Risk guarantee information pushing method and device, storage medium and computer equipment

Abstract

The invention relates to the technical field of computers and financial science and technology, and discloses a risk guarantee information pushing method and device, a storage medium and computer equipment. Identifying risk-related news texts related to the risks and insurance types capable of guaranteeing the risks involved in the risk-related news texts; evaluating the association degree of the risk-related news texts and insurance types, removing low-association-degree texts, obtaining candidate push news texts, analyzing first-level risk causes and second-level risk characteristics of the candidate push news texts, generating risk guarantee warning information, fusing the candidate push news texts, forming directed push news contents, and sending the directed push news contents to a user terminal. According to the technical scheme of the invention, the effective information of the hot news is extracted, the effective news is screened, and the corresponding insurance type is matched, so that the user can more effectively obtain the risk guarantee information meeting the own demand.

Description

Technical Field

[0001] The present application relates to the fields of computer technology and financial technology, and in particular to a method and device for pushing risk protection information, a storage medium, and a computer device. Background Art

[0002] With the improvement of people's living standards and the strengthening of risk awareness, risk protection awareness has become an indispensable part of family financial planning. However, users often face many inconveniences and challenges when understanding and choosing the type of insurance that suits them.

[0003] Users often need to actively search for definitions of various insurance types. This process is not only time-consuming and laborious, but the information obtained is often incomplete. The internet offers a wide variety of insurance information, but the quality varies greatly. This makes it difficult for users to filter out the insurance types that truly meet their needs and accurately determine which insurance types are most critical for their risk protection. Summary of the Invention

[0004] In view of this, the present application provides a risk protection information push method and device, storage medium, and computer equipment. By extracting effective information from hot news, screening effective news, and matching corresponding insurance types, combined with the specific risk content displayed in the news, risk protection warning information is generated in a targeted manner, so that users can more effectively obtain insurance type information that meets their needs.

[0005] According to one aspect of the present application, a method for pushing risk assurance information is provided, the method comprising:

[0006] Obtain multiple original news texts of domestic reports;

[0007] Identifying, from among the plurality of original news texts, risk-related news texts that are relevant to the risk, and types of insurance that can cover the risks involved in the risk-related news texts;

[0008] After evaluating the correlation between each risk-related news text and its corresponding insurance type, the risk-related news texts with low correlation are eliminated to obtain candidate push news texts;

[0009] For any candidate push news text, obtain the primary risk cause and secondary risk characteristics corresponding to the candidate push news text, and generate risk protection warning information based on the primary risk cause, secondary risk characteristics, and insurance type. The primary risk cause includes uncontrollable factors and internal human factors, and the risk manifestation caused by the primary risk cause has secondary risk characteristics, which include suddenness and periodicity.

[0010] After the risk protection warning information is integrated into the candidate push news text, the targeted push news content is obtained and pushed to the user terminal, so that the user receives the targeted push news content containing the risk protection warning information through the user terminal.

[0011] According to another aspect of the present application, a risk protection information push device is provided, the device comprising:

[0012] The original news acquisition module allows users to obtain multiple original news texts of domestic reports;

[0013] A risk news identification module is used to identify risk-related news texts and types of insurance that can protect the risks involved in the risk-related news texts from multiple original news texts;

[0014] The correlation evaluation module is used to evaluate the correlation between each risk-related news text and its corresponding insurance type, and then eliminate the risk-related news texts with low correlation to obtain candidate push news texts;

[0015] A risk information generation module is configured to obtain, for any candidate push news text, the primary risk causes and secondary risk characteristics corresponding to the candidate push news text, and generate risk protection warning information based on the primary risk causes, secondary risk characteristics, and insurance type. The primary risk causes include uncontrollable factors and inherent human factors, and the risk manifestations resulting from the primary risk causes include secondary risk characteristics, which include suddenness and periodicity.

[0016] The targeted news push module is used to integrate the risk protection warning information into the candidate push news text, obtain targeted push news content and push it to the user terminal, so that the user can receive the targeted push news content containing the risk protection warning information through the user terminal.

[0017] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned risk assurance information push method is implemented.

[0018] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned risk assurance information push method when executing the program.

[0019] Through the above-mentioned technical solution, the present application provides a risk protection information push method and device, storage medium, and computer equipment, which improves the pertinence and accuracy of risk protection information push through precise matching of news text and risk protection warning information, while enriching the news content, enhancing the user's reading experience and insurance awareness, and improving user satisfaction.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 A schematic diagram of a process for pushing risk assurance information provided by an embodiment of the present application is shown;

[0023] Figure 2 A flow chart showing another method for pushing risk assurance information provided by an embodiment of the present application is shown;

[0024] Figure 3 A schematic diagram of a risk protection information push framework provided in an embodiment of the present application is shown;

[0025] Figure 4 A schematic structural diagram of a risk protection information push device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0027] In this embodiment, a risk protection information push method is provided, such as Figure 1 As shown, the method includes:

[0028] Step 101: Acquire multiple original news texts of domestic reports.

[0029] Step 102 : Identify risk-related news texts and types of insurance that can protect the risks involved in the risk-related news texts from among the multiple original news texts.

