Method and device for generating insurance recommendation text and related equipment

Abstract

The embodiment of the invention provides an insurance recommendation text generation method and device and related equipment, and belongs to the technical field of large language models. The method comprises the following steps: acquiring historical claim settlement data and external event data of a target object, and constructing multi-source fusion data according to the historical claim settlement data and the external event data; inputting the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates an original recommendation text according to the multi-source fusion data; comparing the multi-source fusion data with the original recommended text to obtain deviation data; inputting the deviation data and the original recommended text into a text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text; and pushing the target recommendation text to the target object. According to the embodiment of the invention, the highly personalized insurance recommendation text can be generated.

Description

Technical Field

[0001] The present invention relates to the technical field of large language models, and in particular to a method and device for generating insurance recommendation texts and related equipment. Background Art

[0002] In the insurance field, the recommendation text of the insurance industry is based on the customer's data and needs, and recommends insurance business to customers through targeted communication. The content of the recommendation text may include the characteristics, advantages, applicable scenarios and how to meet the specific protection needs of customers. The generation of the recommendation text depends on the collected customer information, such as age, gender, occupation, existing insurance products, etc. With this information, insurance companies can design recommendation texts that meet customer expectations and needs to improve the attractiveness of insurance products and sales conversion rates.

[0003] In the related art, traditional recommendation text generation systems often rely on preset static templates and rules, which limits the personalization of the text and makes it difficult to meet the diverse and dynamically changing needs of customers. For example, it is difficult to generate refined personalized recommendation texts for different customers with different claims histories. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and device for generating an insurance recommendation text and related equipment, aiming to generate a highly personalized insurance recommendation text.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for generating an insurance recommendation text, the method comprising:

[0006] Acquire historical claims data and external event data of the target object, and construct multi-source fusion data based on the historical claims data and the external event data;

[0007] Inputting the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data;

[0008] Comparing the multi-source fusion data with the original recommendation text to obtain deviation data;

[0009] Inputting the deviation data and the original recommended text into the text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text;

[0010] The target recommendation text is pushed to the target object.

[0011] In some embodiments, the external event data includes real-time weather data and news data, and the acquiring of historical claims data and external event data of the target object, and constructing multi-source fusion data according to the historical claims data and the external event data, includes:

[0012] Acquire the historical claims data of the target object; wherein the historical claims data includes the historical number of claims, historical claims amount and historical claims type of the target object;

[0013] Acquire real-time weather data within a future preset time period from at least one weather data source;

[0014] Obtaining hot news data and natural disaster loss data caused by historical natural disasters from at least one news data source;

[0015] Multi-source fusion data is constructed according to the historical claims data, the real-time weather data, the natural disaster loss data and the hot news data.

[0016] In some embodiments, constructing multi-source fusion data according to the historical claims data, the real-time weather data, the natural disaster loss data, and the hot news data includes:

[0017] Extracting extreme weather features from the real-time weather data to obtain key weather features;

[0018] Calculate the claim frequency and average claim amount of the target object according to the historical number of claims and the historical claim amount of the target object;

[0019] Performing natural language processing on the hot news data to obtain hot keywords and corresponding event description data;

[0020] Multi-source fusion data is constructed according to the key weather characteristics, the claim frequency, the average claim amount, the natural disaster loss data, the hot keywords and the corresponding event description data.

[0021] In some embodiments, constructing multi-source fusion data according to the historical claims data, the real-time weather data, the natural disaster loss data, and the hot news data includes:

[0022] Generate historical claims data weights based on the historical claims data;

[0023] Generating real-time weather data weights according to the real-time weather data;

[0024] generating natural disaster loss weights according to the natural disaster loss data;

[0025] Generate hot news weights according to the hot news data;

[0026] The multi-source fusion data is constructed according to the historical claims data, the historical claims data weight, the real-time weather data, the real-time weather data weight, the natural disaster loss data, the natural disaster loss weight, the hot news data and the hot news weight.

[0027] In some embodiments, before constructing the multi-source fusion data according to the historical claims data, the real-time weather data, the natural disaster loss data, and the hot news data, the method further includes:

[0028] For the historical claim data, in response to finding that the claim amount in the historical claim data is missing, performing data filling processing on the claim amount;

[0029] For the real-time weather data, in response to retrieving abnormal weather data in the real-time weather data, performing abnormal processing on the abnormal weather data based on a preset weather data threshold;

[0030] The news data is deduplicated.

[0031] In some embodiments, the multi-source fusion data is compared with the original recommendation text to obtain deviation data, including

[0032] Extracting key data points from the original recommendation text;

[0033] The key data points are compared with the corresponding original data points in the historical claims data and the external event data, and all the difference data points are marked. All the difference data points constitute the deviation data.

[0034] In some embodiments, inputting the deviation data and the original recommended text into the text generation model so that the text generation model corrects the original recommended text to obtain the target recommended text includes:

[0035] The original recommended text and the deviation data marking all the difference data points are input into the text generation model together, so that the text generation model corrects the original recommended text according to each difference data point and refers to the corresponding original data point to obtain the target recommended text.

