoDB, Redis) to ensure that the voice data corresponds one-to-one with the user information during subsequent data processing.

[0163] In this embodiment, the system can collect the user's insurance needs in real time through voice interaction with the user, and convert the voice into text through automatic voice recognition technology, reducing the burden of manual input by the user. The pre-processed voice text is associated with the user through a unique identifier, providing accurate data support for subsequent personalized recommendations and analysis.

[0164] In one embodiment, the above S30 includes:

[0165] S301, processing and analyzing the voice text of each user to extract insurance demand features in the voice text;

[0166] S302, analyzing each user's insurance product purchase history and extracting the user's purchase preference characteristics;

[0167] S303, analyzing each user's personal information and extracting the user's personal economic characteristics;

[0168] S304, analyzing the risk assessment data of each user and extracting the risk status characteristics of the user;

[0169] S305, combining the insurance demand characteristics, purchase preference characteristics, personal economic characteristics and risk status characteristics into an original feature vector for each user.

[0170] In this embodiment, the voice text contains the user's natural language interaction content related to insurance, such as the demand for insurance products, consulting questions, etc. By analyzing these voice texts, the system can extract the user's demand for different types of insurance products, such as vehicle extended warranty, accident insurance or roadside assistance services, etc. These demand characteristics are part of the user's personalized characteristics and are used for subsequent product recommendations.

[0171] The system uses natural language processing (NLP) technology to analyze the user's voice text and extract key insurance demand features. Using technologies such as semantic analysis, named entity recognition (NER), and topic models (such as LDA), the system can identify the core content in the voice text and convert it into structured insurance demand features. For example, keywords such as "extended warranty" and "accident insurance" mentioned by the user will be extracted and mapped to specific insurance needs. The system can also identify the user's potential needs and analyze the user's possible future insurance needs by understanding the conversation context.

[0172] Insurance product purchase history records the insurance products that users have purchased in the past. These records can reflect the user's preferences for insurance product types, purchase frequency, policy amounts, and renewal behavior. By analyzing the purchase history, we can understand the user's insurance preference characteristics, such as whether they prefer low-priced basic insurance products or high-end insurance products with comprehensive coverage. The system analyzes the user's insurance purchase history and uses data mining and cluster analysis techniques to extract purchase behavior patterns. By analyzing the user's past purchase frequency, product type, purchase duration and other dimensions, the system can construct the user's purchase preference characteristics. For example, whether the user tends to purchase additional insurance, extended warranty services, or specific insurance company products. The system can also use frequent pattern mining (such as the Apriori algorithm) to identify the user's preferences for insurance product combinations, such as the user may tend to purchase vehicle insurance and accident insurance at the same time.

[0173] The user's personal information, such as age, gender, occupation, income level, place of residence, etc., can reflect their economic status and spending power. This information is crucial for the recommendation of insurance products, because different insurance products are often targeted at different income levels or occupational groups. The extraction of economic characteristics can help the system recommend insurance products that match the user's economic ability. The system analyzes the user's registration information and personal information, and uses statistical analysis and machine learning models (such as decision trees or logistic regression) to extract the user's economic characteristics. For example, income level, occupational nature, and regional economic level will directly affect the user's preference for high-end or basic insurance products. Through this information, the system can infer the user's spending power and acceptance of insurance products in different price ranges.

[0174] Risk assessment data typically includes the user's driving record, health status, claims history, etc. This data reflects the user's risk level in terms of vehicle use, personal health, etc. Users with higher risks may be more inclined to purchase additional insurance protection, while users with lower risks may choose standard insurance products. The system quantifies the user's risk level using a scoring model (such as a credit score or risk assessment model) by accessing the user's driving record, health assessment data, and insurance claims history. By analyzing this data, the system can predict the user's need for additional insurance protection. For example, the system may find that the user has a history of multiple claims, so it recommends that he or she purchase additional accident insurance or high-risk protection insurance. The data is obtained through APIs or integration with third-party services.