[0030] In the above embodiment of the present application, the "risk protection warning information" can be integrated into the news for push, which can improve the user acquisition efficiency through a more accurate and personalized push method. Compared with the method of pushing in combination with news, the push method that does not use the news integration method will produce the following problems:

[0031] Singularity: Unable to fully capture the complex information in news events.

[0032] Lack of real-time performance: Unable to respond to hot news in real time, resulting in poor push timeliness.

[0033] Insufficient personalization: The push algorithm fails to fully consider the user's personalized needs and specific situations.

[0034] Lack of contextual understanding: Lacking in-depth understanding and analysis of news content, it is difficult to accurately determine the relevance of news to insurance types.

[0035] In the above embodiment of the present application, multiple original news texts of domestic reports are first obtained. By filtering domestic news from the massive amount of news, and further filtering news related to specific types of insurance (such as home insurance, medical insurance, accident insurance, pet insurance), etc., preparation can be made for subsequent customized push notifications. In particular, after obtaining the original news text, the captured original news text can also be cleaned, for example, by removing irrelevant information such as HTML tags, special characters, and advertisements, so as to retain the plain text content and improve subsequent recognition accuracy and efficiency.

[0036] Next, a large model (such as Tongyi Qianwen) can be used to identify risk-related news text and insurance types in the original news text that can be used to push risk protection warning information (i.e., risk-related). For example, if the identified risk-related news text is: A fire broke out in a residential complex, causing property damage to multiple households, where there is a causal relationship between "fire" and "family property damage," then for the original news text "fire caused family property damage," home property insurance can be selected as the corresponding insurance type for push notification.

[0037] For example, a risk-related news article might include: "A residential complex suffered flooding in its basement due to heavy rain, damaging several households' home appliances." Using this method, we can also identify home insurance as a suitable insurance type for push notifications, as home insurance typically covers property damage caused by natural disasters.

[0038] Specifically, the process of identifying risk-related news text involves identifying whether the original news text contains descriptions of events or situations that may or have caused loss or damage to household property. These events or situations can be potential, meaning they have not yet occurred but may occur in the future. Once they occur, they may cause financial loss or damage to household property (such as houses, furniture, appliances, vehicles, pets, etc.). Specifically, events or situations that may cause loss or damage to household property include:

[0039] 1. Natural disasters: Such as fire, flood, earthquake, storm, etc. These natural disasters can cause serious damage to family property.

[0040] 2. Man-made accidents: such as theft, destruction, accidental collisions, etc. These man-made factors can also lead to loss of household property.

[0041] 3. Technical failures: Such as electrical short circuits, burst water pipes, equipment damage, etc. If these technical failures are not repaired in time, they can also cause damage to household property.

[0042] 4. Legal liability: If a family member causes injury to others or property loss due to an accident, he or she may be held legally responsible, which is also a potential risk to family property.

[0043] 5. Pet-related risks: Pets biting others, damaging property, etc. These pet-related incidents can also cause financial losses to the family.

[0044] Optionally, step 101 obtains multiple original news texts of domestic reports, specifically including:

[0045] Step 1011: Acquire multiple original news texts to be filtered from domestic reports.

[0046] Step 1012, identifying the time information and / or location information contained in the original news text to be filtered, and filtering out the original news texts that are within a preset time period based on the identified time information, and / or filtering out the original news texts that meet the preset push area based on the identified location information among the multiple original news texts to be filtered.

[0047] In the above embodiment of the present application, when obtaining the original news text, further, a plurality of original news texts to be screened from domestic reports can be obtained first, and then the time information and / or location information in the original news text to be screened can be identified to screen recent news and / or determine whether it meets the requirements of the region to be pushed, so as to match the latest news in the region for different push times and push locations. Specifically, Tongyi Qianwen can be used to extract the time information and / or location information in the original news text to be screened and perform the screening.

[0048] In particular, the time entity recognition algorithm in NLP technology (Natural Language Processing) can also be used to extract time information from the original news text to be screened. Time information includes the time of news release, the time of event occurrence, etc. The formats of time information can include multiple, such as "May 1, 2023", "Yesterday", "Last week", etc., so a flexible time parsing algorithm can be used for processing. Similarly, the named entity recognition (NER) algorithm in NLP technology can also be used to extract location information from the original news text to be screened. Location information can include city names, district and county names, street names, etc. For fuzzy or abbreviated location information, it can be parsed and standardized in combination with a place name dictionary or a geographic information system (GIS).

[0049] Next, you can define the time period that needs to be filtered (preset time period) according to business needs. For example, it can be defined to filter original news texts within the last week, the last month or a specific date range. Then, compare the identified news release time or event occurrence time with the preset time period. If the time information falls within the preset time period, the news text is retained (that is, the original news text is obtained); otherwise, it is eliminated. For example, the preset time period is "April 1, 2023 to April 30, 2023", and the release time of an original news text to be filtered is extracted through the time information recognition algorithm as "April 15, 2023". Since this time falls within the preset time period, the news text is retained.

[0050] Similarly, based on business needs, you can also define the regions to be pushed. For example, you can target a province, city, or specific area for push notifications. The identified news location information is compared with the preset push location. If the location information matches the preset push location, the news article is retained; otherwise, it is discarded. For example, if the preset push location is "Beijing," and the location information recognition algorithm extracts the location information of a news article as "Chaoyang District, Beijing," the news article is retained because this location information matches the preset push location.

