Intelligent claim settlement auxiliary system based on artificial intelligence

Abstract

The invention discloses an intelligent claim settlement auxiliary system based on artificial intelligence, and the system comprises the following modules: a data collection module which is used for collecting claim settlement image data and corresponding insurance clause contents, and distributing a unique index; the clause analysis module is used for carrying out structured analysis on the insurance clauses and generating a claim settlement judgment basis set; the character recognition module is used for recognizing character information in the image data and outputting a structured claim settlement text; the semantic coding module is used for carrying out context semantic coding on the claim settlement text to obtain a semantic representation vector; the clause matching module is used for executing matching judgment of texts and clauses; the compensation judgment module is used for outputting a compensation prediction result, amount estimation and a reason label; and the auxiliary report generation module is used for generating a structured claim auxiliary report. According to the method, an integrated processing flow of automatic image-text information identification, term semantic matching and intelligent claim payment judgment is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and insurance information processing technology, and in particular to an intelligent claim settlement auxiliary system based on artificial intelligence. BACKGROUND

[0002] With the deepening application of artificial intelligence technology in the field of finance and insurance, intelligent claim settlement auxiliary technology based on image recognition and natural language processing has gradually become a key path to improve claim settlement efficiency and decision accuracy. The existing intelligent claim settlement system mainly relies on image and text recognition technology to extract text from claim settlement materials, and performs preliminary screening and judgment by matching simple rules or keyword-based retrieval methods against insurance clauses. However, in actual application, the existing technology generally has the following problems:

[0003] The text information in the image data is complex and diverse, and the traditional OCR model has low recognition accuracy in medical invoices, handwritten documents and other scenarios, making it difficult to ensure the integrity and accuracy of structured data; there is a lack of deep semantic understanding between text and insurance clauses, and the existing key field matching-based solution easily ignores context dependence and polysemy expression, resulting in inaccurate clause application judgment; the claim settlement process usually does not introduce a multi-task modeling strategy, and lacks systematic integration of information such as claim settlement results, amount estimation and reason explanation, making it difficult to support intelligent claim settlement decisions; in addition, some systems do not establish a clear structured output process, and the generated auxiliary report structure is loose and the content is redundant, which is not conducive to manual checking and subsequent audit use.

[0004] Therefore, how to provide an intelligent claim settlement auxiliary system based on artificial intelligence is a problem that those skilled in the art need to solve. SUMMARY

[0005] One object of the present application is to provide an intelligent claim settlement auxiliary system based on artificial intelligence. The present application combines image recognition, context-enhanced semantic encoding, multi-round rule-enhanced matching network and multi-task claim settlement model to comprehensively improve the structured processing capability of claim settlement materials and the accuracy of clause application judgment, realize the comprehensive prediction of whether to pay, the recommended amount of payment and the reason for payment, and generate a structured claim settlement auxiliary report, having the advantages of strong semantic understanding ability, high clause matching accuracy and strong interpretability of judgment results.

[0006] According to an embodiment of the present application, an intelligent claim settlement auxiliary system based on artificial intelligence comprises the following modules:

[0007] A data acquisition module is configured to acquire image data submitted by an insured person for claim settlement and corresponding insurance product clause content, and assign a unique index identifier to each claim settlement application data;

[0008] a clause analysis module configured to perform structural analysis on the insurance product clause content to generate a set of judgment basis for claim settlement;

[0009] a character recognition module configured to perform character recognition and output structured claim settlement text information;

[0010] a semantic encoding module configured to perform context-enhanced feature encoding on the structured claim settlement text information to obtain a context semantic representation vector of each claim settlement text content;

[0011] a clause matching module configured to perform a clause matching task and output a clause matching score and a satisfaction judgment result;

[0012] a payout determination module configured to perform a payout judgment and amount estimation task and output a prediction result of whether to pay, a recommended payout amount, and a payout reason label;

[0013] an auxiliary report generation module configured to generate a structured claim settlement auxiliary report.

[0014] An intelligent claim settlement auxiliary method based on artificial intelligence according to an embodiment of the present application includes the following steps:

[0015] claim settlement image data submitted by an insured person and corresponding insurance product clause content are collected, and each claim settlement application data is assigned a unique index identifier;

[0016] the insurance product clause content is subjected to structural analysis to generate a set of judgment basis for claim settlement;

[0017] the claim settlement image data is input into an improved PaddleOCR model for character recognition to obtain structured claim settlement text information;

[0018] the structured claim settlement text information is subjected to semantic encoding to obtain a context semantic representation vector of each claim settlement text content;

[0019] the context semantic representation vector and the set of judgment basis for claim settlement are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task, and a clause matching score and a satisfaction judgment result are output;

[0020] the structured claim settlement text information, the clause matching score, and the satisfaction judgment result are jointly input into a multi-task semantic-driven payout determination network to perform a payout judgment and amount estimation task, and a prediction result of whether to pay, a recommended payout amount, and a payout reason label are output;

[0021] a structured claim settlement auxiliary report is generated according to the prediction result of whether to pay, the recommended payout amount, and the payout reason label.