[0175] The original feature vector is a multi-dimensional vector, and each dimension represents an important feature of the user, including insurance demand, purchase preference, economic status and risk level. The vector integrates all the key information of the user and provides a basis for subsequent personalized recommendations and analysis. The system combines user features of multiple dimensions into a high-dimensional feature vector through feature splicing. Each dimension corresponds to an important feature of the user, such as the intensity of insurance demand, the category of purchase preference, the level of economic status and the score of risk assessment. Through feature standardization (such as Min-Max standardization or Z-score standardization), the system ensures that features of different dimensions can be calculated on the same scale. The final generated original feature vector will be stored in the vector database for subsequent similar user matching and product recommendations.

[0176] This embodiment can build a complete user portrait by comprehensively analyzing the user's voice text, insurance purchase history, personal information and risk assessment data, reflecting the user's characteristics in terms of insurance needs, purchase behavior, economic status and risk level. The original feature vector provides a solid foundation for personalized recommendation and user demand analysis, enabling the system to more accurately recommend on-board and off-board insurance products that meet user needs, thereby improving user satisfaction and product sales rate.

[0177] In one embodiment, the above S40 includes:

[0178] S401, comparing the original feature vector of the target user with the original feature vectors of all other users one by one;

[0179] S402, using a similarity measurement module to analyze similarity scores between the target user and other users;

[0180] S403, according to a preset similarity threshold, screening out other users whose similarity scores exceed the similarity threshold as similar users of the target user;

[0181] S404: Group the screened similar users into a similar user group.

[0182] In this embodiment, the system first needs to calculate the similarity between the target user and other users in terms of purchase history, risk assessment and other characteristics. Through this comparison, the system can identify other users with similar behavior patterns to the target user. In order to accurately calculate these similarities, the system uses a similarity measurement method based on the Pearson correlation coefficient. Using the Pearson correlation coefficient formula in the figure, calculate the similarity between the target user and other users in insurance purchase records. The formula is as follows:

[0183]

[0184] Among them, A and B represent two users; IAB Represents the set of insurance products purchased by users A and B together; and It is the purchase status of users A and B for insurance product i, 1 represents purchase and 0 represents non-purchase. and Represents the average purchase preference of users A and B. This formula calculates the similarity score by comparing the insurance product purchase records of two users. The system first standardizes each user's purchase record and calculates the user's average purchase preference, and then calculates the similarity between all users one by one.

[0185] After calculating the similarity scores of each user using the Pearson correlation coefficient, the similarity measurement module will evaluate the similarity between each user and the target user based on these scores. The users with higher similarity have purchase behaviors that are closer to the target user, and the system will give priority to the purchase behaviors of these users in subsequent recommendations.

[0186] By calculating the similarity scores derived from the formula (for example, based on the Pearson correlation coefficient), the system can evaluate the similarity between each pair of users and generate a matrix to store the similarities between all users. Each similarity score reflects the degree of similarity between two users in terms of purchasing behavior and risk preferences. The calculation of this matrix can be accelerated by a parallel computing framework, such as using matrix multiplication for batch computing.

[0187] The system sets a similarity threshold to ensure that only users with highly similar behaviors to the target user are screened as similar users. Users with similarity scores exceeding this threshold will be considered similar users of the target user and used as a reference for subsequent recommendations.

[0188] By setting a threshold for the similarity score (e.g., 0.7 or 0.8), the system can filter out users whose similarity to the target user exceeds the threshold. The characteristics and purchasing behaviors of these users will be further analyzed to predict the potential insurance needs of the target user. The similarity screening process can be implemented through SQL queries or Boolean condition filtering methods to ensure efficient screening of similar users who meet the conditions.

[0189] The similar user group is composed of users who are highly similar to the target user in terms of behavior, risk preference, purchase history, etc. By analyzing the behavior patterns of this user group, the system can predict the on-board and non-car insurance products that the target user may purchase and make personalized recommendations.