[0051] In practical applications, time and location filtering can be combined. For example, if you need to filter out news text that occurred in Beijing between April 1, 2023 and April 30, 2023, this goal can be achieved by combining time information recognition, location information recognition, and filtering logic.

[0052] Optionally, the original news text includes a title and a body text. Regarding step 102, "identifying risk-related news texts related to risks and types of insurance that can protect risks involved in the risk-related news texts among the multiple original news texts" specifically includes:

[0053] Step 1021 : extract multiple news keywords from the title and body of each original news text.

[0054] Step 1022: Analyze the news keywords in multiple original news texts using a large language model for identifying risk-related news to determine the risk-related news texts and the types of insurance that can protect the risks involved in the risk-related news texts. The large language model for identifying risk-related news is trained on news texts with preset insurance type labels. During training, the news texts are extracted into multiple news keywords for training.

[0055] In the above embodiments of the present application, a large model (such as Tongyi Qianwen or a separately trained large language model for identifying risk-related news) can be used to analyze news titles and texts, extract keywords, and identify news related to insurance types.

[0056] Specifically, news keyword extraction can be divided into title keyword extraction and text keyword extraction. Title keyword extraction involves extracting representative and key words from the title of the original news text. These words can summarize the main content of the news or highlight the key points of the news. Text keyword extraction involves in-depth analysis of the main body of the original news text to extract keywords that are closely related to the topic and can reflect the details and background of the news. Text keywords can include elements such as event type, location, people, time, cause, and result.

[0057] Next, the news keywords extracted from the title and text can be organized to construct a keyword set. This set will serve as the basis for subsequent news text classification and insurance type matching using a large language model.

[0058] Next, use a large language model for news text classification. Large language models, such as BERT and GPT, are pre-trained on large amounts of text data. During training, the large language model can be fine-tuned to make it more suitable for the specific news text classification task. For example, fine-tuning can use a portion of labeled news text as training data to improve model accuracy.

[0059] The constructed keyword set is fed into a large language model, which then categorizes news text. The goal of this classification is to identify news texts that are relevant to risk events and are likely to trigger risk mitigation alerts.

[0060] For a pre-defined set of insurance types, such as home insurance, medical insurance, accident insurance, pet insurance, etc., each insurance type can have a clear definition and coverage.

[0061] For news articles classified as risk-related, a large language model is further used to match them to corresponding insurance types. This matching process takes into account factors such as the risk type, loss severity, and impact scope described in the news article.

[0062] The final output is the identified risk-related news text and its corresponding insurance type. Specifically, the large language model's parameter settings, training data selection, and processing methods can be continuously optimized to improve the accuracy of news text classification and insurance type matching. Furthermore, the definition and coverage of the insurance type set can be adjusted based on actual circumstances to meet the needs of different users.

[0063] Step 103 : After evaluating the correlation between each risk-related news text and its corresponding insurance type, risk-related news texts with low correlation are eliminated to obtain candidate push news texts.

[0064] Next, the correlation between the news and specific insurance types can be evaluated. This means using a large model to infer the similarity between the news and the insurance types, matching news and pushing information to determine if the insurance type is strongly correlated, while eliminating news with low correlation. For example, a news article and an insurance type description can be fed into a pre-trained large language model as input, and the model outputs a correlation score. Based on the score, the correlation between the news article and the insurance type can be determined. Specifically, consider the following two risk-related news articles and their corresponding insurance types:

[0065] 1. News text 1:

[0066] Content: "A fire broke out in a residential complex, causing damage to the property of several households. Fortunately, no casualties were reported."

[0067] Insurance type: home contents insurance.

[0068] 2. News text 2:

[0069] Content: "A child fell and was injured while playing in the park and was rushed to the hospital for treatment."

[0070] Insurance type: accident insurance.

[0071] Evaluation Process:

[0072] 1. Keyword matching. For news text 1, the keywords "fire", "family", and "property" are extracted and matched with the insurance type description of home insurance. It is found that the matching degree is high.

[0073] 2. For news text 2, the keywords "children", "fall", and "injury" are extracted and matched with the description of the insurance type of accident insurance. It is also found that the matching degree is high.

[0074] Next, news text 1 and the home insurance description are input into the large language model to obtain a relevance score, for example, a score of 0.8 (out of 1). News text 2 and accident insurance are input into the large language model to obtain a relevance score, for example, a score of 0.75.

[0075] According to the results of keyword matching and semantic similarity calculation, it can be judged that news text 1 has a high correlation with home property insurance, and news text 2 has a high correlation with accident insurance.

[0076] A relevance threshold (e.g., 0.6) is preset, and news texts with relevance below the threshold are considered low-relevance texts and removed. In this example, the relevance of the two news texts is both above the threshold, so both are retained as candidate news texts for push notification.

[0077] Optionally, after evaluating the correlation between each risk-related news text and its corresponding insurance type in step 103, risk-related news texts with low correlation are eliminated to obtain candidate push news texts, specifically including:

[0078] Step 1031: Use the large language model for correlation evaluation to evaluate the correlation between each risk-related news text and its corresponding insurance type, then eliminate the risk-related news texts with low correlation to obtain candidate push news texts. The correlation evaluation results include no correlation, weak correlation, correlation, and strong correlation, and no correlation, weak correlation, and correlation belong to low correlation respectively. The large language model for correlation evaluation is trained through evaluation samples, and each evaluation sample corresponds to an insurance type label and a preset correlation definition result.