[0022] Optionally, the insurance product clause content is structured parsed to generate a claim judgment basis set, specifically including:

[0023] The insurance product clause content is structured parsed in a manner combining rule template parsing and syntax dependency analysis, the rule template parsing extracts the pay claim responsibility field, trigger condition field, pay claim proportion field, exclusion responsibility field and proof material requirement field by matching the preset insurance clause language mode;

[0024] If the rule template cannot be matched, the syntax dependency analysis method is used to automatically extract the subject-predicate-object structure and modification components related to the pay claim condition from the insurance product clause content based on the verb center structure recognition and syntax dependency path calculation, and the field extraction result is supplemented;

[0025] The field identification and standard semantic label are given to each field content extracted according to the preset field classification rule to construct a clause structure field set;

[0026] Based on the clause structure field set, the fields are grouped according to the pay claim type similarity, guarantee subject consistency and clause logic dependency relationship, and the semantic association matrix between the fields is constructed by using the sentence vector cosine similarity calculation to generate a claim judgment basis set with structure organization and semantic mapping relationship.

[0027] Optionally, the claim image data is input into the improved PaddleOCR model for character recognition to obtain structured claim text information, specifically including:

[0028] An improved PaddleOCR model is constructed, the improved PaddleOCR model includes a bill structure positioning module, a text detection subnetwork, a text direction classification subnetwork and a character recognition subnetwork;

[0029] The input claim image data is preprocessed by the bill structure positioning module, specifically including performing a layout structure segmentation operation, positioning the medical bill main body area, cropping the image edge interference content, and extracting the suspected information block area as a structured recognition candidate area;

[0030] The structured recognition candidate area is input into the text detection subnetwork, the text detection subnetwork uses a detection structure based on a differential feature pyramid to extract multi-scale character bounding boxes and outputs detection results containing character region coordinates, confidence scores and candidate order labels;

[0031] The detection result is input into the text direction classification subnetwork, the text direction classification subnetwork judges the direction angle of each character candidate box through a lightweight convolution structure, and performs a direction rotation correction operation on the detected character image area to make the character image arranged in a uniform direction;

[0032] The corrected text image region is input into a text recognition sub-network, a convolutional feature extraction structure, a sequence modeling structure, and a character decoding structure, to sequentially complete feature extraction, sequence learning, and text recognition of the text image. In the character decoding process, a medical term dictionary error correction mechanism is introduced. According to the candidate character sequence in the text recognition result and the preset medical term dictionary, when there is an obvious difference between the recognition result and the dictionary entry, the most similar standard term is replaced;

[0033] Finally, structured claim settlement text information is output, including visit time, hospital, diagnosis conclusion, expense item, and amount field.

[0034] Optionally, the structured claim settlement text information is semantically encoded to obtain a context semantic representation vector of each claim settlement text content, specifically including:

[0035] The visit time, medical institution, diagnosis conclusion, medical item, and amount information in the structured claim settlement text information are spliced according to a preset field order to construct a standard input text sequence.

[0036] The input text sequence is input into a context-enhanced feature encoder for semantic encoding. The context-enhanced feature encoder includes a character-level convolutional feature extraction layer, a bidirectional long short-term memory network modeling layer, and a semantic attention aggregation layer. The semantic attention aggregation layer performs weighted combination on the time sequence hidden state to generate a context semantic representation vector that integrates global context.

[0037] Optionally, the context semantic representation vector and the claim settlement judgment basis set are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task, output a clause matching score and a satisfaction judgment result, specifically including:

[0038] The context semantic representation vector and the claim settlement judgment basis set are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task, output a clause matching score and a satisfaction judgment result.

[0039] The multi-round rule-enhanced BERT matching network includes a multi-round semantic interaction modeling layer, a structure field alignment mechanism, and a rule label guided scoring module.

[0040] The multi-round semantic interaction modeling layer is based on a double-input attention mechanism. In each round, a cross-attention weight matrix is constructed between the context semantic representation vector and the clause field semantic vector, and the matching feature representation is iteratively updated in multiple rounds to extract matching relationships at different semantic granularities.

[0041] The structure field alignment mechanism aligns the structure fields in the structured claim settlement text information with the semantic fields in the claim settlement judgment basis set one by one according to a preset field tag mapping rule, and constructs a field-level semantic alignment vector set;

[0042] The rule tag guided scoring module inputs the field-level semantic alignment vector set and the rule tag representations of the claim settlement judgment basis set into a scoring function, and jointly calculates the matching score and satisfaction judgment result of each clause based on the cosine semantic similarity and the rule satisfaction degree. The rule satisfaction degree is logically judged according to the claim settlement judgment basis set corresponding to the clause field. When the structured claim settlement field content satisfies all optional condition tags, the rule satisfaction degree is recorded as 1, otherwise as 0.