[0036] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a device for generating an insurance recommendation text, the device comprising:

[0037] An acquisition module, used to acquire historical claims data and external event data of a target object, and construct multi-source fusion data based on the historical claims data and the external event data;

[0038] A first input module, used for inputting the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data;

[0039] A comparison module, used to compare the multi-source fusion data with the original recommendation text to obtain deviation data;

[0040] A second generation module, configured to input the deviation data and the original recommended text into the text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text;

[0041] A push module is used to push the target recommendation text to the target object.

[0042] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0043] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0044] The method and device for generating insurance recommendation text proposed in this application obtains the historical claims data and external event data of the target object, and constructs multi-source fusion data based on the historical claims data and external event data; inputs the multi-source fusion data into a pre-trained text generation model so that the text generation model generates the original recommendation text based on the multi-source fusion data; compares the multi-source fusion data with the original recommendation text to obtain deviation data; inputs the deviation data and the original recommendation text into the text generation model so that the text generation model corrects the original recommendation text to obtain the target recommendation text; and pushes the target recommendation text to the target object. This application first constructs a multi-source fusion data set by acquiring the historical claims data and external event data of the target object. This multi-source fusion data set not only takes into account the customer's personal historical data, but also integrates external factors that may affect insurance demand, such as weather changes, social events, etc., thereby providing a rich data foundation for generating personalized recommendation texts. Next, the multi-source fusion data is input into a pre-trained text generation model, and the reasoning and text generation capabilities of the large language model are used to generate the original recommendation text based on these data. However, since the model may have certain deviations, the multi-source fusion data is also compared with the original recommendation text to obtain deviation data. This comparison process can identify inaccurate or irrelevant content that may exist in the original recommendation text, thereby providing a basis for correction. Then, the deviation data and the original recommendation text are input into the text generation model again, and the model corrects the original recommendation text based on this information to generate the final target recommendation text. This correction process ensures the accuracy and applicability of the recommendation text and improves the effectiveness of the recommendation. Finally, the corrected target recommendation text is pushed to the target object. By fusing multi-source data and a large language model, the present application can generate highly personalized insurance recommendation texts that dynamically match customer needs and external events. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of a method for generating an insurance recommendation text provided in an embodiment of the present application;

[0046] Figure 2 yes Figure 1 Flow chart of step S101 in FIG.

[0047] Figure 3 yes Figure 2 Flow chart of step S204 in FIG.

[0048] Figure 4 yes Figure 2 Another flow chart of step S204 in;

[0049] Figure 5 yes Figure 2 The flowchart before step S204 in FIG.

[0050] Figure 6 yes Figure 1 Flow chart of step S103 in FIG.

[0051] Figure 7 yes Figure 1 Flow chart of step S104 in FIG.

[0052] Figure 8 It is a structural schematic diagram of a device for generating an insurance recommendation text provided in an embodiment of the present application;

[0053] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0057] First, some nouns involved in this application are analyzed:

[0058] Artificial Intelligence (AI): It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science. AI attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0059] Natural Language Processing (NLP): NLP uses computers to process, understand, and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is commonly used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis, and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0060] Large Language Model (LLM): It is a language model in the field of natural language processing (NLP). It uses deep learning methods to train on massive text data to learn the statistical laws and knowledge of language, and has multiple natural language processing capabilities such as text generation, question answering, summarization, translation, and reasoning. Compared with traditional language models, large language models have the characteristics of large number of parameters (hundreds of millions to hundreds of billions), rich training corpus (hundreds of GB to hundreds of TB), and wide range of adaptability tasks. The emergence of large language models has greatly promoted the progress of artificial intelligence in natural language understanding and generation, and represents the cutting-edge level of current natural language processing technology. Through training on massive text data, the model learns the high-dimensional semantic representation of language and establishes the connection between words, phrases, and sentences. This semantic representation can be mapped to various downstream natural language processing tasks, making large language models have a wide range of adaptability.

[0061] Based on this, the embodiments of the present application provide a method and apparatus for generating an insurance recommendation text and related equipment, aiming to generate a highly personalized insurance recommendation text.

[0062] The method and device for generating the insurance recommendation text and related equipment provided in the embodiments of the present application are specifically described through the following embodiments. First, the recommendation method in the embodiments of the present application is described.

[0063] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0064] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0065] The method for generating insurance recommendation text provided in the embodiment of the present application relates to the technical field of large language models. The method for generating insurance recommendation text provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the method for generating insurance recommendation text, etc., but is not limited to the above forms.

[0066] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0067] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0068] Figure 1 is an optional flowchart of the method for generating an insurance recommendation text provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S105.

[0069] Step S101, obtaining historical claims data and external event data of the target object, and constructing multi-source fusion data based on the historical claims data and external event data.

[0070] Step S102: input the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data.

