A method, device and storage medium for analyzing and mining insurance risk control dialogue data

Abstract

The present invention discloses a method, device and storage medium for analyzing and mining insurance risk control dialogue data, and relates to the technical field of insurance risk control management. The method comprises: reading structured dialogue data preprocessed by a preprocessing strategy for target dialogue data, extracting the first round, analyzing the first loss assessment sentence and the first loss verification sentence in combination with a dependency structure database to obtain a word pair, determining the first insurance item and generating an initial insurance map, calling a feature extraction strategy to extract the named entity of the feature of the second dialogue when it meets the predetermined constraints, rendering to the initial insurance map to obtain a target insurance map and forming a target comprehensive insurance map. It solves the technical problems in the prior art that insurance loss assessment and verification rely on manual labor, the judgment of insurance coverage is cumbersome, resulting in slow speed, low accuracy, lack of integration and low visualization, and achieves the technical effect of improving the accuracy and fairness of insurance claims, improving the efficiency of insurance coverage judgment, and improving the level of insurance risk control.

Description

Technical Field

[0001] The present application relates to the technical field of insurance risk control management, and in particular to an insurance risk control dialogue data analysis and mining method, device and storage medium. Background Art

[0002] In the insurance industry, damage assessment and insurance coverage determination in the insurance claims process are extremely important. The contradiction between insurance risk control resource demand is also prominent. Accurate and efficient damage assessment, reasonable judgment of insurance coverage and risk control have become key links in the development of the insurance industry. Traditional insurance risk control methods often rely more on manual experience, and there are problems of singleness and localization. They only rely on manual judgment of damage assessment and loss results, lack overall control of insurance dialogue data, lack in-depth analysis and utilization of data, and are difficult to fully and accurately determine insurance items and judge insurance coverage. There are unreasonable situations in resource allocation (such as labor costs, etc.), resulting in low efficiency and easy misjudgment. The formulation of insurance risk control plans is relatively simple and fixed, and cannot cope well with complex and changeable insurance business situations.

[0003] At the current stage, relevant technologies have technical problems such as insurance loss assessment and verification relying on manual labor, and insurance coverage judgment being cumbersome, resulting in slow speed, low accuracy and completeness, as well as scattered insurance data with a low degree of integration and visualization. Summary of the invention

[0004] The present application provides an insurance risk control dialogue data analysis and mining method, device and storage medium, adopts a reading preprocessing strategy to preprocess the structured dialogue data of the target dialogue data, extracts the first round, analyzes the first damage assessment sentence and the first loss verification sentence in combination with the dependency structure database to obtain word pairs, determines the first insurance item and generates an initial insurance map, calls the feature extraction strategy to extract the named entity of the feature of the second dialogue when it meets the predetermined constraints, renders it to the target insurance map of the initial insurance map and constructs a target comprehensive insurance map, realizes the technical effect of improving the accuracy and fairness of insurance claims, improving the efficiency of insurance scope judgment, and improving the level of insurance risk control by analyzing the damage assessment and loss verification sentences in combination with the dependency structure database.

[0005] This application provides an insurance risk control dialogue data analysis and mining method, including:

[0006] Read a preprocessing strategy, and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; extract the first round in the structured dialogue data, the first round includes a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first loss verification sentence; analyze the first damage assessment word pair of the first damage assessment sentence and the first loss verification word pair of the first loss verification sentence in sequence in combination with the dependency structure database; analyze the first damage assessment word pair and the first loss verification word pair to determine the first insurance item, and generate a first initial insurance map in combination with the first insurance item; call a feature extraction strategy when the first initial insurance map meets the first predetermined constraint; extract features from the second dialogue according to the feature extraction strategy to obtain a first named entity; render the first named entity to the first initial insurance map to obtain a first target insurance map, and construct a target comprehensive insurance map of the target dialogue data.

[0007] The present application also provides an insurance risk control dialogue data analysis and mining device, including:

[0008] a structured dialogue data acquisition module, the structured dialogue data acquisition module is used to read the preprocessing strategy, and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; a first round extraction module, the first round extraction module is used to extract the first round in the structured dialogue data, the first round includes the first dialogue and the second dialogue, wherein the first dialogue includes the first loss assessment sentence and the first loss verification sentence; a database analysis module, the database analysis module is used to sequentially combine the dependency structure database analysis to obtain the first loss assessment word pair of the first loss assessment sentence and the first loss verification word pair of the first loss verification sentence; a first insurance item determination module, the first insurance The project determination module is used to analyze the first loss assessment word pair and the first loss verification word pair to determine the first insurance project, and generate a first initial insurance map in combination with the first insurance project; the strategy extraction module, the strategy extraction module is used to call the feature extraction strategy when the first initial insurance map meets the first predetermined constraint; the first named entity acquisition module, the first named entity acquisition module is used to extract features from the second dialogue according to the feature extraction strategy to obtain the first named entity; the insurance map component module, the insurance map component module is used to render the first named entity to the first initial insurance map to obtain the first target insurance map, and construct the target comprehensive insurance map of the target dialogue data.