[0043] Optionally, the multi-task semantic driven claim settlement judgment network specifically includes:

[0044] A multi-task semantic driven claim settlement judgment network is constructed, and the network includes a claim settlement judgment sub-network, an amount estimation sub-network and a claim settlement reason tag generation sub-network.

[0045] The structured claim settlement text information, the clause matching score and the satisfaction judgment result are spliced in the field dimension to form a joint semantic feature vector.

[0046] The joint semantic feature vector is input into the claim settlement judgment sub-network. The claim settlement judgment sub-network adopts a double-tower semantic classification structure, each tower is composed of two layers of Transformer encoder and feedforward network, the coupling relationship between claim settlement semantics and judgment basis is established by respectively modeling claim settlement text semantics and clause matching information, and the prediction result of whether to claim is output.

[0047] The field information related to the amount in the joint semantic feature vector is input into the amount estimation sub-network. The amount estimation sub-network sequentially transforms the amount field, treatment method and claim settlement ratio through a fully connected layer and a ReLU activation function, and outputs a predicted value of the recommended claim settlement amount.

[0048] The joint semantic feature vector is input into the claim settlement reason tag generation sub-network. The claim settlement reason tag generation sub-network specifically inputs the joint semantic feature vector into a layer of feedforward fully connected network to perform nonlinear feature transformation, and outputs the claim settlement reason tag processed by a Sigmoid activation function.

[0049] Optionally, the structured claim settlement auxiliary report is generated according to the prediction result of whether to claim, the recommended claim settlement amount and the claim settlement reason tag, specifically including:

[0050] Integrate the prediction result of whether to pay, the recommended compensation amount and the compensation reason label to generate a claim settlement decision summary content;

[0051] Combine the clause matching score and the satisfaction judgment result to generate clause explanation information corresponding to the compensation reason label;

[0052] In the case of containing a refusal or risk-related label in the compensation reason label, generate structured risk prompt information;

[0053] According to the preset claim settlement auxiliary report format template, a structured claim settlement auxiliary report containing a claim settlement application unique index, insurance product clause information, claim settlement decision summary content, recommended compensation amount and clause explanation information is constructed.

[0054] The beneficial effects of the present application are:

[0055] (1) Improve the claim settlement data structuring and semantic understanding ability: by inputting the claim settlement image data into the improved OCR model for structured text extraction, and introducing the context-enhanced semantic encoder, the non-standard format text and medical terminology expression can be effectively parsed, and the representation vector with consistent context semantics is generated, which significantly improves the machine-parsable nature and semantic understanding accuracy of the claim settlement data.

[0056] (2) Enhance the applicability judgment ability of clause matching: the present application constructs a multi-round rule-enhanced matching network, which integrates claim settlement text and clause semantic information, and performs clause matching task by combining semantic similarity and rule satisfaction degree, which not only improves the accuracy of clause adaptation, but also has a controllable judgment strategy, realizing the fine-grained identification of the applicability of compensation under complex clause conditions.

[0057] (3) Realize the intelligentization and explainable output of claim settlement decision: based on the multi-task semantic-driven claim settlement decision network, the present application simultaneously outputs the claim settlement decision, amount estimation and compensation reason label, and realizes the intelligent closed loop from data recognition, clause judgment to decision suggestion through the structured auxiliary report automatic generation mechanism, which has high automation level and good explainability, significantly improves the efficiency and decision quality of the insurance industry claim settlement process. BRIEF DESCRIPTION OF DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0059] Figure 1 A structural schematic diagram of an intelligent claim settlement auxiliary system based on artificial intelligence is proposed for the present application;

[0060] Figure 2A flow chart of an intelligent claim settlement auxiliary method based on artificial intelligence is provided for the present application.

[0061] Figure 3 A framework diagram of an improved PaddleOCR model in an intelligent claim settlement auxiliary method based on artificial intelligence is provided for the present application. DETAILED DESCRIPTION

[0062] The present application will now be further described in detail with reference to the drawings. These drawings are simplified schematic diagrams and only schematically show the basic structure of the present application, and thus only show the configurations related to the present application.