[0079] In the above-mentioned embodiments of the present application, a large language model for relevance evaluation can be trained to perform a specific evaluation process. Specifically, a set of sample data (evaluation samples) with insurance type labels and preset relevance definition results is obtained. The sample data can be obtained from historical push data, user feedback, or expert annotations.

[0080] Example of sample data:

[0081] Sample 1: The news article title is "Typhoon Landfall, Damage to Homes in Many Areas," and the text describes the damage to homes caused by the typhoon. The insurance type label is "Home Insurance," and the preset correlation definition result is "Strong Correlation."

[0082] Sample 2: The news article is titled "Stock Market Fluctuates, Investors Should Be Cautious," and the text discusses stock market fluctuations. The insurance type label is "Investment Insurance," and the default correlation definition is "Unrelated."

[0083] Sample 3: The news article is titled "New Cars Launched, Safety Improved," and the main text describes the safety features of new cars. The insurance type label is "Auto Insurance," and the preset correlation definition is "Weak Correlation."

[0084] Using this sample data, we can train a large language model to assess the correlation between news articles and insurance types. During training, the model learns how to determine the correlation between news articles and insurance types based on their content.

[0085] Next, the trained model is used to evaluate the relevance of each risk-related news text in order to screen out candidate push news texts.

[0086] Step 104: For any candidate pushed news text, obtain the primary risk cause and secondary risk characteristics corresponding to the candidate pushed news text, and generate risk protection warning information based on the primary risk cause, secondary risk characteristics and insurance type. The primary risk cause includes uncontrollable factors and internal human factors, and the risk manifestation caused by the primary risk cause has secondary risk characteristics, which include suddenness and periodicity.

[0087] In the above embodiment of the present application, the first-level risk causes and second-level risk characteristics are obtained step by step to prepare for the subsequent generation of specific risk protection warning information. In particular, the first-level risk causes specifically include:

[0088] 1. Uncontrollable factors, that is, factors caused by external factors that are difficult for individuals or organizations to predict or control, such as natural disasters (such as floods, earthquakes, typhoons, heavy rains, etc.).

[0089] 2. Internal human factors, which are caused by the behavior or decision-making of individuals or organizations, can be reduced through measures such as improved management and enhanced awareness, such as traffic accidents, fire, theft, health risks (such as disease, accidental injury, etc.), and business operation risks (such as poor management, investment errors, etc.).

[0090] Secondary risk characteristics include:

[0091] 1. Sudden: This refers to an unexpected event that occurs without obvious signs or warning periods, causing immediate or short-term losses to the victims. Examples include natural disasters (such as earthquakes and floods), traffic accidents, fires, and sudden illnesses.

[0092] 2. Cyclicality: This refers to the presence of significant time cycles that may fluctuate with factors such as seasons and economic cycles, resulting in long-term or recurring impacts on victims. Examples include seasonal diseases (such as influenza and dengue fever), unemployment risks caused by economic fluctuations, and cyclical business risks (such as cyclical industry recessions).

[0093] Optionally, refer to Figure 2 As shown, in step 104, "generating risk protection warning information based on the primary risk cause, secondary risk characteristics and insurance type" specifically includes:

[0094] Step 1041: Perform semantic analysis on the candidate pushed news text to obtain the primary risk cause that leads to the risk.

[0095] Step 1042: Based on the acquired primary risk causes, a secondary semantic analysis is performed on the candidate pushed news text to obtain the secondary risk characteristics of the risks caused by the primary risk causes. Among them, the secondary risk characteristics of the risks caused by uncontrollable factors are suddenness or periodicity, and the secondary risk characteristics of the risks caused by internal human factors are suddenness.

[0096] Step 1043: Based on named entity recognition technology, entities are extracted from the candidate pushed news text, and the loss subject corresponding to the candidate pushed news text is determined according to the extracted entities, wherein the loss subject includes at least one of a person, a house, and a pet. When the determined loss subject is a person, the age and gender of the loss subject are further determined according to the extracted entities.

[0097] Step 1044, based on the obtained secondary risk characteristics, the candidate push news text is semantically analyzed three times to obtain a primary risk factor, and based on the primary risk factor, the candidate push news text is semantically analyzed four times to obtain a secondary risk factor, wherein the risks caused by different primary risk causes are manifested as the same or different secondary risk characteristics. When the risks caused by the same primary risk cause are manifested as different secondary risk characteristics, the different secondary risk characteristics correspond to different primary risk factors, and different primary risk factors correspond to different secondary risk factors.

[0098] Step 1045: Generate risk protection warning information based on the loss subject, primary risk factors, secondary risk factors, and insurance type.

[0099] In the above-mentioned embodiment of the present application, semantic analysis technology can accurately identify the primary risk causes that lead to risks in the candidate push news text, which helps to more deeply understand the nature of the risk event and provide accurate basic data for subsequent risk assessment and risk protection information push. Then, based on the identification of the primary risk causes, secondary semantic analysis is further used to further obtain the secondary risk characteristics of the risk. The level-by-level detailed analysis helps to more comprehensively understand the possible manifestations and development trends of the risk event. After understanding the primary causes and secondary characteristics of the risk, it is possible to more accurately match the user's risk protection needs.