[0009] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0010] Read a preprocessing strategy, and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; extract the first round in the structured dialogue data, the first round includes a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first loss verification sentence; analyze the first damage assessment word pair of the first damage assessment sentence and the first loss verification word pair of the first loss verification sentence in sequence in combination with the dependency structure database; analyze the first damage assessment word pair and the first loss verification word pair to determine the first insurance item, and generate a first initial insurance map in combination with the first insurance item; call a feature extraction strategy when the first initial insurance map meets the first predetermined constraint; extract features from the second dialogue according to the feature extraction strategy to obtain a first named entity; render the first named entity to the first initial insurance map to obtain a first target insurance map, and construct a target comprehensive insurance map of the target dialogue data.

[0011] The present application proposes an insurance risk control dialogue data analysis and mining method, device and storage medium. First, the structured dialogue data pre-processed by the pre-processing strategy for the target dialogue data is read, and the first round is extracted. The first damage assessment sentence and the first loss verification sentence are analyzed in combination with the dependency structure database to obtain word pairs, the first insurance item is determined and an initial insurance map is generated. When the predetermined constraints are met, the feature extraction strategy is called to extract the named entity of the feature of the second dialogue, and the target insurance map is rendered to the initial insurance map and a target comprehensive insurance map is constructed. By analyzing the damage assessment and loss verification sentences in combination with the dependency structure database, automatically judging the insurance scope and generating a visual insurance map, the technical effect of improving the accuracy and fairness of insurance claims, improving the efficiency of insurance scope judgment, and improving the level of insurance risk control is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the device according to the embodiment of the present application. It should be understood that the previous or following operations are not necessarily performed precisely in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0013] Figure 1 A flowchart of an insurance risk control dialogue data analysis and mining method provided in an embodiment of the present application;

[0014] Figure 2 A schematic diagram of the structure of an insurance risk control dialogue data analysis and mining device provided in an embodiment of the present application.

[0015] Explanation of the accompanying drawings: structured dialogue data acquisition module 10, first round extraction module 20, database analysis module 30, first insurance item determination module 40, policy extraction module 50, first named entity acquisition module 60, insurance graph component module 70. DETAILED DESCRIPTION

[0016] 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.

[0017] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0018] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, device, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0019] The present application embodiment provides a method for analyzing and mining insurance risk control dialogue data, such as Figure 1 As shown, the method includes:

[0020] Step S100, read the preprocessing strategy, and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data. Specifically, read the preprocessing strategy that has been set in advance, which is a series of rules and methods formulated to convert the original target dialogue data into a structured form that is easier to analyze and process. The strategy includes data extraction rules, cleaning standards, and conversion methods, etc. According to the extraction rules, the key contents related to insurance risk control are screened out from the target dialogue data, such as loss assessment, loss verification, and insurance project description statements. Multi-level extraction is performed on complex dialogue data to ensure that key information is not omitted. According to the cleaning standards, noise is removed, such as extra spaces, special characters, repeated words, etc., and sentences with grammatical errors or incomplete sentences are corrected or marked. Finally, according to the conversion method, the cleaned dialogue data is converted into a structured format, such as a table or a specific data structure, so as to obtain structured dialogue data, which provides a clear data foundation for subsequent analysis and mining, and ensures that the whole process proceeds smoothly.

[0021] In one possible implementation, the preprocessing strategy is read, and the target dialogue data is preprocessed according to the preprocessing strategy to obtain structured dialogue data. Step S100 further includes step S110, and the preprocessing strategy refers to the preprocessing of extraction, cleaning and conversion. Specifically, in the analysis and mining of insurance risk control dialogue data, the preprocessing strategy includes three steps: extraction, cleaning and conversion. During extraction, key information is extracted from the original target dialogue data based on specific rules and standards, such as keywords such as "damage assessment", "loss verification", "insurance project", etc. Multi-level extraction can be performed for complex dialogue scenarios, and targeted extraction can also be performed based on participants, timestamps, etc. Cleaning removes noise such as irrelevant punctuation, special characters, and extra spaces, processes repeated words and sentences, and checks and corrects grammatical errors, incomplete sentences, or ambiguous expressions. Conversion is the process of converting the cleaned conversation data into a structured form suitable for subsequent analysis, such as recording key information in a table, or converting it into a tree or graph structure to reflect logical relationships. Data type conversion can also be performed. Through the preprocessing strategy of these three steps, the original data can be converted into a high-quality, structured data set, providing a basis for subsequent analysis and mining.

[0022] Step S200, extract the first round in the structured dialogue data, the first round includes the first dialogue and the second dialogue, wherein the first dialogue includes the first damage assessment sentence and the first loss verification sentence. Specifically, in the analysis and mining of insurance risk control dialogue data, the first round is extracted from the structured dialogue data, the first round consists of the first dialogue and the second dialogue, the first dialogue includes the first damage assessment sentence to determine the insured part and the actual degree of damage and the first loss verification sentence to verify and confirm it, through natural language processing technology, specific insurance field knowledge and combined with contextual information, the description of the insured part and the degree of damage is accurately extracted, and the loss verification sentence is compared and analyzed with the damage assessment sentence to ensure accuracy, the second dialogue is used to verify the total insurance amount of the project, it is necessary to extract the digital information and cost description in the dialogue, combine the degree of damage to the insured part, the repair cost standard and the insurance contract compensation ratio, etc. to determine the total insurance amount and accurately understand the language expression, and extract the content of the first round to provide key basic data for subsequent steps.