[0063] REFERENCE Figure 1 An intelligent claim settlement auxiliary system based on artificial intelligence comprises the following modules:

[0064] A data collection module is configured to collect claim image data submitted by an insured person and corresponding insurance product clause content, and assign a unique index identifier to each claim application data;

[0065] A clause analysis module is configured to perform structured analysis on the insurance product clause content to generate a claim judgment basis set;

[0066] A character recognition module is configured to perform character recognition and output structured claim text information;

[0067] A semantic encoding module is configured to perform context-enhanced feature encoding on the structured claim text information to obtain a context semantic representation vector of each claim text content;

[0068] A clause matching module is configured to perform a clause matching task and output a clause matching score and a satisfaction judgment result;

[0069] A claim settlement module is configured to perform claim settlement judgment and amount estimation tasks and output a prediction result of whether to claim, a recommended claim amount, and a claim reason label;

[0070] An auxiliary report generation module is configured to generate a structured claim settlement auxiliary report.

[0071] REFERENCE Figures 2-3 An intelligent claim settlement auxiliary method based on artificial intelligence comprises the following steps:

[0072] Step 1: Collect claim image data submitted by an insured person and corresponding insurance product clause content, and assign a unique index identifier to each claim application data;

[0073] Step 2: Perform structured analysis on the insurance product clause content to generate a claim judgment basis set;

[0074] Step three: input the claim image data into the improved PaddleOCR model for text recognition to obtain structured claim text information;

[0075] Step four: perform semantic coding on the structured claim text information to obtain a context semantic representation vector for each claim text content;

[0076] Step five: input the context semantic representation vector and the claim judgment basis set into the multi-round rule-enhanced BERT matching network to perform a clause matching task, and output a clause matching score and a satisfaction judgment result;

[0077] Step six: input the structured claim text information, the clause matching score and the satisfaction judgment result into the multi-task semantic-driven payout determination network to perform a payout judgment and amount estimation task, and output a prediction result of whether to pay, a recommended payout amount and a payout reason label;

[0078] Step seven: generate a structured claim assistance report according to the prediction result of whether to pay, the recommended payout amount and the payout reason label.

[0079] In the embodiment, the structured analysis of the insurance product clause content to generate the claim judgment basis set specifically includes:

[0080] The insurance product clause content is analyzed by combining rule template analysis and syntax dependency analysis. The rule template analysis extracts the payout responsibility field, trigger condition field, payout proportion field, exclusion responsibility field and proof material requirement field by matching the pre-set insurance clause language mode;

[0081] If the rule template cannot be matched, the syntax dependency analysis method is used to automatically extract the subject-predicate-object structure and modification components related to the payout condition from the insurance product clause content based on the verb center structure recognition and syntax dependency path calculation, and complete the field extraction result;

[0082] The field identification and standard semantic label are given to each field content extracted according to the pre-set field classification rule to construct a clause structure field set;

[0083] Based on the clause structure field set, the fields are grouped according to the payout type similarity, guarantee subject consistency and clause logical dependency relationship, and the semantic association matrix between the fields is constructed by using the sentence vector cosine similarity calculation to generate a claim judgment basis set with structure organization and semantic mapping relationship.

[0084] The embodiment combines rule template parsing and syntax dependency analysis method to structurally parse the content of the insurance product clauses, which not only ensures the accurate extraction of known semantic patterns, but also intelligently completes complex or non-standard expressions, significantly improves the completeness and semantic accuracy of the claim-related field extraction, and effectively solves the poor adaptability of traditional parsing methods to flexible clause expressions.

[0085] In the embodiment, the claim image data is input into the improved PaddleOCR model for character recognition to obtain structured claim text information, which specifically includes:

[0086] An improved PaddleOCR model is constructed, which includes a bill structure positioning module, a text detection subnetwork, a text direction classification subnetwork, and a character recognition subnetwork.

[0087] The input claim image data is preprocessed by the bill structure positioning module, which specifically includes performing layout structure segmentation, positioning the medical bill main body area, cropping the image edge interference content, and extracting the suspected information block area as a structured recognition candidate area.

[0088] The structured recognition candidate area is input into the text detection subnetwork, which uses a detection structure based on a differential feature pyramid to extract multi-scale text bounding boxes and outputs detection results containing text region coordinates, confidence scores, and candidate order labels.

[0089] The detection results are input into the text direction classification subnetwork, which uses a lightweight convolution structure to determine the direction angle of each text candidate box and performs direction rotation correction on the detected text image area to align the text images in the same direction.

[0090] The corrected text image area is input into the character recognition subnetwork, which includes a convolution feature extraction structure, a sequence modeling structure, and a character decoding structure to sequentially complete feature extraction, sequence learning, and character recognition operations. In the character decoding process, a medical term dictionary error correction mechanism is introduced to match the candidate character sequence in the character recognition result with the preset medical term dictionary. When there is a significant difference between the recognition result and the dictionary entry, the most similar standard term is replaced.

[0091] Finally, the structured claim text information is output, including the fields of visit time, visit hospital, diagnosis conclusion, expense item, and amount.