[0100] Next, named entity recognition is used to further identify the subject of the loss, so that risk protection warning information can be pushed in a more targeted manner. Named Entity Recognition (NER) is a key technology in natural language processing (NLP), which is used to identify entities with specific meanings from text, such as names of people, places, names of organizations, time, date, currency, percentages, etc. These entities usually represent key information in the text and are crucial for understanding the meaning and context of the text. Named entity recognition technology can accurately identify named entities in the text and classify them into predefined entity types by analyzing and understanding the language structure, contextual information and entity features in the text.

[0101] Next, we will continue to determine more "detailed" first-level risk factors and second-level risk factors based on the second-level risk characteristics, so that we can more accurately match and push risk protection warning information. Regarding the relationship between the first-level risk causes, second-level risk characteristics, first-level risk factors and second-level risk factors, as shown in Table 1, when the second-level risk characteristics of the risks caused by uncontrollable factors are sudden, the first-level risk factors include natural disasters, bursting of water and heating pipes, falling objects in the air, falling external objects, and collapse of external objects. When the second-level risk characteristics of the risks caused by uncontrollable factors are cyclical, the first-level risk factors include aging of home furnishings and appliances. The second-level risk characteristics of the risks caused by internal human factors are sudden, and the corresponding first-level risk factors include water leakage, fire, gas leakage, explosion and property theft. The second-level risk factors of natural disasters include heavy rain, hail, mudslides, Lightning strikes, strong winds, floods, snowstorms, cliff collapses, icicles, ground subsidence or sinking; secondary risk factors for bursting of water and heating pipes include bursting of tap water pipes, bursting of heating pipes, bursting of sewer pipes, and bursting of indoor and outdoor pipes of solar water heaters; secondary risk factors for aging of home appliances include aging of indoor units of air conditioners, aging of floor heating manifolds, and leakage of household appliances; secondary risk factors for water leakage include leaking water and heating pipes and forgetting to turn off the faucet; secondary risk factors for fire include fires caused by household gas appliances, fires caused by short circuits of household electrical appliances or circuits, and fires caused by other internal or external fire sources in the home; secondary risk factors for gas leakage include gas poisoning; secondary risk factors for explosion include explosions caused by the use of natural gas or liquefied gas or gas leakage.

[0102] Table 1

[0103]

[0104]

[0105] To this end, risk protection push information can be further generated based on risk factors, insurance types, loss entities, etc.

[0106] In particular, different loss entities (people, houses, pets) will also correspond to different types of insurance. For example, human losses may involve life insurance, health insurance, etc.; house losses may involve home insurance, earthquake insurance, etc.; pet losses may involve pet insurance. The causes of primary risks (uncontrollable factors / internal human factors) may also affect the choice of insurance types. For example, risks caused by uncontrollable factors may require more comprehensive protection, while risks caused by internal human factors may require specific liability insurance. Similarly, the characteristics of secondary risks (suddenness / cyclicality) will also affect the choice of insurance types. For example, sudden risks may require high compensation capabilities, while cyclical risks may require long-term protection plans.

[0107] Step 105 , after integrating the risk protection warning information into the candidate push news text, the obtained targeted push news content is pushed to the user terminal, so that the user receives the targeted push news content containing the risk protection warning information through the user terminal.

[0108] In the above-described embodiment of the present application, by incorporating risk protection warning information into candidate push news text, users can be more aware of potential risks, enhance their insurance awareness, and better understand the role and value of risk protection. Targeted push news content is customized based on user interests and needs, meeting their personalized needs and enabling them to receive information defining relevant insurance types, improving the relevance and effectiveness of the information.

[0109] Optionally, the insurance type includes at least one of home insurance, medical insurance, accident insurance, and pet insurance. Different insurance types correspond to insurance type definition information. Regarding step 1045, "generating risk protection warning information based on the loss subject, primary risk factor, secondary risk factor, and insurance type" specifically includes:

[0110] Step 10451: construct a risk protection warning information framework, wherein the risk protection warning information framework includes an opening statement, risk warnings, an introduction to the definition of insurance types, and the role of risk protection.

[0111] Step 10452: Generate risk protection warning information based on the risk protection warning information framework, insurance type definition information, loss subject, primary risk factor, secondary risk factor and candidate push news text.

[0112] In the above embodiments of the present application, a risk protection warning information framework may be constructed, which may include:

[0113] 1. Opening remarks design:

[0114] Purpose: To attract users' attention and bring up topics.

[0115] Content: Briefly introduce the event or risk scenario in the candidate push news text, which is related to the user's daily life or concerns and resonates with the user.

[0116] Example: "Have you recently paid attention to the XX incident in XX region? This type of risk is actually all around us and cannot be ignored."

[0117] 2. Risk Warning:

[0118] Purpose: To emphasize the severity and prevalence of risks and enhance users' crisis awareness.

[0119] Content: Based on specific cases in news texts, analyze the primary and secondary risk factors and point out the losses these risks may bring to users.