[0023] Step S300, analyze the first loss assessment word pair of the first loss assessment sentence and the first loss verification word pair of the first loss verification sentence in combination with the dependency structure database. Specifically, in the analysis of insurance risk control dialogue data, for the first loss assessment sentence and the first loss verification sentence, first associate them with the dependency structure database respectively, analyze the dependency relationship of the words in the sentence according to the rules and patterns in the database to understand the grammatical structure and semantic connotation, and extract the first loss assessment word pair of the first loss assessment sentence and the first loss verification word pair of the first loss verification sentence based on this. The loss assessment word pair and the loss verification word pair are usually composed of core words and modifiers, which respectively reflect the actual situation of the insurance part and the loss verification requirements. The word pairs provide accurate language elements for subsequent steps such as determining insurance items and generating insurance maps, which are crucial to improving the accuracy and reliability of insurance risk control dialogue data analysis and mining.

[0024] Step S400, analyze the first loss assessment word pair and the first loss verification word pair to determine the first insurance item, and generate a first initial insurance map in combination with the first insurance item. Specifically, in the insurance risk control dialogue data analysis, after obtaining the first loss assessment word pair and the first loss verification word pair, comprehensively interpret and compare and integrate the information of the two, and determine the first insurance item based on the insurance industry standards and rules and the reflected state of the insurance subject. After determination, generate the first initial insurance map with the first insurance item as the core content, present it in a visual way, add relevant detailed information, and supplement and adjust it in combination with other relevant information. The insurance map finally generated provides a visual basis for subsequent analysis and processing.

[0025] In a possible implementation, the first damage assessment word pair and the first damage assessment word pair are analyzed to determine the first insurance item, and a first initial insurance map is generated in combination with the first insurance item. Step S400 further includes step S410, step a: the first damage assessment word pair includes a first damage assessment modifier and a first damage assessment core word. Specifically, in the analysis of insurance risk control dialogue data, the composition structure of the first damage assessment word pair is first clarified. The first damage assessment word pair describes the characteristics of the insured object from the perspective of damage assessment, wherein the first damage assessment modifier is used to further explain the specific state of the insured part or situation represented by the first damage assessment core word. For example, if the first damage assessment core word is "vehicle part", then the first damage assessment modifier may be "severely damaged", which is combined to describe the degree of damage to a specific vehicle part in the insured object.

[0026] Step S420, step b: the first damage word pair includes a first damage modifier and a first damage core word. Specifically, the first damage word pair describes the insured subject from the damage perspective, and the first damage modifier supplements the first damage core word to jointly determine the status of the insured subject in the damage link. For example, the first damage core word is "body part", and the first damage modifier may be "slightly scratched", reflecting the actual situation of the body part determined by the damage.

[0027] Step S430, step c: extract the first predetermined word pair from the dependency structure database, the first predetermined word pair including the first predetermined modifier and the first predetermined core word. Specifically, the first predetermined word pair is extracted from the dependency structure database as the basis for subsequent judgment and matching. The first predetermined word pair represents a known insurance-related word combination pattern. For example, the first predetermined core word may be "common insurance parts", and the first predetermined modifier may be "typical damage conditions". This predetermined word pair can help the system quickly identify and match related insurance items in subsequent steps.

[0028] Step S440, step d: retrieve the first predetermined synonymous core word library of the first predetermined core word, and traverse the first damage assessment core word and the first nuclear damage core word in the first predetermined synonymous core word library to obtain a first traversal result. Specifically, in order to determine the degree of semantic similarity between the first damage assessment core word and the first nuclear damage core word and the predetermined core word, the system retrieves the first predetermined synonymous core word library of the first predetermined core word, and traverses the first damage assessment core word and the first nuclear damage core word in this synonymous core word library. If the first damage assessment core word is "car engine" and the first nuclear damage core word is "vehicle engine", words with similar semantics to them may be found in the synonymous core word library. Through traversal, the system can determine the degree of semantic association between the two core words and obtain the first traversal result.

[0029] Step S450, step e: retrieve the first predetermined synonymous modifier word library of the first predetermined modifier, and traverse the first damage assessment modifier word and the first verification damage modifier word in the first predetermined synonymous modifier word library to obtain a second traversal result. Specifically, in order to determine the similarity between the modifiers of damage assessment and verification damage and the predetermined modifiers, the system retrieves the first predetermined synonymous modifier word library of the first predetermined modifier, and traverses the first damage assessment modifier word and the first verification damage modifier word in this synonymous modifier word library. For example, if the first damage assessment modifier word is "moderately damaged" and the first verification damage modifier word is "damaged to a certain extent", their similarity can be judged in the synonymous modifier word library to obtain the second traversal result.