[0071] Step S103, comparing the multi-source fusion data with the original recommendation text to obtain deviation data.

[0072] Step S104: input the deviation data and the original recommended text into a text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text.

[0073] Step S105: Push the target recommendation text to the target object.

[0074] Steps S101 to S106 shown in the embodiment of the present application, by acquiring the historical claims data and external event data of the target object, construct a multi-source fusion data set. This multi-source fusion data set not only takes into account the personal historical data of the customer, but also integrates external factors that may affect insurance demand, such as weather changes, social events, etc., thereby providing a rich data basis for generating personalized recommendation texts. Next, the multi-source fusion data is input into a pre-trained text generation model, and the reasoning and text generation capabilities of the large language model are used to generate the original recommendation text based on these data. However, since the model may have certain deviations, the multi-source fusion data is also compared with the original recommendation text to obtain deviation data. This comparison process can identify inaccurate or irrelevant content that may exist in the original recommendation text, thereby providing a basis for correction. Then, the deviation data and the original recommendation text are input into the text generation model again, and the model corrects the original recommendation text based on this information to generate the final target recommendation text. This correction process ensures the accuracy and applicability of the recommendation text and improves the effectiveness of the recommendation. Finally, the corrected target recommendation text is pushed to the target object. By integrating multi-source data and large language models, it is possible to generate highly personalized insurance recommendation texts that dynamically match customer needs and external events.

[0075] In step S101 of some embodiments, historical claims data of the target object may be collected first, including but not limited to the number of claims, claim amount, claim type and other information. These data can reflect the customer's previous insurance needs and claims tendencies, and provide an important basis for generating personalized recommendation texts. At the same time, external event data related to insurance needs may also be collected, such as weather changes, natural disasters, social news, etc., which may affect the customer's insurance needs. After the collection is completed, these data are fused to construct a multi-source fusion data set. This data set combines the customer's personal historical data and external environment data, providing rich contextual information for subsequent text generation.

[0076] See also Figure 2 In some embodiments, step S101 may include but is not limited to steps S201 to S204:

[0077] Step S201, obtaining historical claims data of the target object.

[0078] Step S202: acquiring real-time weather data within a future preset time period from at least one weather data source.

[0079] Step S203: acquiring hot news data and natural disaster loss data caused by historical natural disasters from at least one news data source.

[0080] Step S204, constructing multi-source fusion data based on historical claims data, real-time weather data, natural disaster loss data, and hot news data.

[0081] In step S201 of some embodiments, historical claims data is a record of historical interactions between customers and insurance companies. This information helps the text generation model understand the customer's insurance needs and claims tendencies. Specifically, historical claims data includes the target object's historical number of claims, historical claim amounts, and historical claim types. The number of claims can reflect the frequency of unexpected events encountered by the customer, the claim amount can indicate the severity of the event, and the claim type reveals the type of risk that the customer may face. These data are crucial for generating targeted insurance recommendation texts.

[0082] In step S202 of some embodiments, real-time weather data is obtained from at least one weather data source, and the data covers a preset time period in the future. Weather conditions are an important factor affecting insurance demand, especially in areas where natural disasters are frequent. Real-time weather data includes temperature, precipitation, wind speed, etc., as well as predictions of specific weather events, such as typhoons, heavy rains, floods, etc. These data are helpful in predicting possible future events, thereby providing timely insurance advice to customers.

[0083] In some embodiments, weather conditions within the next week can be dynamically acquired, with particular emphasis on monitoring and forecasting of major weather events. Weather data is acquired in real time through an interface connection with at least one weather data source, which may be a weather service provider or other reliable weather information provider. Not only conventional weather data such as temperature, humidity, wind speed and direction are collected, but also special attention is paid to major weather events that may cause changes in insurance demand. These events include, but are not limited to, extreme weather conditions such as typhoons, heavy rains, blizzards, hail, tornadoes, etc. By monitoring these data in real time, the system is able to predict and identify high-risk weather events that may affect customers.

[0084] In step S203 of some embodiments, hot news data may be obtained from data sources such as news websites, social media platforms, news subscription services, or other news aggregators. Such data may include current news, economic trends, public safety warnings, etc., which are essential for understanding the risks and concerns that customers may currently face. Hot news data provides a macro perspective of the socio-economic environment, which helps predict and evaluate external factors that may affect insurance demand.

[0085] Secondly, we will also collect data on losses caused by historical natural disasters. These data record the economic losses, casualties, and property damage caused by natural disasters in the past. By analyzing these historical data, we can identify the impact of natural disasters in specific regions or specific types on insurance demand, as well as the type of insurance that customers may need.

[0086] In some embodiments, natural language processing (NLP) technology can be used to analyze and extract key information and trends in news data. NLP technology can help the system identify keywords, topics, and emotional colors in news reports, so as to more accurately understand the essence and impact of news content.