[0190] After similarity screening, the system integrates all users who meet the similarity criteria into a similar user group. Next, the system will mark the common characteristics of the group, such as purchase frequency, product preferences, price sensitivity, etc. The creation of similar user groups helps with subsequent personalized recommendations, especially when recommending products, the system can give priority to the most popular products in these user groups.

[0191] In this embodiment, by comparing the original feature vector of the target user with the feature vectors of other users, the system can effectively identify user groups with similar needs. The construction of similar user groups provides a basis for the personalized recommendation system. The system can recommend the vehicle-mounted non-vehicle insurance products that best meet the needs of the target user by analyzing the behavior and purchase history of these similar users.

[0192] In one embodiment, the above S50 includes:

[0193] S501, collecting behavior data of each similar user in the similar user group, including purchase frequency and preference change information;

[0194] S502, using a deep learning model to analyze the behavior data of similar user groups to extract hidden behavior pattern features and preference features;

[0195] S503, identifying common features of similar users in a similar user group, wherein the common features include common purchasing habits or preference trends;

[0196] S504, standardizing the extracted behavior pattern features, preference features, and common features;

[0197] S505, integrating the standardized behavior pattern features, preference features and common features into a complete potential feature vector.

[0198] In this embodiment, the behavior data of each similar user includes their purchase frequency, purchase cycle, product type, etc. of vehicle-related non-vehicle insurance products. These data reflect the user's insurance purchasing habits and demand changes. By collecting this information, the system can further understand the potential needs of target users and provide a data basis for personalized recommendations.

[0199] The system extracts the purchase behavior data of similar users by mining their purchase history. This data may include the frequency of users purchasing non-vehicle insurance (such as monthly or annual purchases), the categories of products purchased (such as extended warranty, roadside assistance, etc.), and the changing trends of purchase decisions (such as switching from basic products to advanced products). The system obtains this information through database queries or log analysis tools and stores it in the behavior data set for subsequent deep learning model analysis.

[0200] Behavioral pattern features are extracted by analyzing user behavior data, such as changes in users' insurance purchasing behavior in a specific time period or under specific circumstances. Preference features reflect users' preferences for different types of insurance products, such as a preference for high-risk products or high-security products. Through deep learning models, the system can discover these hidden patterns and preference features.

[0201] The system uses deep learning models (such as RNN, LSTM or Transformer) to analyze the behavioral data of similar user groups to extract complex behavioral patterns and changing trends. By analyzing time-series behavioral data, deep learning models can identify user behavior patterns, such as users may be more inclined to purchase additional insurance or other high-security products during a specific period. The model can also extract preference features, such as user preferences for price, coverage, service quality and other factors. Through training, deep learning models can adaptively adjust parameters to extract potential features that reflect users' actual needs.

[0202] In similar user groups, some users may have the same purchasing habits or preferences, such as regularly purchasing extended warranty products or preferring cost-effective insurance products. Common characteristics reflect the purchasing behavior and product selection tendencies that most people in the user group have in common.

[0203] The system uses clustering algorithms (such as K-means or DBSCAN) to cluster the behavioral data of similar user groups and identify common features in the groups. For example, the system may find that some similar user groups pay more attention to price factors when purchasing insurance, while other groups tend to buy products with high protection and wider coverage. By identifying these common features, the system can further improve the accuracy of demand prediction and personalized recommendations for target users.

[0204] In order to ensure that different feature dimensions have consistent scales, the system standardizes the extracted behavioral pattern features, preference features, and common features. Standardization ensures that the values ​​of feature vectors are within the same range, avoiding unnecessary deviations in the results due to excessively large or small values ​​of certain features during the calculation process.

[0205] The system uses common standardization methods, such as Min-Max standardization or Z-score standardization, to scale the data range of all features. For example, Min-Max standardization maps feature data to the interval [0,1]. For behavioral pattern features and preference features, the system can calculate the maximum and minimum values ​​of each feature, or calculate the mean and standard deviation, and standardize each feature to ensure that all features are analyzed at the same scale.