[0030] Step S460, step f: when the first traversal result and the second traversal result meet the second predetermined constraint, reversely match the first predetermined insurance item of the first predetermined word pair. Specifically, after the first traversal result and the second traversal result are generated, it is necessary to determine whether they meet the second predetermined constraint. The second predetermined constraint stipulates that the first damage assessment core word and the first core loss word both exist in the first predetermined synonymous core word library, and the first damage assessment modifier word and the first core loss modifier word both exist in the first predetermined synonymous modifier word library. The judgment process introduces a state model, and the training process of the state model may include the following steps: First, a large number of insurance risk control dialogue data samples are collected, and these data are preprocessed and word pairs are extracted to obtain various damage assessment word pairs, core loss word pairs and predetermined word pairs. According to known insurance items and constraints, these word pairs are labeled to determine which word pairs meet the second predetermined constraint and which do not. Machine learning algorithms, such as support vector machines, neural networks, etc., are used to train the labeled data to let the model learn how to judge whether the word pairs meet the second predetermined constraint. During the training process, the parameters of the model are continuously adjusted to improve the accuracy and generalization ability of the model. In the actual judgment process, the state model will track each information exchange and state change. For example, when the core word traversal and modifier traversal begin, the state model records the current steps and states. If the traversal result meets the second predetermined constraint, the state model will be updated to a state that meets the constraint and record the change. If it does not meet the constraint, the state model will also record the non-compliant state accordingly and may trigger further processing, such as re-traversal or adjustment of word pairs.

[0031] Step S470, step g: record the first predetermined insurance item as the first insurance item. Specifically, the system records the first predetermined insurance item that meets the conditions as the first insurance item, providing a basis for subsequent insurance analysis and processing. The first insurance item is obtained by analyzing the loss assessment word pair and the loss verification word pair and matching with the predetermined word pair, and it represents the specific insurance item corresponding to the current insurance subject.

[0032] In a possible implementation, step g: record the first predetermined insurance item as the first insurance item, and step S470 further includes step S471, when the first traversal result and the second traversal result do not meet the second predetermined constraint, repeat step c to step g to obtain the first insurance item. Specifically, when the first traversal result and the second traversal result do not meet the second predetermined constraint, the system will repeatedly extract the first predetermined word pair from the dependency structure database, and then call the first predetermined synonym core word library of the new first predetermined core word, traverse the first damage assessment core word and the first nuclear damage core word in it to obtain a new first traversal result, and at the same time call the first predetermined synonym modifier word library of the new first predetermined modifier, traverse the first damage assessment modifier word and the first nuclear damage modifier word in it to obtain a new second traversal result, and then determine whether the new traversal result meets the second predetermined constraint. If it meets the constraint, reverse match the first predetermined insurance item of the current first predetermined word pair, and finally record it as the first insurance item. By repeating these steps, the system continues to try different predetermined word pairs and synonym word libraries until the first insurance item that meets the second predetermined constraint is found, providing an accurate basis for subsequent insurance analysis and processing.

[0033] Step S500, when the first initial insurance map meets the first predetermined constraint, the feature extraction strategy is retrieved. Specifically, after the first initial insurance map is generated, the system determines whether the insurance map meets the first predetermined constraint based on the terms of the auto insurance contract purchased by the car owner, that is, whether the items in the map are within the insurance coverage. For example, engine wear and tear is usually not within the insurance coverage, but accidental scratches on the car body may be within the insurance coverage. A state model is introduced to track the judgment progress and status, collect and organize the information on the terms of the auto insurance contract and convert it into judgment rules and standards, and record each judgment process and result. When the first initial insurance map meets the first predetermined constraint, that is, all items are within the insurance coverage, the system will retrieve the feature extraction strategy, which may include extracting key features of the insurance items, providing detailed information for subsequent claims assessment and decision-making, and providing strong support for insurance risk control and claims processing.

[0034] In a possible implementation, when the first initial insurance map meets the first predetermined constraint, the feature extraction strategy is called, and step S500 further includes step S510, obtaining the target vehicle insurance terms of the target vehicle, and the target vehicle refers to the vehicle in the target conversation data. Specifically, in the analysis of insurance risk control conversation data, the concept of the target vehicle must be clarified first. The target vehicle is the vehicle involved in the target conversation data. The target conversation data is the communication content between the customer and the insurance customer service, the description in the claim application, etc. In order to accurately determine whether the insurance items related to the target vehicle meet the specific constraints, it is necessary to obtain the insurance terms of the target vehicle. The process can be carried out by querying the insurance database, connecting with the insurance company's system, or extracting from the insurance documents provided by the customer. For example, the system can locate and obtain the specific insurance terms of the target vehicle in the insurance database based on information such as the vehicle identification code or policy number in the target conversation data.

[0035] Step S520, generating the first predetermined constraint according to the insurance terms of the target vehicle. Specifically, once the insurance terms of the target vehicle are obtained, the first predetermined constraint can be generated according to these terms. The first predetermined constraint is an important basis for judging whether each item in the first initial insurance map generated subsequently is within the insurance scope. The insurance terms clearly stipulate which situations fall within the insurance liability scope and which situations are not within the insurance scope. For example, if the insurance terms of the target vehicle stipulate that the damage to the vehicle body caused by accidental collision is within the insurance scope, but the natural wear of the engine is not within the insurance scope, then the first predetermined constraint can be generated according to these terms, that is, for the insurance items involving the target vehicle, the items related to the accidental collision of the vehicle body are within the insurance scope, while the items related to the natural wear of the engine are not within the insurance scope. In actual operation, the system can convert the provisions in the terms into specific constraints by performing text analysis and semantic understanding on the insurance terms, so that they can be accurately applied in the subsequent judgment of the first initial insurance map.