[0087] In step S204 of some embodiments, the data collected in steps S201, S202, and S203 are integrated to construct a multi-source fusion data set. This data set integrates the client's personal history data and external environment data, providing comprehensive contextual information for generating personalized insurance recommendation texts. The construction of multi-source fusion data is a key step, which ensures that the recommendation text is not only based on the client's personal history, but also takes into account current environmental factors and social events that may affect insurance demand. In this way, the generated recommendation text can more accurately reflect the actual needs of the client and improve the relevance and effectiveness of the recommendation text.

[0088] See also Figure 3 In some embodiments, step S204 may include but is not limited to steps S301 to S304:

[0089] Step S301, extract extreme weather features from real-time weather data to obtain key weather features.

[0090] Step S302, calculating the claim frequency and average claim amount of the target object according to the historical number of claims and historical claim amounts of the target object.

[0091] Step S303: perform natural language processing on the hot news data to obtain hot keywords and corresponding event description data.

[0092] Step S304, constructing multi-source fusion data based on key weather characteristics, claim frequency, average claim amount, natural disaster loss data, hot keywords and corresponding event description data.

[0093] In step S301 of some embodiments, real-time weather data obtained from at least one weather data source is analyzed to identify and extract features of extreme weather events. These features may include the path, intensity, and estimated arrival time of a typhoon, the precipitation and duration of a rainstorm, and other weather conditions that may affect insurance demand. Through in-depth analysis of real-time weather data, key weather features can be extracted to provide a basis for generating personalized insurance recommendation texts.

[0094] In step S302 of some embodiments, the historical claims data of the target object is analyzed, including the number of claims and the amount of claims. Based on these data, the claim frequency of the target object, that is, the number of claims in a specific time period, and the average claim amount, that is, the average of all claim amounts, are obtained. These indicators help to assess the risk level and insurance needs of the target object, and provide a basis for generating personalized insurance recommendation texts.

[0095] In step S303 of some embodiments, natural language processing (NLP) may be performed on the hot news data obtained from the news data source. NLP technology is used to identify hot keywords and phrases in the news text, and extract event descriptions related to these keywords. These keywords and descriptions provide insights into the current concerns and important events of society, and help generate insurance recommendation texts related to the customer's current environment and concerns.

[0096] In step S304 of some embodiments, the data extracted and calculated in steps S301 to S303 are integrated to construct a multi-source fusion data set. This data set includes key weather characteristics, the frequency and average amount of claims of the target object, historical natural disaster loss data, and keywords and event descriptions in hot news. Through this fusion of multi-source data, a comprehensive context information can be generated to provide rich input data for the subsequent text generation model.

[0097] Through steps S301 to S304, a multi-source fusion dataset is constructed that integrates customer personal history, real-time weather conditions, social hot events, and historical natural disaster data. This dataset provides a solid foundation for generating personalized, timely, and persuasive insurance recommendation texts.

[0098] See also Figure 4 In some embodiments, step S204 may include but is not limited to steps S401 to S405:

[0099] Step S401, generating historical claim data weights based on historical claim data.

[0100] Step S402: Generate real-time weather data weights based on the real-time weather data.

[0101] Step S403: Generate natural disaster loss weights based on natural disaster loss data.

[0102] Step S404: generating hot news weights according to the hot news data.

[0103] Step S405, constructing multi-source fusion data based on historical claims data, historical claims data weights, real-time weather data, real-time weather data weights, natural disaster loss data, natural disaster loss weights, hot news data and hot news weights.

[0104] In step S401 of some embodiments, weights may be assigned to historical claims data according to the size of the claims amount, the frequency of claims, or other relevant factors. For example, if a customer has multiple claims records in the past year and the claims amount is high, a higher weight may be assigned to the customer's historical claims data, indicating that the historical claims data has a higher influence in generating the recommendation text.

[0105] In step S402 of some embodiments, a weight may be assigned to the real-time weather data according to the severity, urgency or other relevant factors of the weather event. For example, if a typhoon is predicted to land within the next week, and the intensity and path of the typhoon coincide with the area where the client is located, a higher weight will be assigned to the real-time weather data.

[0106] In step S403 of some embodiments, the loss data caused by historical natural disasters are analyzed, and weights are assigned to the natural disaster loss data according to the severity of the disaster, the size of the economic loss, or other relevant factors. For example, if the historical data shows that the area where customer A is located has been hit by a major typhoon, causing serious economic losses, a higher weight will be assigned to the natural disaster loss data. For example, if the historical data shows that the area where customer A is located has been hit by two major typhoons in the past five years, with a total economic loss of 50 million yuan, a high weight will be assigned to the natural disaster loss data.

[0107] In step S404 of some embodiments, weights may be assigned to hot news data based on the attention, influence, or other relevant factors of the news event. For example, if a news event is widely discussed or frequently reported on social media and is related to insurance, a higher weight will be assigned to the hot news data. For example, if the news currently widely discussed on social media is about new traffic regulations that are about to be implemented, which may affect the demand for auto insurance, the weight corresponding to the news will be increased.