[0206] The latent feature vector is a comprehensive representation of user behaviors, preferences, and common features, reflecting the user's complex needs and purchasing tendencies. By integrating all extracted features into a latent feature vector, the system can provide a complete user feature description for subsequent product recommendations.

[0207] The system combines all standardized features (including behavioral pattern features, preference features, and common features) into a complete vector through feature fusion methods. Each feature dimension represents an important attribute of the user, such as purchase frequency, product preference, etc. By combining these features, the system can generate a potential feature vector for each similar user. This potential feature vector will be used to measure the degree of match between users and products in the subsequent product matching and recommendation process.

[0208] By extracting behavioral pattern features, preference features, and common features from similar user groups, the system can fully capture the user's purchasing behavior and demand changes. Standardization ensures that all features are compared at the same scale to avoid calculation bias caused by feature differences. The final generated latent feature vector provides an accurate user portrait for the subsequent recommendation process, making the recommendation results more personalized, meeting the actual needs of users, and improving user satisfaction and product promotion rate.

[0209] In one embodiment, the above S70 includes:

[0210] S701, merging the potential feature vector extracted from the similar user group with the original feature vector of the target user to form a preliminary merged user feature vector;

[0211] S702, fusing the product key feature vector with the initially merged user feature vector to form a comprehensive feature vector;

[0212] S703: Generate an enhanced feature vector through a feature fusion module based on the comprehensive feature vector.

[0213] In this embodiment, the system extracts potential features related to the target user by analyzing similar user groups, which reflect the behavior patterns and preferences of similar users. These potential features are combined with the original feature vector of the target user (including its purchase history, risk assessment, personal information, etc.) to form a preliminary feature vector of the target user.

[0214] Use latent feature vectors extracted from similar user groups (obtained through a deep learning model), and the original feature vector u of the target user i To merge, the formula is as follows:

[0215] in, is the potential feature vector of similar users extracted by deep learning; u u is the original feature vector of the target user; α is the weight parameter that controls the ratio of potential features to original features.

[0216] The merging process combines the latent features with the original feature vector in a simple linear weighted manner to generate a preliminary user feature vector that can better reflect the user's potential needs.

[0217] The key features of non-car insurance products (such as price, historical sales data, etc.) are one of the main factors that determine product recommendations. In order to generate an enhanced feature vector, the system needs to fuse product features with user features. This step combines the preliminary feature vector of the target user with the key feature vector of the non-car product to form a comprehensive feature vector.

[0218] The system converts the key feature vector of the product The initial feature vector of the user Fusion is performed to generate a comprehensive feature vector:

[0219] in, It is the key feature vector extracted from the vehicle-attached non-vehicle product data, representing the product’s price, function, coverage, and other information; It is the user's preliminary feature vector, which contains information such as user preferences, behavior patterns, and purchase history. By combining user features and product features through vector inner product, the system generates a comprehensive feature vector This vector contains the matching information between user needs and product features.

[0220] The enhanced feature vector is the optimized user feature vector generated by the system. It combines the original features of the user, the potential features of similar users, and the key features of the product. The enhanced feature vector can be used to make accurate recommendations for users on vehicle-related and non-vehicle products.

[0221] Comprehensive feature vector Further processing is performed through the feature fusion module to generate the final enhanced feature vector The feature fusion module can use deep learning technology to further extract and optimize the feature vector. The formula is as follows:

[0222] It is a feature fusion function, which can be an activation function of a deep neural network, used to extract and optimize the comprehensive feature vector of the input. It can be achieved through a deep neural network (such as a multi-layer perceptron) or other feature fusion techniques (such as an autoencoder) to further refine the user's deep needs and purchase intentions, thereby generating an optimized enhanced feature vector.

[0223] In this embodiment, by combining the original features of the user, the potential features of similar user groups, and the key features of the non-vehicle products, the system can generate a highly accurate enhanced feature vector. This vector not only contains the user's historical behavior and preferences, but also combines the user's potential needs and product features, so that the recommendation system can more accurately recommend the non-vehicle insurance products that best meet the user's needs. The generation of enhanced feature vectors significantly improves the accuracy of the recommendation system and user satisfaction.