[0036] Step S600, extract features from the second dialogue according to the feature extraction strategy to obtain a first named entity. Specifically, in the analysis of insurance risk control dialogue data, feature extraction is performed on the second dialogue involving important information such as the total amount of insurance approved according to the feature extraction strategy. First, the feature extraction strategy and goals are clarified, and the content of the second dialogue is deeply analyzed, including digital information, cost descriptions, and specific instructions for insurance items. Then, the feature extraction strategy is applied, including keyword extraction to identify key nouns, extract digital information, and extract features based on grammatical structures and semantic relationships. By extracting features from the second dialogue, the first named entity is finally obtained. The named entity summarizes the key features extracted from the second dialogue, provides a specific information basis for subsequent insurance analysis and decision-making, and helps to more accurately evaluate insurance items and determine the amount of claims.

[0037] In a possible implementation, the second dialogue is feature extracted according to the feature extraction strategy to obtain the first named entity, and step S600 further includes step S610, wherein the feature extraction strategy includes an accessory name extraction strategy and a labor cost extraction strategy, wherein the accessory name extraction strategy stores the accessory name feature, and the labor cost extraction strategy stores the labor cost feature. Specifically, in the analysis of insurance risk control dialogue data, it is clear that the feature extraction strategy consists of the accessory name extraction strategy and the labor cost extraction strategy, and the accessory name extraction strategy stores the accessory name feature, which is usually composed of nouns and may contain special characters such as brackets and hyphens, which are used to accurately represent specific accessory models or location information, such as "engine" or "brake pad (front left wheel)". The labor cost extraction strategy stores the labor cost feature, which is generally composed of numbers, currency symbols and possible time units (such as "yuan" and "hour"). For example, "500 yuan / hour" or "total labor cost 2000 yuan".

[0038] Step S620, analyzing the second dialogue based on the accessory name feature to obtain the first target accessory name. Specifically, when analyzing the second dialogue, the accessory name feature is used for extraction, the second dialogue is segmented, the sentence is split into individual words, part-of-speech tagging is performed, the part-of-speech of each word is determined, and the words or phrases that meet the features are screened from the results of word segmentation and part-of-speech tagging according to the template of the accessory name feature. For example, if "need to replace the brake pads (front left wheel)" appears in the second dialogue, the system will identify the word or phrase "brake pads (front left wheel)" as the first target accessory name according to the accessory name feature.

[0039] Step S630: Analyze the second dialogue based on the labor cost feature to obtain the first target labor cost. Specifically, analyze the second dialogue based on the labor cost feature, and first perform word segmentation and part-of-speech tagging according to the steps similar to the extraction of the accessory name, and extract the required combination of numbers, currency symbols and time units from the text according to the template of the labor cost feature. For example, if the second dialogue contains "the total maintenance labor cost is 1,500 yuan", the system will extract "1,500 yuan" as the first target labor cost.

[0040] Step S640, the first target part name and the first target man-hour cost form the first named entity. Specifically, the obtained first target part name and the first target man-hour cost are combined to form the first named entity. The named entity integrates part information and man-hour cost information, providing a more specific and comprehensive information basis for subsequent insurance analysis and decision-making. For example, if the first target part name is "Brake pads (front left wheel)" and the first target man-hour cost is "500 yuan", then the first named entity can be "Brake pads (front left wheel) repair cost (500 yuan)".

[0041] In a possible implementation, based on the part name feature, the second dialogue is analyzed to obtain the first target part name. Step S620 further includes step S621 of performing word segmentation on the second dialogue to obtain a set of words. Specifically, when analyzing the second dialogue, word segmentation is first performed to split the continuous text in the second dialogue into individual words. For example, if the second dialogue is "Need to replace a part of the car engine", after word segmentation, the set of words obtained may be "Need", "Replace", "Car", "Engine", "Of", "A certain", "Part", etc. Through word segmentation, complex text can be transformed into a set of individual words that are easier to analyze, laying a foundation for subsequent processing steps.

[0042] Step S622, stop words are removed from the set of words to obtain a set of target words, and the set of target words includes multiple target words. Specifically, stop words are removed from the obtained set of words. Stop words refer to words that appear frequently in the text but usually contribute less to semantic analysis, such as "Of", "Is", "In", etc. By removing stop words, interference from irrelevant information can be reduced, improving the accuracy and efficiency of subsequent analysis. After removing stop words, a set of target words is obtained, and the set contains multiple target words. The target words are more likely to be related to key information such as part names and man-hour costs. For example, after removing stop words from the set of words in the previous step, the set of target words obtained may be "Need", "Replace", "Engine", "Part", etc.

[0043] Step S623, obtain the first target part of speech of the first target word, where the first target word refers to any one of the multiple target words. Specifically, any one word is selected from the set of target words as the first target word, and its part of speech is obtained. Part-of-speech tagging is an important step in natural language processing, which can determine the grammatical role of each word in a sentence. For example, "Car" is tagged as a noun, and "Replace" may be tagged as a verb. By obtaining the part of speech of the first target word, the nature of the word and its role in the sentence can be further understood, providing a basis for extracting key information such as part names in the subsequent steps.

[0044] Step S624, combining the accessory name feature with the first target part of speech, to obtain the first target accessory name. Specifically, the first target accessory name is extracted by combining the accessory name feature and the part of speech of the first target word. The accessory name feature is usually composed of specific nouns, including special characters such as brackets and hyphens. According to the result of part of speech tagging, if the part of speech of the first target word is a noun, and its form conforms to the template of the accessory name feature, then it is possible to be determined as the first target accessory name. For example, if the first target word is "engine", its part of speech is a noun, and it conforms to the accessory name feature, then this word can be extracted as the first target accessory name. In the above manner, the system can accurately identify key information related to the accessory name from the target word set.