[0108] In step S405 of some embodiments, the weights generated in steps S401 to S404 are combined with the corresponding data to construct a multi-source fusion data set. This data set includes the customer's historical claims data, historical compensation data and its weights, real-time weather data and its weights, natural disaster loss data and its weights, and hot news data and its weights. In this way, the importance and urgency of different data sources can be comprehensively considered to generate a comprehensive data set.

[0109] See also Figure 5 In some embodiments, before step S204, the following steps may also be performed, including but not limited to steps S501 to S503:

[0110] Step S501, for historical claim data, in response to finding that the claim amount in the historical claim data is missing, data filling processing is performed on the claim amount.

[0111] Step S502: for the real-time weather data, in response to finding that there is abnormal weather data in the real-time weather data, performing abnormal processing on the abnormal weather data based on a preset weather data threshold.

[0112] Step S503, deduplication processing is performed on the news data.

[0113] In step S501 of some embodiments, missing data is a common problem in data processing, which may be caused by incomplete records or data transmission errors. In order to solve this problem, data filling processing methods such as mean filling method, interpolation method or inference method based on other features can be used to fill in missing values. For example, if the customer's claim record lacks amount information, the average claim amount of the customer's previous claims or the average claim amount of similar customers can be used to fill it. This processing ensures the integrity of the data and provides an accurate data basis for subsequent analysis. For example, in the historical claims data of customer B, there is a record that lacks the claim amount. By analyzing the claims records of customer B in the past year, the average claim amount is calculated to be 2,000 yuan, and this value is used to fill the missing claim amount.

[0114] In step S502 of some embodiments, outlier detection and processing are performed on real-time weather data. Outliers may be caused by sensor failure, data input errors, or other reasons. In order to ensure the accuracy of the data, a reasonable threshold (such as the maximum possible rainfall is 1000mm) is set to mark and process these outliers. If it is found that the rainfall record on a certain day is an abnormal 5000mm, this value is marked as abnormal, and appropriate processing measures are taken, such as replacing or deleting the record with a historical average. This abnormal processing ensures the reliability of weather data and provides accurate data for subsequent weather analysis and prediction.

[0115] In step S503 of some embodiments, on social media and news platforms, the same news event may be recorded or reported multiple times, resulting in data duplication. In order to ensure the accuracy of data analysis, duplicate data records are identified and deleted by unique identifiers (such as news titles and timestamps). This deduplication process helps to reduce data redundancy and improve the efficiency of data processing. For example, data is collected from multiple news sources and it is found that three records all report the same earthquake event. By comparing the news titles and timestamps, it is identified that these three records are duplicated, and only one record is retained for subsequent analysis.

[0116] In some embodiments, the data also needs to be standardized. For example, different data sources may use different units to represent temperature (such as Celsius and Fahrenheit). Before analysis, all temperature data need to be converted into a unified unit (such as all converted to Celsius) to ensure data consistency. And in customer information, gender may be recorded as "male" or "female", but may be recorded as "1" or "0" in another data source. By uniformly coding (such as using "1" to represent male and "0" to represent female), it can be ensured that these data will not be ambiguous in analysis. In addition, the time formats of different data sources may be inconsistent, such as "YYYY-MM-DD" and "DD / MM / YYYY". Before analysis, all time data need to be converted into a unified format (such as ISO 8601 format: "YYYY-MM-DD") for subsequent processing.

[0117] Through the data cleaning and standardization processing of steps S501 to S503, the system ensures the accuracy, completeness and consistency of historical claims data, real-time weather data and news data.

[0118] Through steps S201 to S204, a comprehensive data set can be constructed, providing a rich information basis for subsequent text generation. These data include not only the customer's personal history, but also external factors that may affect insurance demand, making the generated recommendation text more personalized and persuasive.

[0119] In step S102 of some embodiments, the multi-source fusion data constructed in step S101 is input into a carefully trained text generation model. This text generation model is based on a large language model (LLM) architecture, such as Transformer-based models such as BERT, GPT, etc. These models have been pre-trained on a large amount of text data and have the ability to understand and generate natural language.

[0120] The pre-training process involves two main stages:

[0121] Fine-tuning phase: In this phase, the model is further trained to adapt to a specific task or domain, namely, insurance recommendation text generation in this scenario. To this end, the model is fed with specific customer data and external event data. Customer data may include historical claims records, customer feedback, customer preferences, etc., while external event data may cover weather reports, news events, market trends, etc. During the fine-tuning process, the model learns how to combine the customer's personal history with external events to generate recommendation text that meets the customer's current situation and needs. In this way, the model is able to understand the importance of specific data and learn how to transform this data into meaningful and persuasive text.