[0120] Example: "This incident was caused by XX (a primary risk factor), and this type of risk often carries the risk of XX (a secondary risk factor). If it occurs, it could cause you significant losses."

[0121] 3. Insurance type definition information display:

[0122] Purpose: To introduce in detail the characteristics, coverage and advantages of the insurance types.

[0123] Content: Define the information by type of insurance, highlighting how this type of insurance provides protection against the risks mentioned in the news text.

[0124] Example: "XX type of insurance can provide protection against XX risk. The specific definition of this type is XX."

[0125] 4. Explanation of risk protection function:

[0126] Purpose: To explain the specific risk protection function and improve the details.

[0127] Content: This section explains the role of risk protection by combining the specific needs of the loss subject (such as people, houses, pets), and the analysis of primary and secondary risk factors.

[0128] Example: "Considering that your XX (loss subject) may be exposed to the risk of XX (secondary risk factor) caused by XX (primary risk factor), XX type of insurance can provide you with comprehensive protection and greater peace of mind."

[0129] Finally, generate a risk protection warning message, integrating the opening statement, risk warning, insurance type definition information, and risk protection function to form a complete risk protection warning message. Polish the language to ensure that the message is clear, fluent, and attractive. Example:

[0130] Opening remarks: "Have you recently paid attention to the XX incident in XX region? This kind of risk is actually all around us and cannot be ignored."

[0131] Risk Warning: "This incident was caused by XX (a primary risk factor), and this type of risk often carries the risk of XX (a secondary risk factor). If it occurs, it may cause you considerable losses."

[0132] Insurance type definition information: "XX type of insurance provides protection against XX risk. The specific definition of the type is XX."

[0133] Risk protection: "Considering that your XX (loss subject) may face the risk of XX (secondary risk factor) caused by XX (primary risk factor), XX type of insurance can provide you with comprehensive protection and make your life more secure."

[0134] Through the above steps, risk protection warning information that meets the user's actual needs can be generated.

[0135] By applying the technical solution of this embodiment, in order to address the current situation where it is impossible to utilize massive amounts of hot news and combine the risks therein to push risk protection warning information, the above-mentioned embodiment of this application extracts effective information from hot news, screens effective news, and matches corresponding insurance types, and pushes risk protection warning information in a targeted manner, so that users can more effectively obtain information on insurance types that meet their needs.

[0136] In a specific embodiment, a "multi-agent model" can also be used to match and screen news and risk protection warning information. If only one large model agent is used, the news screening will not be refined enough, that is, the existing system cannot accurately distinguish which news is related to a specific type of insurance when screening news. The existing risk factor assessment method is too simple and cannot fully cover the risk factors in the news. It cannot fully utilize the specific information in the news, such as the loss subject, location, date, etc., resulting in inaccurate results. The existing speech generation method also lacks an in-depth understanding of the news content, and the generated speech is not personalized and targeted enough. Through the collaborative work of the multi-agent system, the modules can be expanded and adjusted more flexibly and easily. Specifically, the multi-agent system can include a news screening agent, an insurance type correlation analysis agent, a loss subject identification agent, a time and location analysis agent, an age and gender identification agent (for the case where the loss subject is a person), a risk factor assessment agent, an information push agent and a speech generation agent. The functions of each agent are, for example, as follows:

[0137] 1. News screening agent:

[0138] Function: Filter domestic news from massive news, and further filter news related to specific types of insurance (such as home insurance, medical insurance, accident insurance, and pet insurance).

[0139] Implementation: Use a large model to analyze news headlines and text, extract keywords, and identify news related to insurance types.

[0140] 2. Insurance type correlation analysis agent:

[0141] Function: Evaluate the relevance of news to a specific type of insurance.

[0142] Implementation: Use a large model to infer the similarity between news and the insurance types matched by the news screening agent, that is, whether the matching news and the insurance type pushed are strongly correlated, and eliminate news with low correlation.

[0143] 3. Loss subject identification agent:

[0144] Function: Identify loss entities (such as people, houses, pets, etc.) in news.

[0145] Implementation: Use named entity recognition (NER) technology to extract entities from news and determine the subject type.

[0146] 4. Time and place analysis agent:

[0147] Function: Identify the time and location information in the news, filter recent news, and determine whether it meets the push area (that is, for different push times and push locations, match the latest news in the area).

[0148] Implementation: Use a large model to extract time and location information from news and perform filtering.

[0149] 5. Age and gender identification agent (for cases where the loss subject is a person):

[0150] Function: Identify the age and gender of the subject in the news.

[0151] Implementation: Use text analysis techniques to extract age and gender information from news.

[0152] 6. Risk Factor Assessment Agent (step by step):

[0153] Function: Evaluate risk classification and risk factors in news, and finally output primary and secondary risk factors.

[0154] Implementation: Use a large model to analyze news content and select matching risk factors based on pre-set risk factors. Specifically, to ensure the accuracy of first- and second-level risk factor classification, first perform inference on the first- and second-level risk classifications (corresponding to first-level risk causes and second-level risk characteristics), then perform inference on the first- and second-level risk factors. For scenarios requiring hierarchical classification, a step-by-step inference agent is more effective due to the attribution of hierarchical relationships.

[0155] 7. Information push agent:

[0156] Function: Determine appropriate risk protection warning information based on risk factors, insurance types, loss subjects, etc.