[0092] The embodiment introduces mechanisms such as bill structure positioning, differential feature pyramid detection and direction classification correction by constructing an improved PaddleOCR model, accurately extracts the text area in the medical bill, and optimizes the recognition result through character decoding and medical term correction, effectively improves the accuracy of image text recognition and the effect of field structured extraction, especially in the face of complex bill format, blurred and inclined text and other actual scenes, it can still output high-quality structured claim settlement text information, providing reliable data support for subsequent clause matching and payment determination.

[0093] In the embodiment, the structured claim settlement text information is semantically encoded to obtain a context semantic representation vector of each claim settlement text content, specifically including:

[0094] The visit time, medical institution, diagnosis conclusion, medical project and amount information in the structured claim settlement text information are spliced according to a preset field order to construct a standard input text sequence;

[0095] The input text sequence is input into a context-enhanced feature encoder for semantic encoding, the context-enhanced feature encoder includes a character-level convolution feature extraction layer, a bidirectional long short-term memory network modeling layer and a semantic attention aggregation layer, the semantic attention aggregation layer performs weighted combination on the time sequence hidden state to generate a context semantic representation vector fused with the global context.

[0096] The embodiment constructs a context-enhanced feature encoder to perform multi-layer semantic modeling on the standard input text sequence, introduces bidirectional long short-term memory network and semantic attention aggregation mechanism on the basis of character-level convolution extraction, realizes global context modeling of key information such as visit time, medical institution and diagnosis conclusion in the claim settlement text, and effectively improves the accuracy and discriminability of text semantic representation.

[0097] In the embodiment, the context semantic representation vector and the claim settlement judgment basis set are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task, output a clause matching score and a satisfaction judgment result, specifically including:

[0098] The context semantic representation vector and the claim settlement judgment basis set are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task, output a clause matching score and a satisfaction judgment result;

[0099] The multi-round rule-enhanced BERT matching network includes a multi-round semantic interaction modeling layer, a structure field alignment mechanism and a rule label guided scoring module;

[0100] The multi-round semantic interaction modeling layer is based on a double-input attention mechanism, constructs a cross-attention weight matrix between the context semantic representation vector and the clause field semantic vector in each round, and iteratively updates the matching feature representation in multiple rounds to extract the matching relationship at different semantic granularities.

[0101] The structure field alignment mechanism aligns the structure fields in the structured claim settlement text information with the semantic fields in the claim settlement judgment basis set one by one according to a preset field label mapping rule, and constructs a field-level semantic alignment vector set.

[0102] The rule label guided scoring module inputs the field-level semantic alignment vector set and the rule label representation of the compensation condition field, the exclusion liability field and the proof material requirement field in the claim settlement judgment basis set into a scoring function, calculates the matching score and the satisfaction judgment result of each clause based on the cosine semantic similarity and the rule satisfaction degree, and the rule satisfaction degree is logically judged according to the claim settlement judgment basis set corresponding to the clause field. When the structured claim settlement field content satisfies all optional condition labels, the rule satisfaction degree is recorded as 1, otherwise as 0.

[0103] The present embodiment introduces a multi-round rule-enhanced BERT matching network to model the deep matching between the context semantic representation vector and the claim settlement judgment basis set, and fuses the field alignment and rule label guided scoring mechanism. This not only improves the semantic accuracy of the clause applicability determination, but also enhances the structured logical judgment ability of the compensation conditions and the exclusion liability, effectively supporting the clause adaptation and intelligent decision-making process in high complexity scenarios.

[0104] In the present embodiment, the multi-task semantic driven compensation determination network specifically includes:

[0105] A multi-task semantic driven compensation determination network is constructed, which includes a compensation judgment sub-network, an amount estimation sub-network and a compensation reason label generation sub-network.

[0106] The structured claim settlement text information, the clause matching score and the satisfaction judgment result are spliced according to the field dimension to form a joint semantic feature vector.

[0107] The joint semantic feature vector is input into the compensation judgment sub-network, and the compensation judgment sub-network adopts a double-tower semantic classification structure, each tower being composed of two layers of Transformer encoder and feedforward network. By respectively modeling the claim settlement text semantics and the clause matching information, and introducing a cross-attention mechanism to establish the coupling relationship between the compensation semantics and the judgment basis, the prediction result of whether to compensate is output.

[0108] The field information related to the amount in the joint semantic feature vector is input into the amount estimation subnetwork, and the amount estimation subnetwork sequentially transforms the features of the amount field, treatment method, and payment ratio through a fully connected layer and a ReLU activation function, and outputs a predicted value of the recommended payment amount.

[0109] The joint semantic feature vector is input into the payment reason label generation subnetwork, which specifically performs nonlinear feature transformation on the joint semantic feature vector through a one-layer feedforward fully connected network, and outputs a payment reason label processed by a Sigmoid activation function.