[0122] In order to make the model better generate effective text, a specific loss function can be used to measure the gap between the model generation results and the ideal results. Commonly used loss functions include cross-entropy loss for generative tasks, and BLEU or ROUGE-based indicators for evaluating the quality of text. In order to accelerate the convergence of the model, a dynamic learning rate adjustment strategy can also be used, such as gradually reducing the learning rate (Learning Rate Schedule) during training to avoid overfitting and improve the generalization ability of the model.

[0123] The pre-trained and fine-tuned text generation model receives multi-source fusion data as input. The complex algorithms and deep learning mechanisms inside the model will conduct in-depth analysis of this data, extract key information and features, and generate preliminary recommendation text based on this information. This original recommendation text is a direct response of the model to the input data. It tries to capture the key points in the data and transform it into fluent, relevant and attractive text.

[0124] In step S103 of some embodiments, the generated multi-source fusion data is carefully compared and analyzed with the original recommendation text generated by the text generation model. The purpose of this step is to identify and quantify any deviations between the original recommendation text and the actual data, which may be caused by various factors, including misunderstanding or omission of important information when the model generates the text. When comparing, each data and statement in the text will be checked in detail to ensure that it is consistent with the information in the multi-source fusion data set. For example, if the original recommendation text mentions the risk of a specific weather event, but the risk of the event is actually assessed to be low in the multi-source fusion data, the system will identify this deviation and record it as deviation data.

[0125] See also Figure 6 In some embodiments, step S103 may include but is not limited to steps S601 to S602:

[0126] Step S601: extract key data points from the original recommendation text.

[0127] Step S602, compare the key data points with the corresponding original data points in the historical claims data and the external event data, mark all the difference data points, and all the difference data points constitute the deviation data.

[0128] In step S601 of some embodiments, because the large language model (LLM) performs well in text generation, it may also produce inaccurate or inconsistent content, especially when it comes to specific values. Extract key data points from the original recommendation text generated by the text generation model. These data points may include but are not limited to the claim amount, number of claims, description of weather events, date, etc. This step usually involves natural language processing technology, which is used to identify and extract important information such as values ​​and proper nouns in the text.

[0129] In step S602 of some embodiments, the key data points extracted in step S601 are compared with the original data points in the historical claims data and the external event data. The purpose of the comparison is to identify and mark any inconsistencies, which are deviation data. This step ensures the accuracy and reliability of the recommendation text.

[0130] Exemplarily, a large language model is used to generate original recommendation text based on multi-source data provided by the customer. For example, if the customer's historical claims record shows two claims last year, with amounts of 50,000 yuan and 25,000 yuan respectively, the model may generate the following text: "Dear Mr. Li, according to our records, you have made two claims in the past year, with a total amount of 80,000 yuan. In order to ensure smooth claims in the future, we recommend that you upgrade your insurance plan." After the original text is generated, the original data comparison check will be automatically performed to extract all key data points (such as the number of claims, claim amounts, dates, weather events, etc.) from the initially generated marketing text, and compare this information with the original data, focusing on the accuracy of numbers, time, event descriptions, etc. For example, if the text generated by the model mentions that the claim amount is 80,000 yuan, and the original data is 75,000 yuan, this difference needs to be identified and marked, and then adjustments will be made based on the identified errors and the original text to generate the corrected final text. After correction, the final text generated by the model is: "Dear Mr. Li, according to our records, you have made two claims in the past year, with a total amount of 75,000 yuan. To ensure smooth claims in the future, we recommend that you upgrade your insurance plan." Such text is more accurate in content and data.

[0131] In step S104 of some embodiments, the deviation data identified in step S103 is input into the text generation model together with the original recommended text. The model uses this information to correct the original recommended text to generate a target recommended text that is more accurate and more consistent with the current data. The correction process involves the model analyzing the deviation data and adjusting the text content accordingly. The model will modify or completely rewrite certain parts of the text to ensure the accuracy and relevance of the final text.

[0132] See also Figure 7 In some embodiments, step S104 may include but is not limited to step S701:

[0133] Step S701, input the original recommended text and the deviation data marking all the difference data points into the text generation model, so that the text generation model corrects the original recommended text according to each difference data point and refers to the corresponding original data point to obtain the target recommended text.

[0134] In step S701 of some embodiments, the original recommended text and deviation data (including all identified difference data points) are input into the text generation model. The purpose of this step is to use the model to accurately correct the original text. The correction strategy includes:

[0135] Accurate citation: When it comes to numerical values, time and event descriptions, directly quote the original data to reduce errors in the generation process.

[0136] Contextual consistency: Ensure that the corrected text is semantically and contextually consistent with the original recommendation text while ensuring data accuracy.

[0137] Avoid duplicate corrections: When generating corrected text, avoid making multiple corrections on the same data point to ensure the fluency and consistency of the text.

[0138] Through step S701, accurate target recommendation texts can be generated, which are consistent with the original data points in terms of data.

[0139] In step S105 of some embodiments, the corrected target recommendation text is pushed to the target object, and the push can be completed through various channels, including but not limited to email, SMS, mobile application notification or direct display in the insurance company's customer relationship management (CRM) system.