[0157] Implementation: Use push algorithms to comprehensively consider multiple factors and push the most suitable information.

[0158] 8. Speech generation agent:

[0159] Function: Generate push statements based on the pushed risk protection warning information and news content.

[0160] Implementation: Use the big model to generate personalized sentences related to the push information.

[0161] By applying the technical solution of this embodiment, through refined news screening and risk factor assessment, information push results are more accurate and responsive. By comprehensively considering specific information in the news, such as the cause of loss, location, and date, the results are more personalized. Push statements generated based on the news content are more targeted, improving user comprehension. The collaborative operation of the multi-agent system makes the system more flexible, allowing for easy expansion and adjustment of modules. Through more accurate and personalized push notifications, user satisfaction and trust are improved.

[0162] In another specific embodiment, it is also possible to build Figure 3 The framework shown in the figure is used to implement the process of combining news screening with risk protection warning information for push notification. Specifically, through systematic process design, the time and cost of manual operations are reduced, making the entire workflow more efficient. The process includes key factor extraction (corresponding to risk factors) and news sentiment analysis steps, which help to more accurately understand user needs and thus provide more accurate risk protection information push notifications. By polishing news content and generating short copy, the attractiveness and dissemination effect of the copy can be significantly improved, making information communication more effective. The use of emoticons is added to the process to help enhance the user's interactive experience, making the information (corresponding to insurance protection information) more vivid and interesting, and improving user participation and satisfaction.

[0163] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a risk protection information push device, such as Figure 4 As shown, the device includes:

[0164] The original news acquisition module 201 is used to acquire multiple original news texts reported domestically;

[0165] The risk news identification module 202 is used to identify risk-related news texts and types of insurance that can protect the risks involved in the risk-related news texts from multiple original news texts;

[0166] The correlation evaluation module 203 is used to evaluate the correlation between each risk-related news text and its corresponding insurance type, and eliminate risk-related news texts with low correlation to obtain candidate push news texts;

[0167] The risk information generation module 204 is configured to obtain, for any candidate push news text, the primary risk cause and secondary risk characteristics corresponding to the candidate push news text, and generate risk protection warning information based on the primary risk cause, secondary risk characteristics, and insurance type. The primary risk cause includes uncontrollable factors and inherent human factors, and the risk manifestation resulting from the primary risk cause includes secondary risk characteristics, which include suddenness and periodicity.

[0168] The targeted news push module 205 is used to integrate the risk protection warning information into the candidate push news text, obtain targeted push news content and push it to the user terminal, so that the user can receive the targeted push news content containing the risk protection warning information through the user terminal.

[0169] It should be noted that for other corresponding descriptions of the functional units involved in the risk protection information push device provided in the embodiment of the present application, please refer to Figures 1 to 2 The corresponding description in the method will not be repeated here.

[0170] Based on the above Figures 1 to 2 The method shown in FIG. 1 is a method for performing the above-mentioned operation. Accordingly, the embodiment of the present application further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned operation is performed. Figures 1 to 2 The risk protection information push method shown.

[0171] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0172] Based on the above Figures 1 to 2 The method shown, and Figure 4 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figures 1 to 2 The risk protection information push method shown.

[0173] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a Wi-Fi interface), etc.

[0174] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0175] The storage medium may also include an operating system and a network communication module. An operating system is a program that manages and stores the hardware and software resources of a computer device, supporting the execution of information processing programs and other software and / or programs. The network communication module facilitates communication between components within the storage medium, as well as with other hardware and software within the physical device.

[0176] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware to obtain multiple original news texts, identify the risk-related news texts and corresponding insurance types that are related to the risk; evaluate the correlation between the risk-related news texts and the insurance types, eliminate the low-correlation texts, obtain candidate push news texts, analyze their primary risk causes and secondary risk characteristics, determine the risk protection warning information that can be pushed, integrate it into the candidate push news texts, and form a targeted push news content to be sent to the user terminal. By extracting effective information from hot news, screening effective news, and matching the corresponding insurance types, and pushing risk protection warning information in a targeted manner, users can more effectively obtain insurance type definition information that meets their needs.

[0177] Those skilled in the art will understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application. Those skilled in the art will understand that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0178] The serial numbers of the above application are for descriptive purposes only and do not represent the advantages or disadvantages of the implementation scenarios. The above disclosures are only a few specific implementation scenarios of the present application, but the present application is not limited thereto, and any changes that can be made by those skilled in the art should fall within the scope of protection of the present application.

Claims

1. A method for pushing risk protection information, characterized in that: The method comprises: Obtain multiple original news texts of domestic reports; Identifying, from among the plurality of original news texts, risk-related news texts that are relevant to the risk, and types of insurance that can cover the risks involved in the risk-related news texts; After evaluating the correlation between each risk-related news text and its corresponding insurance type, the risk-related news texts with low correlation are eliminated to obtain candidate push news texts; For any candidate push news text, obtain the primary risk cause and secondary risk characteristics corresponding to the candidate push news text, and generate risk protection warning information based on the primary risk cause, secondary risk characteristics, and insurance type. The primary risk cause includes uncontrollable factors and internal human factors, and the risk manifestation caused by the primary risk cause has secondary risk characteristics, which include suddenness and periodicity. After the risk protection warning information is integrated into the candidate push news text, the targeted push news content is obtained and pushed to the user terminal, so that the user receives the targeted push news content containing the risk protection warning information through the user terminal.