[0110] The present embodiment improves the accuracy and robustness of payment determination by constructing a multi-task semantic-driven payment determination network, fusing the matching semantics of reasonable payment text and clauses through a double-tower structure, introducing an amount estimation and reason label generation subnetwork to realize intelligent prediction of payment amount and explainable output of reasons, and enhancing the fine processing ability and decision transparency of the intelligent claims settlement system, effectively supporting efficient and reliable claims settlement assistance services.

[0111] In the present embodiment, the structured claims settlement assistance report generated according to the prediction result of whether to pay, the recommended payment amount, and the payment reason label includes:

[0112] The prediction result of whether to pay, the recommended payment amount, and the payment reason label are integrated to generate a claims determination summary;

[0113] The clause explanation information corresponding to the payment reason label is generated in combination with the clause matching score and the satisfaction judgment result;

[0114] In the case that the payment reason label includes a refusal or risk-related label, structured risk prompt information is generated;

[0115] According to the preset claims settlement assistance report format template, a structured claims settlement assistance report is constructed, which includes a claims settlement application unique index, insurance product clause information, claims determination summary, recommended payment amount, and clause explanation information.

[0116] The present embodiment generates a structured claims settlement assistance report, which uniformly organizes and formats the payment prediction result, amount estimation value, and payment reason label, significantly improves the readability and review efficiency of claims settlement decision output, generates a clear semantic chain in combination with the structure field and matching basis, facilitates verification and traceability, effectively assists claims settlement auditors to quickly understand the model determination logic, and enhances the human-machine collaboration ability and business adaptability of the intelligent claims settlement system.

[0117] Example 1:

[0118] To verify the feasibility of the present application in implementation, the present application is applied to the actual business process of a certain large commercial health insurance company in a certain geographic claim service center, and is deployed and piloted around the high-frequency claim scenario of outpatient and emergency expense reimbursement. The number of outpatient and emergency claim applications received by the insurance company per day exceeds 2000, which contains a large amount of mixed structured and unstructured content, including image bills, hospital diagnosis sheets, treatment expense lists, and corresponding clause texts. In the traditional manual review process, the claim officer needs to manually check the image information and the policy clauses, the process is tedious and the accuracy depends on experience, and it is easy to cause underpayment, overpayment or inconsistent claim opinions.

[0119] In the pilot, an intelligent claim assistance system based on artificial intelligence proposed by the present application is deployed to process the claim bill image and insurance clause information uploaded by the insured. The system automatically collects image data and clause content, and assigns a unique index identifier. In the image processing link, the improved PaddleOCR model is used to complete the structure positioning, text recognition and term correction of medical bill images, and extract key fields such as hospital, visit time, diagnosis result, expense item and amount. Then the extracted structured text is input into the context-enhanced feature encoder to obtain the context semantic vector, and combined with the claim judgment basis set generated by the clause analysis module, the clause matching task is completed through multiple rounds of rule-enhanced BERT matching network.

[0120] The matching score and judgment result are combined into the payment judgment module, and the system outputs whether to pay, the recommended payment amount and the payment reason label based on the double-tower structure and semantic cross-modeling method, and generates a structured claim assistance report by the auxiliary report module for the claim officer to review and make decisions.

[0121] To further quantify the performance of the present application in real business processes, key indicators such as processing efficiency, payment accuracy and manual review burden before and after the pilot are recorded, and the statistical results are shown in Table 1:

[0122] Table 1 Comparison of effect data before and after deployment of intelligent claim assistance system

[0123] Indicator category Before deployment (manual processing) After deployment (system assisted by the present application) Promotion range Daily average processing applications 2080 3420 +64.4％ Average processing time 18.6 minutes 7.2 minutes -61.3％ Clause matching accuracy 82.5％ 95.1％ +12.6％ Suggested compensation consistency 79.3％ 92.4％ +13.1％ Manual review ratio 100％ 36.7％ -63.3％ User complaint rate 4.8％ 1.5％ -3.3％ Report generation time Not applicable 1.2 minutes Significant optimization

[0124] From the data in Table 1, after the system is deployed, the number of claim processing, processing time and review efficiency are significantly improved. The daily application processing capacity is increased by 64.4%, which shows that the system has strong concurrent processing capacity and stability; the average processing time of each claim is shortened to 38.7% of the original, which greatly improves the operation efficiency. Through the clause structure analysis and semantic alignment enhancement mechanism, the clause matching accuracy is increased to 95.1%, the consistency of the recommended compensation result and the artificial judgment is increased by 13.1 percentage points, which significantly enhances the performance of the system in complex text understanding and semantic reasoning.