[0140] The following is a specific example of generating insurance recommendation text:

[0141] Customer Background:

[0142] Age: 35

[0143] Gender: Male

[0144] Region: Coastal cities

[0145] Vehicle Type: SUV

[0146] Historical claims record: Last year, the vehicle was damaged by a typhoon and the claim amount was 100,000 yuan.

[0147] Real-time weather forecast:

[0148] There may be a typhoon passing through in the next week, and rainfall is expected to exceed 200mm.

[0149] Generate recommendation text:

[0150] Hello, Mr. Zhang. We have noticed that there may be a typhoon passing through in the next week. Last year, a similar typhoon caused serious damage to your vehicle, but fortunately you purchased our car insurance in time and received a claim of 100,000 yuan. This year's typhoon may bring more serious impacts. In order to ensure the safety of your vehicle and property, we recommend that you renew or upgrade your insurance plan as soon as possible. This can avoid unnecessary financial losses and ensure that you are more calm when facing natural disasters.

[0151] The embodiment of the present application first constructs a multi-source fusion data set by acquiring the historical claims data and external event data of the target object. This multi-source fusion data set not only considers the personal historical data of the customer, but also integrates external factors that may affect the insurance demand, such as weather changes, social events, etc., thereby providing a rich data basis for generating personalized recommendation texts. Next, the multi-source fusion data is input into a pre-trained text generation model, and the reasoning and text generation capabilities of the large language model are used to generate the original recommendation text based on these data. However, since the model may have certain deviations, the multi-source fusion data is also compared with the original recommendation text to obtain deviation data. This comparison process can identify inaccurate or irrelevant content that may exist in the original recommendation text, thereby providing a basis for correction. Then, the deviation data and the original recommendation text are input into the text generation model again, and the model corrects the original recommendation text based on this information to generate the final target recommendation text. This correction process ensures the accuracy and applicability of the recommendation text and improves the effectiveness of the recommendation. Finally, the corrected target recommendation text is pushed to the target object. By fusing multi-source data and a large language model, the present application can generate highly personalized insurance recommendation texts that dynamically match customer needs and external events.

[0152] By integrating multi-source data and using a large language model, the system can generate highly personalized recommendation texts. These texts can not only better match the needs and background of customers, but can also be dynamically adjusted according to real-time information, making recommendations more accurate and efficient. The system can obtain and process data such as weather forecasts, social media, and news events in real time to ensure that the generated recommendation texts keep up with the latest information and improve the timeliness and relevance of recommendations. Combined with historical claims data and real-time economic loss analysis, the generated recommendation texts are more persuasive. Through specific cases and data, customers can more intuitively understand the necessity of purchasing insurance, thereby increasing the insurance purchase rate. At the same time, the automatic generation of recommendation texts can greatly reduce the workload of insurance operators, allowing them to focus more on customer communication and services, thereby improving overall work efficiency.

[0153] See also Figure 8 The embodiment of the present application further provides a device for generating an insurance recommendation text, which can implement the above-mentioned method for generating an insurance recommendation text, and the device includes:

[0154] The acquisition module is used to obtain the historical claims data and external event data of the target object, and construct multi-source fusion data based on the historical claims data and external event data.

[0155] The first input module is used to input the multi-source fusion data into the pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data.

[0156] The comparison module is used to compare the multi-source fusion data with the original recommendation text to obtain deviation data;

[0157] The second generation module is used to input the deviation data and the original recommended text into the text generation model, so that the text generation model corrects the original recommended text to obtain the target recommended text.

[0158] The push module is used to push the target recommendation text to the target object.

[0159] The specific implementation of the device for generating the insurance recommendation text is substantially the same as the specific implementation of the method for generating the insurance recommendation text described above, and will not be described in detail herein.

[0160] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned method for generating insurance recommendation text when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0161] See also Fig. 9 , Fig. 9The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0162] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0163] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the method for generating the insurance recommendation text in the embodiment of this application;

[0164] Input / output interface 903, used to implement information input and output;

[0165] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0166] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0167] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0168] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned method for generating the insurance recommendation text.

[0169] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0170] The generation method of insurance recommendation text, the generation device of insurance recommendation text and related equipment provided in the embodiment of the present application obtain the historical claims data and external event data of the target object, and construct multi-source fusion data according to the historical claims data and external event data; input the multi-source fusion data into the pre-trained text generation model so that the text generation model generates the original recommendation text according to the multi-source fusion data; compare the multi-source fusion data with the original recommendation text to obtain deviation data; input the deviation data and the original recommendation text into the text generation model so that the text generation model corrects the original recommendation text to obtain the target recommendation text; push the target recommendation text to the target object. The present application first constructs a multi-source fusion data set by obtaining the historical claims data and external event data of the target object. This multi-source fusion data set not only takes into account the customer's personal historical data, but also integrates external factors that may affect insurance demand, such as weather changes, social events, etc., thereby providing a rich data basis for generating personalized recommendation texts. Next, the multi-source fusion data is input into a pre-trained text generation model, and the reasoning and text generation capabilities of the large language model are used to generate the original recommendation text based on these data. However, since the model may have certain deviations, the multi-source fusion data is also compared with the original recommendation text to obtain deviation data. This comparison process can identify inaccurate or irrelevant content that may exist in the original recommendation text, thereby providing a basis for correction. Then, the deviation data and the original recommendation text are input into the text generation model again, and the model corrects the original recommendation text based on this information to generate the final target recommendation text. This correction process ensures the accuracy and applicability of the recommendation text and improves the effectiveness of the recommendation. Finally, the corrected target recommendation text is pushed to the target object. By fusing multi-source data and a large language model, the present application can generate highly personalized insurance recommendation texts that dynamically match customer needs and external events.