[0224] In one embodiment, the above S80 includes:

[0225] S801, using enhanced feature vectors to perform matching scores on all non-vehicle products in the insurance product database;

[0226] S802, collecting insurance product purchase history data of similar users from the similar user group;

[0227] S803, analyzing the purchase history of similar users, identifying the product categories with the highest purchase frequency and specific products that have been purchased multiple times in the similar user group;

[0228] S804, comprehensively matching the scores, the product categories with the highest purchase frequency, and the specific products that have been purchased multiple times, and screening out the non-car-related products that are most suitable for the target user;

[0229] S805, output and display the most suitable vehicle-related non-vehicle products to the target user.

[0230] In this embodiment, the enhanced feature vector has combined the user's historical behavior and potential needs, and also integrated the key features of the product. At this time, the system matches the enhanced feature vector with the feature vector of the non-vehicle product and calculates the matching score between the two. The matching score can be achieved by the inner product between the enhanced feature vectors, and the formula is as follows: The score reflects the user's potential interest in the product. The higher the score, the more the product meets the user's needs.

[0231] The system needs to collect the insurance product purchase history of similar users from the previously identified similar user groups. This data provides the behavior patterns of similar users and helps further determine which products are suitable for target users. The system will retrieve the purchase history records in the similar user groups, obtain the list of products purchased by each similar user, and record their purchase preference data for each product (such as the number of purchases, comments, etc.).

[0232] The purchase history of similar users provides a valuable reference for the system. By analyzing the purchase frequency and product categories of similar users, the system can identify which product categories are most popular among similar users, and also identify which products are purchased multiple times by similar users. The system can count the number of times similar users purchase each category of products and sort them. For example, it can identify the product categories with the highest purchase frequency and the specific products in this category that have been purchased multiple times. These products usually have a higher recommendation value.

[0233] The system will combine the matching score with the purchase history of similar users to select the most suitable products for the target user from all products. The matching score measures the matching degree between the user and the product, while the purchase behavior of similar users provides a strong reference for recommendation. Combining these two indicators, the system selects the best non-car products through a sorting algorithm. These products not only have a high score in the matching score, but also have a high purchase frequency among similar user groups.

[0234] Finally, the system will generate a recommendation list and display these recommended products to the target users. The display form may include pushing to users through APP, website, or email. The system will output the recommended list of filtered vehicle-mounted and non-vehicle products, and users can view these products through the interactive interface and make further purchase decisions.

[0235] By combining the enhanced feature vector and the purchase history of similar users, the system can more accurately recommend vehicle and non-vehicle insurance products to users. The matching score provides the system with a quantitative user-product matching degree, while the behavior data of similar users enhances the credibility of the recommendation.

[0236] In one embodiment, a device for outputting a non-vehicle product accompanying a vehicle is provided, and the device for outputting a non-vehicle product accompanying a vehicle corresponds one-to-one to the method for outputting a non-vehicle product accompanying a vehicle in the above embodiment. Figure 3 , Figure 3 This is a functional module diagram of a preferred embodiment of the vehicle-mounted non-vehicle product output device of the present invention. The voice recognition module 10, the user data collection module 20, the original feature extraction module 30, the user group classification module 40, the potential feature extraction module 50, the product feature extraction module 60, the feature fusion module 70 and the product output module 80. The detailed description of each functional module is as follows:

[0237] A speech recognition module 10, used to obtain speech data related to insurance products from each user and convert the speech data into speech text;

[0238] A user data collection module 20 is used to obtain each user's insurance product purchase history and collect each user's personal information and risk assessment data;

[0239] The original feature extraction module 30 is used to analyze each user's voice text, insurance product purchase history, personal information and risk assessment data to form an original feature vector for each user;

[0240] A user group division module 40 is used to analyze the similarity between the original feature vector of the target user and the original feature vectors of other users, and determine a similar user group of the target user;

[0241] A potential feature extraction module 50, used to extract potential features from the similar user group, wherein the potential features include behavior pattern features, preference features and common features;

[0242] A product feature extraction module 60, for extracting key product features from the vehicle-attached non-vehicle product data, wherein the key product features include price and historical sales;

[0243] A feature fusion module 70 is used to combine the original feature vector, potential features and product key features to generate an enhanced feature vector;

[0244] The product output module 80 is used to determine and output the most suitable non-vehicle accompanying product for the target user based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group.