[0045] In a possible implementation, the second conversation is analyzed based on the labor cost feature to obtain a first target labor cost, and step S630 further includes step S631, combining the labor cost feature with the first target part of speech to obtain the first target labor cost. Specifically, in the process of determining the first target labor cost, the labor cost feature is usually composed of numbers, currency symbols and time units, which are used to identify content related to labor costs. The first target part of speech is the part of speech of a word in the target word set, which can help determine its grammatical role. If the first target part of speech is number-related, currency unit-related or time unit-related, the possibility of matching with each part of the labor cost feature is increased respectively. The system analyzes each word in the target word set one by one in combination with the labor cost characteristics. When the part of speech of a word matches a certain part of the labor cost characteristics and the surrounding words also meet the other parts, the word or group of words is determined to be the first target labor cost. For example, in "the total maintenance cost is 800 yuan", the quantitative part of speech of "800" and the monetary unit part of speech of "yuan" combined with the labor cost characteristics can determine "800 yuan" as the first target labor cost, providing key information for insurance analysis and decision-making.

[0046] Step S700, render the first named entity to the first initial insurance map to obtain the first target insurance map, and build the target comprehensive insurance map of the target dialogue data. Specifically, first make it clear that the first named entity contains key information such as the target accessory name and the target labor cost. After obtaining the first initial insurance map, render the first named entity to the first initial insurance map, that is, add the specific information represented by the first named entity to the insurance map, so that the insurance map can reflect the actual situation of the insurance subject and the claim cost in more detail and accurately. After rendering the first named entity to the first initial insurance map, start to build the target comprehensive insurance map of the target dialogue data, integrate the first initial insurance map and the rendered information, continue to analyze other information in the target dialogue data and gradually add it to the insurance map, and form a target comprehensive insurance map after multiple analyses and integrations, providing a comprehensive and accurate basis for insurance risk control and claims decisions.

[0047] The embodiment of the present application adopts a reading preprocessing strategy to preprocess the structured dialogue data of the target dialogue data, extracts the first round, analyzes the first damage assessment sentence and the first loss verification sentence in combination with the dependency structure database to obtain word pairs, determines the first insurance item and generates an initial insurance map, and calls the feature extraction strategy to extract the named entity of the feature of the second dialogue when the predetermined constraints are met, renders the target insurance map of the initial insurance map and constructs a target comprehensive insurance map, analyzes the damage assessment and loss verification sentences in combination with the dependency structure database, automatically determines the insurance scope and generates a visual insurance map, thereby achieving the technical effect of improving the accuracy and fairness of insurance claims, improving the efficiency of insurance scope judgment, and improving the level of insurance risk control.

[0048] In the above, refer to Figure 1 A method for analyzing and mining insurance risk control dialogue data according to an embodiment of the present invention is described in detail. Figure 2 An insurance risk control dialogue data analysis and mining device according to an embodiment of the present invention is described.

[0049] An insurance risk control dialogue data analysis and mining device according to an embodiment of the present invention is used to solve the technical problems in the prior art that insurance loss assessment and verification relies on manual work, the judgment of insurance coverage is cumbersome, resulting in slow speed, low accuracy and integrity, scattered insurance data lacking integration, and low visualization. By combining the dependency structure database to analyze loss assessment and verification sentences, automatically judge the insurance coverage and generate a visual insurance map, the technical effect of improving the accuracy and fairness of insurance claims, improving the efficiency of insurance coverage judgment, and improving the level of insurance risk control is achieved. An insurance risk control dialogue data analysis and mining device includes: a structured dialogue data acquisition module 10, a first round extraction module 20, a database analysis module 30, a first insurance project determination module 40, a strategy extraction module 50, a first named entity acquisition module 60, and an insurance map component module 70.

[0050] The structured dialogue data acquisition module 10 is used to read the preprocessing strategy and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data.

[0051] The first round extraction module 20 is used to extract the first round in the structured dialogue data, where the first round includes a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first damage verification sentence.

[0052] The database analysis module 30 is used to sequentially analyze the dependency structure database to obtain the first damage assessment word pair of the first damage assessment sentence and the first damage verification word pair of the first damage verification sentence.

[0053] The first insurance item determination module 40 is used to analyze the first loss assessment word pair and the first loss verification word pair to determine the first insurance item, and generate a first initial insurance map in combination with the first insurance item.

[0054] The strategy extraction module 50 is used to call a feature extraction strategy when the first initial insurance map meets a first predetermined constraint.

[0055] The first named entity acquisition module 60 is used to extract features from the second conversation according to the feature extraction strategy to obtain a first named entity.

[0056] The insurance map component module 70 is used to render the first named entity to the first initial insurance map to obtain a first target insurance map, and to construct a target comprehensive insurance map of the target conversation data.

[0057] The specific configuration of the structured dialogue data acquisition module 10 will be described in detail below. As described above, the preprocessing strategy is read, and the target dialogue data is preprocessed according to the preprocessing strategy to obtain the structured dialogue data. The structured dialogue data acquisition module 10 further includes: a preprocessing strategy component unit, and the preprocessing strategy component unit is used for the preprocessing of the preprocessing strategy, which refers to extraction, cleaning and conversion.