[0171] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0172] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0173] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0174] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0175] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0176] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one item (one) of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0177] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0178] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0179] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0180] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0181] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for generating an insurance recommendation text, characterized in that: include: Acquire historical claims data and external event data of the target object, and construct multi-source fusion data based on the historical claims data and the external event data; Inputting the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data; Comparing the multi-source fusion data with the original recommendation text to obtain deviation data; Inputting the deviation data and the original recommended text into the text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text; The target recommendation text is pushed to the target object.

2. The method according to claim 1, characterized in that The external event data includes real-time weather data and news data. The historical claims data and external event data of the target object are obtained, and multi-source fusion data is constructed according to the historical claims data and the external event data, including: Acquire the historical claims data of the target object; wherein the historical claims data includes the historical number of claims, historical claims amount and historical claims type of the target object; Acquire real-time weather data within a future preset time period from at least one weather data source; Obtaining hot news data and natural disaster loss data caused by historical natural disasters from at least one news data source; Multi-source fusion data is constructed according to the historical claims data, the real-time weather data, the natural disaster loss data and the hot news data.

3. The method according to claim 2, characterized in that The multi-source fusion data is constructed according to the historical claims data, the real-time weather data, the natural disaster loss data and the hot news data, including: Extracting extreme weather features from the real-time weather data to obtain key weather features; Calculate the claim frequency and average claim amount of the target object according to the historical number of claims and the historical claim amount of the target object; Performing natural language processing on the hot news data to obtain hot keywords and corresponding event description data; Multi-source fusion data is constructed according to the key weather characteristics, the claim frequency, the average claim amount, the natural disaster loss data, the hot keywords and the corresponding event description data.

4. The method according to claim 2, characterized in that: The multi-source fusion data is constructed according to the historical claims data, the real-time weather data, the natural disaster loss data and the hot news data, including: Generate historical claims data weights based on the historical claims data; Generating real-time weather data weights according to the real-time weather data; generating natural disaster loss weights according to the natural disaster loss data; Generate hot news weights according to the hot news data; The multi-source fusion data is constructed according to the historical claims data, the historical claims data weight, the real-time weather data, the real-time weather data weight, the natural disaster loss data, the natural disaster loss weight, the hot news data and the hot news weight.

5. The method according to claim 2, characterized in that: Before constructing multi-source fusion data according to the historical claims data, the real-time weather data, the natural disaster loss data and the hot news data, it also includes: For the historical claim data, in response to finding that the claim amount in the historical claim data is missing, performing data filling processing on the claim amount; For the real-time weather data, in response to retrieving abnormal weather data in the real-time weather data, performing abnormal processing on the abnormal weather data based on a preset weather data threshold; The news data is deduplicated.

6. The method according to claim 1, characterized in that The multi-source fusion data is compared with the original recommendation text to obtain deviation data, including Extracting key data points from the original recommendation text; The key data points are compared with the corresponding original data points in the historical claims data and the external event data, and all the difference data points are marked. All the difference data points constitute the deviation data.

7. The method according to claim 6, characterized in that The step of inputting the deviation data and the original recommended text into the text generation model so that the text generation model corrects the original recommended text to obtain a target recommended text includes: The original recommended text and the deviation data marking all the difference data points are input into the text generation model together, so that the text generation model corrects the original recommended text according to each difference data point and refers to the corresponding original data point to obtain the target recommended text.

8. A device for generating an insurance recommendation text, characterized in that: include: An acquisition module, used to acquire historical claims data and external event data of a target object, and construct multi-source fusion data based on the historical claims data and the external event data; A first input module, used for inputting the multi-source fusion data into a pre-trained text generation model, so that the text generation model generates original recommendation text according to the multi-source fusion data; A comparison module, used to compare the multi-source fusion data with the original recommendation text to obtain deviation data; A second generation module, configured to input the deviation data and the original recommended text into the text generation model, so that the text generation model corrects the original recommended text to obtain a target recommended text; A push module is used to push the target recommendation text to the target object.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor implements the method for generating an insurance recommendation text as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the method for generating an insurance recommendation text as described in any one of claims 1 to 7.

Images