[0245] In one embodiment, the speech recognition module 10 is specifically used for:

[0246] Conduct voice interactions with users related to insurance products through the communication platform and collect users’ voice data;

[0247] Converting the speech data into speech text using an automatic speech recognition module;

[0248] Performing a preprocessing operation on the speech text, wherein the preprocessing operation includes removing noise, correcting speech recognition errors, and standardizing text format;

[0249] The pre-processed speech text is associated with the corresponding user through the user's unique identifier.

[0250] In one embodiment, the original feature extraction module 30 is specifically used for:

[0251] Process and analyze each user's voice text to extract insurance demand characteristics from the voice text;

[0252] Analyze each user's insurance product purchase history and extract the user's purchase preference characteristics;

[0253] Analyze each user's personal information and extract the user's personal economic characteristics;

[0254] Analyze each user’s risk assessment data and extract the user’s risk profile characteristics;

[0255] The insurance demand characteristics, purchase preference characteristics, personal economic characteristics and risk status characteristics are combined into an original feature vector for each user.

[0256] In one embodiment, the user group classification module 40 is specifically used to:

[0257] Compare the target user's original feature vector with all other users' original feature vectors one by one;

[0258] Analyzing the similarity scores between the target user and other users using a similarity measurement module;

[0259] According to a preset similarity threshold, other users whose similarity scores exceed the similarity threshold are screened out as similar users of the target user;

[0260] The screened similar users are organized into similar user groups.

[0261] In one embodiment, the potential feature extraction module 50 is specifically used to:

[0262] Collect behavioral data of each similar user in the similar user group, including purchase frequency and preference change information;

[0263] Use deep learning models to analyze the behavioral data of similar user groups and extract hidden behavioral pattern features and preference characteristics;

[0264] Identify common features of similar users in a similar user group, wherein the common features include common purchasing habits or preference trends;

[0265] Standardize the extracted behavior pattern features, preference features and common features;

[0266] The standardized behavior pattern features, preference features and common features are integrated into a complete potential feature vector.

[0267] In one embodiment, the feature fusion module 70 is specifically configured to:

[0268] Merge the latent feature vector extracted from the similar user group with the original feature vector of the target user to form a preliminary merged user feature vector;

[0269] Merging the product key feature vector with the initially merged user feature vector to form a comprehensive feature vector;

[0270] An enhanced feature vector is generated based on the comprehensive feature vector through a feature fusion module.

[0271] In one embodiment, the product output module 80 is specifically used for:

[0272] Use enhanced feature vectors to match and score all non-car products in the insurance product database;

[0273] Collecting insurance product purchase history data of similar users from the similar user group;

[0274] Analyze the purchase history of similar users to identify the product categories with the highest purchase frequency and specific products with multiple purchases among the similar user groups;

[0275] Comprehensive matching scores, product categories with the highest purchase frequency, and specific products with multiple purchases are used to select the most suitable non-car-related products for the target user;

[0276] Output and display the most suitable vehicle-related and non-vehicle products to the target users.

[0277] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external user terminal through a network connection. When the computer program is executed by the processor, it realizes the functions or steps of a service-side method for outputting non-vehicle products with the vehicle.