[0058] The specific configuration of the first insurance item determination module 40 will be described in detail below. As described above, the first damage assessment word pair and the first nuclear damage word pair are analyzed to determine the first insurance item, and a first initial insurance map is generated in combination with the first insurance item. The first insurance item determination module 40 further includes: a first damage assessment word composition unit, the first damage assessment word composition unit is used in step a: the first damage assessment word pair includes a first damage assessment modifier and a first damage assessment core word; a first nuclear damage word composition unit, the first nuclear damage word composition unit is used in step b: the first nuclear damage word pair includes a first nuclear damage modifier and a first nuclear damage core word; a first predetermined word pair extraction unit, the first predetermined word pair extraction unit is used in step c: extracting the first predetermined word pair in the dependency structure database, the first predetermined word pair includes a first predetermined modifier and a first predetermined core word; a word library traversal unit, the word library traversal unit is used in step d: retrieving the first predetermined core word A predetermined synonymous core word library is provided, and the first damage assessment core word and the first nuclear damage core word are traversed in the first predetermined synonymous core word library to obtain a first traversal result; a second traversal result acquisition unit, the second traversal result acquisition unit is used in step e: calling the first predetermined synonymous modifier word library of the first predetermined modifier, and traversing the first damage assessment modifier word and the first nuclear damage modifier word in the first predetermined synonymous modifier word library to obtain a second traversal result; a first predetermined insurance item matching unit, the first predetermined insurance item matching unit is used in step f: when the first traversal result and the second traversal result meet the second predetermined constraint, reverse matching the first predetermined word pair with the first predetermined insurance item; a first insurance item determination unit, the first insurance item determination unit is used in step g: recording the first predetermined insurance item as the first insurance item.

[0059] Among them, step g: record the first predetermined insurance item as the first insurance item, and the first insurance item determination unit further includes: a traversal result judgment subunit, and the traversal result judgment subunit is used to repeat steps c to g to obtain the first insurance item when the first traversal result and the second traversal result do not meet the second predetermined constraint.

[0060] The specific configuration of the strategy extraction module 50 will be described in detail below. As described above, when the first initial insurance graph meets the first predetermined constraint, the feature extraction strategy is called, and the strategy extraction module 50 further includes: an insurance clause acquisition unit, the insurance clause acquisition unit is used to acquire the target vehicle insurance clause of the target vehicle, the target vehicle refers to the vehicle in the target dialogue data; and a first predetermined constraint generation unit, the first predetermined constraint generation unit is used to generate the first predetermined constraint according to the target vehicle insurance clause.

[0061] The specific configuration of the first named entity acquisition module 60 will be described in detail below. As described above, the feature extraction is performed on the second dialogue according to the feature extraction strategy to obtain the first named entity, and the first named entity acquisition module 60 further includes: a feature extraction strategy composition unit, the feature extraction strategy composition unit is used for the feature extraction strategy including the accessory name extraction strategy and the labor cost extraction strategy, wherein the accessory name extraction strategy stores the accessory name feature, and the labor cost extraction strategy stores the labor cost feature; a first target accessory name acquisition unit, the first target accessory name acquisition unit is used to analyze the second dialogue based on the accessory name feature to obtain the first target accessory name; a first target labor cost acquisition unit, the first target labor cost acquisition unit is used to analyze the second dialogue based on the labor cost feature to obtain the first target labor cost; a first named entity determination unit, the first named entity determination unit is used to form the first named entity with the first target accessory name and the first target labor cost.

[0062] Among them, the second dialogue is analyzed based on the accessory name feature to obtain the first target accessory name, and the first target accessory name acquisition unit further includes: a word set acquisition subunit, the word set acquisition subunit is used to perform word segmentation processing on the second dialogue to obtain a word set; a stop word removal subunit, the stop word removal subunit is used to remove stop words from the word set to obtain a target word set, and the target word set includes multiple target words; a first target part-of-speech acquisition subunit, the first target part-of-speech acquisition subunit is used to obtain a first target part-of-speech of a first target word, and the first target word refers to any one of the multiple target words; a first target accessory name acquisition subunit, the first target accessory name acquisition subunit is used to combine the accessory name feature with the first target part-of-speech to obtain the first target accessory name.

[0063] Among them, the second dialogue is analyzed based on the labor cost characteristics to obtain the first target labor cost, and the first target labor cost acquisition unit further includes: a first target labor cost acquisition sub-unit, and the first target labor cost acquisition sub-unit is used to combine the labor cost characteristics with the first target part of speech to obtain the first target labor cost.

[0064] An insurance risk control dialogue data analysis and mining device provided in an embodiment of the present invention can execute an insurance risk control dialogue data analysis and mining method provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0065] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when executing the computer program: reading a preprocessing strategy, and preprocessing the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; extracting a first round in the structured dialogue data, wherein the first round includes a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first loss verification sentence; analyzing the first damage assessment word pair of the first damage assessment sentence and the first loss verification word pair of the first loss verification sentence in sequence in combination with a dependency structure database; analyzing the first damage assessment word pair and the first loss verification word pair to determine a first insurance item, and generating a first initial insurance map in combination with the first insurance item; calling a feature extraction strategy when the first initial insurance map meets a first predetermined constraint; performing feature extraction on the second dialogue according to the feature extraction strategy to obtain a first named entity; rendering the first named entity to the first initial insurance map to obtain a first target insurance map, and constructing a target comprehensive insurance map of the target dialogue data.