[0125] In the aspect of artificial review, since the system can directly output whether to compensate, the recommended amount and the reason label, a large number of routine cases realize automatic preliminary judgment, only 36.7% of the applications need to enter the artificial review process, effectively relieving the work intensity of the claim personnel. Especially by introducing the structured presentation of the compensation reason label, the system can clearly give the clause basis for refusing or limiting compensation, reducing user doubts and disputes, so that the user complaint rate is reduced from 4.8% to 1.5%, the claim transparency and service satisfaction are greatly improved.

[0126] The report automatic generation capability also shows strong advantages. The system can automatically summarize the prediction results and matching clause content, output complete claim assistance report combining with standard template, greatly reduce the traditional report writing and checking time, improve the case closing efficiency, and show that the system has good promotion adaptability and landing value in actual business.

[0127] The embodiment effectively verifies the intelligent processing capability of the intelligent claim assistance system based on artificial intelligence in the whole process of image recognition, semantic matching, clause reasoning and report generation, significantly improves the claim efficiency, accuracy and user experience, and has good engineering implementability and commercial application prospect.

[0128] The above describes only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. An intelligent claims assistance system based on artificial intelligence, characterized in that, Includes the following modules: The data acquisition module is used to collect the claim image data submitted by the insured and the corresponding insurance product terms and conditions, and to assign a unique index identifier to each claim application. The terms and conditions parsing module is used to perform structured parsing of the terms and conditions of insurance products and generate a set of criteria for claims judgment. The text recognition module is used for text recognition and outputs structured claims text information; The semantic encoding module is used to perform context-enhanced feature encoding on structured claims text information to obtain the context semantic representation vector of each claims text content; The clause matching module is used to perform clause matching tasks and output clause matching scores and satisfaction status results; The compensation determination module is used to perform compensation determination and amount estimation tasks, and output the prediction result of whether to pay, the suggested compensation amount, and the compensation reason label; The auxiliary report generation module is used to generate structured claims auxiliary reports.

2. The intelligent claims assistance system based on artificial intelligence according to claim 1, characterized in that, The modules are connected in the following way: Collect the image data of the claim submitted by the insured and the corresponding insurance product terms, and assign a unique index identifier to each claim application; The terms and conditions of insurance products are analyzed in a structured manner to generate a set of criteria for claims judgment; The claim image data is input into the improved PaddleOCR model for text recognition to obtain structured claim text information; The structured claims text information is semantically encoded to obtain the contextual semantic representation vector of each claims text content; The context semantic representation vector and the set of claims judgment criteria are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task and output the clause matching score and the result of the satisfaction status judgment. The structured claims text information, clause matching score, and satisfaction judgment result are jointly input into a multi-task semantic-driven claims determination network to perform claims determination and amount estimation tasks, and output the prediction result of whether to pay, the suggested payout amount, and the claim reason label. A structured claims assistance report is generated based on the predicted payout amount, the suggested payout amount, and the payout reason label.

3. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The structured parsing of insurance product terms and conditions to generate a set of claims judgment criteria specifically includes: The content of insurance product terms is structured by combining rule template parsing and syntactic dependency analysis. The rule template parsing extracts fields such as compensation liability, triggering conditions, compensation ratio, exclusions, and supporting documentation requirements by matching preset insurance term language patterns. If the rule template cannot be matched, the syntactic dependency analysis method is used. Based on verb center structure recognition and grammatical dependency path calculation, the subject-verb-object structure and modifiers related to the compensation conditions are automatically extracted from the content of the insurance product terms and conditions to complete the field extraction results. Each extracted field is assigned a field identifier and standard semantic tag according to a preset field classification rule to construct a set of clause structure fields. Based on the set of clause structure fields, the fields are grouped according to the similarity of compensation type, the consistency of the insured subject, and the logical dependency of the clause. The semantic association matrix between the fields is constructed by calculating the cosine similarity of sentence vectors, and a set of claims judgment criteria with structural organization and semantic mapping relationship is generated.

4. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The process of inputting claim image data into the improved PaddleOCR model for text recognition to obtain structured claim text information specifically includes: An improved PaddleOCR model is constructed, which includes a ticket structure localization module, a text detection subnetwork, a text orientation classification subnetwork, and a character recognition subnetwork. The input claim image data is preprocessed through the bill structure localization module. Specifically, this includes performing page structure segmentation, locating the main area of ​​the medical bill, cropping interfering content at the image edges, and extracting suspected information block areas as candidate areas for structured recognition. The structured recognition candidate regions are input into the text detection subnetwork, which adopts a detection structure based on differential feature pyramids to extract multi-scale text bounding boxes and output detection results including text region coordinates, confidence scores and candidate order labels. The detection results are input into the text orientation classification sub-network, which uses a lightweight convolutional structure to determine the orientation angle of each text candidate box and performs orientation rotation correction on the detected text image regions to make the text images be arranged in a uniform orientation. The corrected text image region is input into the text recognition sub-network. The convolutional feature extraction structure, sequence modeling structure and character decoding structure complete the text image feature extraction, sequence learning and text recognition operations in sequence. During the character decoding process, a medical terminology dictionary error correction mechanism is introduced. The candidate character sequence in the text recognition result is matched with the preset medical terminology dictionary. When there is a significant difference between the recognition result and the dictionary entry, it is replaced with the most similar standard term. The final output is structured claim text information, including the time of visit, the hospital visited, the diagnosis, and the fields for cost items and amount.

5. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The step of semantically encoding the structured claims text information to obtain the contextual semantic representation vector of each claims text content specifically includes: The medical time, medical institution, diagnosis, medical items and amount information in the structured claim text are concatenated according to the preset field order to construct a standard input text sequence; The input text sequence is fed into a context-enhanced feature encoder for semantic encoding. The context-enhanced feature encoder includes a character-level convolutional feature extraction layer, a bidirectional long short-term memory network modeling layer, and a semantic attention aggregation layer. The semantic attention aggregation layer performs weighted combination of temporal hidden states to generate a fused global contextual semantic representation vector.

6. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The BERT matching network, which inputs the context semantic representation vector and the set of claims judgment criteria into a multi-round rule-enhanced system, performs a clause matching task and outputs a clause matching score and a satisfaction judgment result, specifically including: The context semantic representation vector and the set of claims judgment criteria are input into a multi-round rule-enhanced BERT matching network to perform a clause matching task and output the clause matching score and the result of the satisfaction status judgment. The multi-round rule-enhanced BERT matching network includes a multi-round semantic interaction modeling layer, a structural field alignment mechanism, and a rule label-guided scoring module; The multi-round semantic interaction modeling layer is based on a dual-input attention mechanism. In each round, it constructs a cross-attention weight matrix between the context semantic representation vector and the clause field semantic vector, and iteratively updates the matching feature representation in multiple rounds to extract matching relationships at different semantic granularities. The structure field alignment mechanism aligns the structure fields in the structured claims text information with the semantic fields in the claims judgment basis set according to the preset field label mapping rules, and constructs a set of field-level semantic alignment vectors. The rule-label-guided scoring module inputs the rule label representations of the field-level semantic alignment vector set and the claim judgment basis set, including the claim conditions field, exclusion liability field, and supporting document requirement field, into the scoring function. Based on cosine semantic similarity and rule satisfaction, it calculates the matching score and satisfaction judgment result for each clause. The rule satisfaction is logically judged based on the claim judgment basis set corresponding to the clause field. When the structured claim field content satisfies all the required condition labels, the rule satisfaction is recorded as 1; otherwise, it is recorded as 0.

7. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The multi-task semantic-driven compensation determination network specifically includes: A multi-task semantic-driven compensation determination network is constructed, which includes a compensation determination subnetwork, an amount estimation subnetwork, and a compensation reason label generation subnetwork. The structured claims text information, clause matching scores, and satisfaction judgment results are concatenated by field dimension to form a joint semantic feature vector; The joint semantic feature vector is input into the compensation judgment sub-network, which adopts a dual-tower semantic classification structure. Each tower consists of two layers of Transformer encoder and feedforward network. By modeling the semantics of the claims text and the clause matching information respectively, and introducing a cross-attention mechanism to establish the coupling relationship between the semantics of compensation and the judgment basis, the prediction result of whether to pay compensation is output. The information of the amount-related fields in the joint semantic feature vector is input into the amount estimation sub-network. The amount estimation sub-network sequentially performs feature transformation on the amount field, treatment method, and compensation ratio through a fully connected layer and a ReLU activation function, and outputs the suggested compensation amount prediction value. The joint semantic feature vector is input into the compensation reason label generation sub-network. Specifically, the compensation reason label generation sub-network inputs the joint semantic feature vector into a one-layer feedforward fully connected network to perform nonlinear feature transformation and outputs the compensation reason label after being processed by the Sigmoid activation function.

8. The intelligent claims assistance system based on artificial intelligence according to claim 2, characterized in that, The process of generating a structured claims assistance report based on the prediction of whether compensation will be paid, the suggested compensation amount, and the claim reason label specifically includes: The prediction of whether compensation will be paid, the suggested compensation amount, and the compensation reason label are integrated to generate a summary of the claims determination. Based on the clause matching score and the satisfaction judgment result, generate clause explanation information corresponding to the claim reason label; If the claim reason label includes a claim rejection or risk-related label, generate structured risk warning information; Based on the preset claims assistance report format template, a structured claims assistance report is constructed, which includes a unique index identifier for the claims application, insurance product terms and conditions information, a summary of the claims determination, a recommended compensation amount, and explanatory information on the terms and conditions.

Images