[0278] In one embodiment, a computer device is provided. The computer device may be a user terminal, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the user side of a method for outputting non-vehicle products with a vehicle

[0279] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0280] Analyze customers’ personal information, historical purchase records, and browsing behaviors, build customer profiles, and extract customer profile features including age, income level, and risk tolerance;

[0281] Extracting insurance features of each insurance product from the non-auto property insurance product database, wherein the insurance features include suitable population categories, insurance identification values, and coverage;

[0282] Initialize particles in the particle swarm optimization method and randomly generate multiple particles, each particle represents a non-auto property insurance combination plan;

[0283] Compare the insurance characteristics of the non-auto property insurance combination plan in each particle with the customer portrait characteristics, analyze the matching degree of each non-auto property insurance combination plan, and determine the initial individual optimal position of each particle based on the matching degree analysis result;

[0284] In each iteration of the particle swarm optimization method, the speed and position of each particle in the particle swarm are updated, and the matching degree analysis is re-performed for each updated particle position;

[0285] According to the re-determined matching analysis results, the individual optimal position of each particle and the global optimal position of the particle group are updated;

[0286] After the iteration is completed, the non-auto property insurance combination plan corresponding to the global optimal position of the particle swarm is selected as the final output.

[0287] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0288] Analyze customers’ personal information, historical purchase records, and browsing behaviors, build customer profiles, and extract customer profile features including age, income level, and risk tolerance;

[0289] Extracting insurance features of each insurance product from the non-auto property insurance product database, wherein the insurance features include suitable population categories, insurance identification values, and coverage;

[0290] Initialize particles in the particle swarm optimization method and randomly generate multiple particles, each particle represents a non-auto property insurance combination plan;

[0291] Compare the insurance characteristics of the non-auto property insurance combination plan in each particle with the customer portrait characteristics, analyze the matching degree of each non-auto property insurance combination plan, and determine the initial individual optimal position of each particle based on the matching degree analysis result;

[0292] In each iteration of the particle swarm optimization method, the speed and position of each particle in the particle swarm are updated, and the matching degree analysis is re-performed for each updated particle position;

[0293] According to the re-determined matching analysis results, the individual optimal position of each particle and the global optimal position of the particle group are updated;

[0294] After the iteration is completed, the non-auto property insurance combination plan corresponding to the global optimal position of the particle swarm is selected as the final output.

[0295] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the user side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0296] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0297] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0298] It should be noted that if software tools or components other than those of the Company appear in the embodiments of the present application, they are only used for illustration and do not represent actual use. The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the above-mentioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the above-mentioned embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for outputting non-vehicle products attached to a vehicle, characterized in that: The following steps are involved: Acquire voice data related to insurance products for each user, and convert the voice data into voice text; Obtain each user's insurance product purchase history and collect each user's personal information and risk assessment data; Analyze each user’s voice text, insurance product purchase history, personal information, and risk assessment data to form each user’s original feature vector; Analyze the similarity between the original feature vector of the target user and the original feature vectors of other users to determine a similar user group of the target user; Extracting potential features from the similar user group, wherein the potential features include behavior pattern features, preference features and common features; Extracting key product features from the vehicle-attached non-vehicle product data, wherein the key product features include price and historical sales; Combining the original feature vector, potential features and product key features to generate an enhanced feature vector; Based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group, the vehicle-related non-vehicle product that best suits the target user is determined and output.

2. The method for outputting non-vehicle products attached to a vehicle according to claim 1, characterized in that: Acquire voice data related to insurance products for each user and convert the voice data into voice text, including: Conduct voice interactions with users related to insurance products through the communication platform and collect users’ voice data; Converting the speech data into speech text using an automatic speech recognition module; Performing a preprocessing operation on the speech text, wherein the preprocessing operation includes removing noise, correcting speech recognition errors, and standardizing text format; The pre-processed speech text is associated with the corresponding user through the user's unique identifier.

3. The method for outputting non-vehicle products attached to a vehicle according to claim 1, characterized in that: Analyze each user’s voice text, insurance product purchase history, personal information, and risk assessment data to form each user’s original feature vector, including: Process and analyze each user's voice text to extract insurance demand characteristics from the voice text; Analyze each user's insurance product purchase history and extract the user's purchase preference characteristics; Analyze each user's personal information and extract the user's personal economic characteristics; Analyze each user’s risk assessment data and extract the user’s risk profile characteristics; The insurance demand characteristics, purchase preference characteristics, personal economic characteristics and risk status characteristics are combined into an original feature vector for each user.