[0066] Although the present application makes various references to certain modules in the device according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0067] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for analyzing and mining insurance risk control dialogue data, characterized in that: include: Reading a preprocessing strategy, and preprocessing the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; Extracting a first round in the structured dialogue data, the first round including a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first damage verification sentence; Combining the dependency structure database to analyze in sequence, the first damage assessment word pair of the first damage assessment sentence and the first damage verification word pair of the first damage verification sentence are obtained; Analyze the first loss assessment word pair and the first loss verification word pair to determine a first insurance item, and generate a first initial insurance map in combination with the first insurance item; When the first initial insurance graph meets the first predetermined constraint, calling the feature extraction strategy; Extract features from the second conversation according to the feature extraction strategy to obtain a first named entity; Rendering the first named entity to the first initial insurance graph to obtain a first target insurance graph, and constructing a target comprehensive insurance graph of the target conversation data; The step of analyzing the first loss assessment word pair and the first loss verification word pair to determine the first insurance item includes: Step a: the first damage assessment word pair includes a first damage assessment modifier word and a first damage assessment core word; Step b: the first core damage word pair includes a first core damage modifier and a first core damage core word; Step c: extracting a first predetermined word pair from the dependency structure database, wherein the first predetermined word pair includes a first predetermined modifier word and a first predetermined core word; Step d: Retrieving a first predetermined synonymous core word library of the first predetermined core word, and traversing the first damage assessment core word and the first verification damage core word in the first predetermined synonymous core word library to obtain a first traversal result; Step e: Retrieving a first predetermined synonymous modifier word library of the first predetermined modifier, and traversing the first damage assessment modifier word and the first nuclear damage modifier word in the first predetermined synonymous modifier word library to obtain a second traversal result; Step f: when the first traversal result and the second traversal result meet the second predetermined constraint, reversely matching the first predetermined insurance item of the first predetermined word pair; Step g: Recording the first scheduled insurance item as the first insurance item; When the first traversal result and the second traversal result do not meet the second predetermined constraint, repeat step c to step g to obtain the first insurance item.

2. According to claim 1, a method for analyzing and mining insurance risk control dialogue data is characterized in that: The preprocessing strategy refers to extraction, cleaning and conversion preprocessing.

3. According to claim 1, the method for analyzing and mining insurance risk control dialogue data is characterized in that: include: Obtaining a target car insurance clause for a target car, wherein the target car refers to a car in the target conversation data; The first predetermined constraint is generated according to the insurance terms of the target vehicle.

4. According to claim 1, the method for analyzing and mining insurance risk control dialogue data is characterized in that: include: The feature extraction strategy includes a component name extraction strategy and a labor cost extraction strategy, wherein the component name extraction strategy stores a component name feature, and the labor cost extraction strategy stores a labor cost feature; Analyze the second conversation based on the accessory name feature to obtain a first target accessory name; Analyze the second conversation based on the labor cost feature to obtain a first target labor cost; The first target accessory name and the first target labor cost form the first named entity.

5. According to claim 4, a method for analyzing and mining insurance risk control dialogue data is characterized in that: include: Performing word segmentation processing on the second conversation to obtain a word set; Eliminating stop words from the word set to obtain a target word set, wherein the target word set includes a plurality of target words; Obtaining a first target part of speech of a first target word, where the first target word refers to any one of the multiple target words; The first target accessory name is obtained by combining the accessory name feature with the first target part of speech.

6. The method for analyzing and mining insurance risk control dialogue data according to claim 5, characterized in that: The first target labor cost is obtained by combining the labor cost feature with the first target part of speech.

7. An insurance risk control dialogue data analysis and mining device, characterized in that: The device is used to implement the insurance risk control dialogue data analysis and mining method according to any one of claims 1 to 6, and the device includes: A structured dialogue data acquisition module, the structured dialogue data acquisition module is used to read the preprocessing strategy and preprocess the target dialogue data according to the preprocessing strategy to obtain structured dialogue data; A first round extraction module, the first round extraction module is used to extract the first round in the structured dialogue data, the first round includes a first dialogue and a second dialogue, wherein the first dialogue includes a first damage assessment sentence and a first damage verification sentence; A database analysis module, the database analysis module is used to sequentially analyze the dependency structure database to obtain a first damage assessment word pair of the first damage assessment sentence and a first damage verification word pair of the first damage verification sentence; A first insurance item determination module, the first insurance item determination module is used to analyze the first loss assessment word pair and the first loss verification word pair to determine a first insurance item, and generate a first initial insurance map in combination with the first insurance item; A strategy extraction module, the strategy extraction module is used to call a feature extraction strategy when the first initial insurance map meets a first predetermined constraint; a first named entity acquisition module, the first named entity acquisition module being used to extract features from the second conversation according to the feature extraction strategy to obtain a first named entity; An insurance map component module, the insurance map component module is used to render the first named entity to the first initial insurance map to obtain a first target insurance map, and to construct a target comprehensive insurance map of the target conversation data.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images