2. The method according to claim 1, characterized in that The obtaining of the primary risk causes and secondary risk characteristics corresponding to the candidate pushed news text includes: Performing semantic analysis on the candidate pushed news text to obtain the primary risk cause that leads to the risk; Based on the acquired primary risk causes, a secondary semantic analysis is performed on the candidate push news text to obtain the secondary risk characteristics of the risks caused by the primary risk causes. Among them, the secondary risk characteristics of the risks caused by uncontrollable factors are suddenness or periodicity, and the secondary risk characteristics of the risks caused by internal human factors are suddenness.

3. The method according to claim 1, characterized in that The risk protection warning information generated based on the primary risk cause, secondary risk characteristics and insurance type includes: Based on named entity recognition technology, entities are extracted from the candidate pushed news text, and the loss subject corresponding to the candidate pushed news text is determined based on the extracted entities, wherein the loss subject includes at least one of a person, a house, and a pet. When the determined loss subject is a person, the age and gender of the loss subject are further determined based on the extracted entities; Based on the obtained secondary risk characteristics, the candidate pushed news text is subjected to three semantic analyses to obtain a primary risk factor, and based on the primary risk factor, the candidate pushed news text is subjected to four semantic analyses to obtain a secondary risk factor, wherein the risks resulting from different primary risk causes are manifested as the same or different secondary risk characteristics. When the risks resulting from the same primary risk cause are manifested as different secondary risk characteristics, the different secondary risk characteristics correspond to different primary risk factors, and the different primary risk factors correspond to different secondary risk factors. Generate risk protection warning information based on the loss subject, primary risk factors, secondary risk factors and insurance type.

4. The method according to claim 3, characterized in that The insurance types include at least one of home insurance, medical insurance, accident insurance, and pet insurance. Different insurance types correspond to insurance type definition information. The risk protection warning information generated based on the loss subject, primary risk factor, secondary risk factor, and insurance type includes: Constructing a risk protection warning information framework, wherein the risk protection warning information framework includes an opening statement, risk warnings, an introduction to insurance types, and the role of risk protection; Risk protection warning information is generated based on the risk protection warning information framework, insurance type definition information, loss subject, primary risk factors, secondary risk factors and candidate push news texts.

5. The method according to claim 1, characterized in that The original news text includes a title and a body text. The risk-related news texts related to the risk are identified in the plurality of original news texts, and the types of insurance that can protect the risks involved in the risk-related news texts include: Extracting multiple news keywords from the title and body of each original news text; In multiple original news texts, a large language model for identifying risk-related news is used to analyze the news keywords to determine the risk-related news texts and the types of insurance that can protect the risks involved in the risk-related news texts. The large language model for identifying risk-related news is trained by news texts with preset insurance type labels. During training, the news texts are extracted into multiple news keywords for training.

6. The method according to any one of claims 1 to 5, characterized in that After evaluating the correlation between each risk-related news text and its corresponding insurance type, the risk-related news texts with low correlation are eliminated to obtain candidate push news texts, including: Using the large language model for correlation evaluation, the correlation between each risk-related news text and its corresponding insurance type is evaluated respectively, and then the risk-related news texts with low correlation are eliminated to obtain candidate push news texts. Among them, the correlation evaluation results include no correlation, weak correlation, correlation and strong correlation. No correlation, weak correlation and correlation belong to low correlation respectively. The large language model for correlation evaluation is trained through evaluation samples, and each evaluation sample corresponds to an insurance type label and a preset correlation definition result.

7. The method according to claim 6, characterized in that The multiple original news texts of domestic reports are obtained, including: Obtain multiple original news texts to be screened from domestic reports; Identify time information and / or location information contained in the original news text to be filtered, and among multiple original news texts to be filtered, filter the original news texts that are within a preset time period based on the identified time information, and / or filter the original news texts that meet the preset push area based on the identified location information.

8. A risk protection information push device, characterized in that: The device comprises: The original news acquisition module is used to obtain multiple original news texts of domestic reports; A risk news identification module is used to identify risk-related news texts and types of insurance that can protect the risks involved in the risk-related news texts from multiple original news texts; The correlation evaluation module is used to evaluate the correlation between each risk-related news text and its corresponding insurance type, and then eliminate the risk-related news texts with low correlation to obtain candidate push news texts; A risk information generation module is configured to obtain, for any candidate push news text, the primary risk causes and secondary risk characteristics corresponding to the candidate push news text, and generate risk protection warning information based on the primary risk causes, secondary risk characteristics, and insurance type. The primary risk causes include uncontrollable factors and inherent human factors, and the risk manifestations resulting from the primary risk causes include secondary risk characteristics, which include suddenness and periodicity. The targeted news push module is used to integrate the risk protection warning information into the candidate push news text, obtain targeted push news content and push it to the user terminal, so that the user can receive the targeted push news content containing the risk protection warning information through the user terminal.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for pushing risk assurance information according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, wherein: When the processor executes the computer program, the method for pushing risk assurance information according to any one of claims 1 to 7 is implemented.

Images