4. The method for outputting non-vehicle products attached to a vehicle according to claim 1, characterized in that: Analyzing the similarity between the original feature vector of the target user and the original feature vectors of other users to determine a similar user group of the target user includes: Compare the target user's original feature vector with all other users' original feature vectors one by one; Analyzing the similarity scores between the target user and other users using a similarity measurement module; According to a preset similarity threshold, other users whose similarity scores exceed the similarity threshold are screened out as similar users of the target user; The screened similar users are organized into similar user groups.

5. The method for outputting non-vehicle products attached to a vehicle according to claim 1, characterized in that: Extracting potential features from the similar user group, the potential features include behavior pattern features, preference features and common features, including: Collect behavioral data of each similar user in the similar user group, including purchase frequency and preference change information; Use deep learning models to analyze the behavioral data of similar user groups and extract hidden behavioral pattern features and preference characteristics; Identify common features of similar users in a similar user group, wherein the common features include common purchasing habits or preference trends; Standardize the extracted behavior pattern features, preference features and common features; The standardized behavior pattern features, preference features and common features are integrated into a complete potential feature vector.

6. The method for outputting non-vehicle products attached to a vehicle as claimed in claim 1, characterized in that: The original feature vector, potential features and product key features are combined to generate an enhanced feature vector, including: Merge the latent feature vector extracted from the similar user group with the original feature vector of the target user to form a preliminary merged user feature vector; Merging the product key feature vector with the initially merged user feature vector to form a comprehensive feature vector; An enhanced feature vector is generated based on the comprehensive feature vector through a feature fusion module.

7. The method for outputting non-vehicle products attached to a vehicle according to claim 1, characterized in that: Based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group, determining and outputting the vehicle-related non-vehicle product that best suits the target user, including: Use enhanced feature vectors to match and score all non-car products in the insurance product database; Collecting insurance product purchase history data of similar users from the similar user group; Analyze the purchase history of similar users to identify the product categories with the highest purchase frequency and specific products with multiple purchases among the similar user groups; Comprehensive matching scores, product categories with the highest purchase frequency, and specific products with multiple purchases are used to select the most suitable non-car-related products for the target user; Output and display the most suitable vehicle-related and non-vehicle products to the target users.

8. A non-vehicle product output device, characterized in that: The vehicle-mounted non-vehicle product output device comprises: A speech recognition module, used to obtain speech data related to insurance products from each user and convert the speech data into speech text; A user data collection module is used to obtain each user's insurance product purchase history and collect each user's personal information and risk assessment data; The original feature extraction module is used to analyze each user's voice text, insurance product purchase history, personal information and risk assessment data to form the original feature vector of each user; A user group division module is used to analyze the similarity between the original feature vector of the target user and the original feature vectors of other users, and determine a similar user group of the target user; A potential feature extraction module, used to extract potential features from the similar user group, wherein the potential features include behavior pattern features, preference features and common features; A product feature extraction module is used to extract key product features from the vehicle-attached non-vehicle product data, wherein the key product features include price and historical sales; A feature fusion module, used to combine the original feature vector, potential features and product key features to generate an enhanced feature vector; The product output module is used to determine and output the most suitable non-vehicle accompanying product for the target user based on the enhanced feature vector and the insurance product purchase history of similar users in the similar user group.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a vehicle-related non-vehicle product output program stored in the memory and executable on the processor. When the vehicle-related non-vehicle product output program is executed by the processor, the steps of the vehicle-related non-vehicle product output method as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a vehicle-attached non-vehicle product output program, and when the vehicle-attached non-vehicle product output program is executed by the processor, the steps of the vehicle-attached non-vehicle product output method according to any one of claims 1 to 7 are implemented.

Images