Data analysis method and device, electronic equipment, storage medium and program product

Abstract

The present disclosure provides a data analysis method, device, electronic equipment, storage medium and program product, the method comprising: obtaining input data; according to the input data, using graph sampling and aggregation model, data matching is carried out in knowledge graph, according to the matching result, vector representation conversion and text representation conversion are carried out, the vector representation and the text representation are spliced to obtain intermediate data; determining whether the intermediate data conforms to the set rule; in response to the intermediate data conforming to the set rule, determining the calculation rule of the analysis result in the knowledge graph according to the intermediate data, determining the analysis result according to the calculation rule and the intermediate data, outputting the analysis result.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a data analysis method, apparatus, electronic device, storage medium, and program product. Background Technology

[0002] With the rapid development of the insurance industry, insurance claims processing is gradually shifting from traditional manual review to intelligent and automated processes, driven by the rapid development of digitalization.

[0003] However, although related technologies have achieved a certain degree of automation in their respective fields, the overall framework still relies heavily on manual review. For example, in the traditional claims process, from the user submitting materials to the final determination of whether a claim can be made, the basis for the claim, and the claim result, the reliance on human intervention is still relatively high, which greatly affects the overall efficiency and accuracy.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] In view of this, the present disclosure provides a data analysis method, apparatus, electronic device, storage medium, and program product to solve or partially solve the above-mentioned problems.

[0006] To achieve the above objectives, this disclosure provides a data analysis method, including:

[0007] Get the input data;

[0008] Based on the input data, a graph sampling and aggregation model is used to perform data matching in the knowledge graph. Based on the matching results, vector representation transformation and text representation transformation are performed, and the vector representation and text representation are concatenated to obtain intermediate data.

[0009] Determine whether the intermediate data conforms to the set rules;

[0010] In response to the intermediate data conforming to the set rules, the calculation rules for the analysis results are determined in the knowledge graph based on the intermediate data, the analysis results are determined based on the calculation rules and the intermediate data, and the analysis results are output.

[0011] In some exemplary embodiments, prior to data matching in the knowledge graph, the method further includes:

[0012] Determine the basic data required for the knowledge graph;

[0013] The system obtains structured data that meets the set conditions from the basic data through a preset interface, and obtains unstructured data that does not meet the set conditions from the basic data using a web crawler framework, thereby obtaining all the basic data.

[0014] In response to the completion of acquiring the basic data, the basic data is encoded using a pre-trained model, and the encoding result is extracted using a bidirectional long short-term memory model to obtain the recognition result;

[0015] The recognition results are classified into relationships using an attention mechanism to generate a triplet structure for assembling the knowledge graph.

[0016] The knowledge graph is generated by collecting and storing the triplet structures using a graph database.

[0017] In some exemplary embodiments, after performing data matching in the knowledge graph, the method further includes:

[0018] The confidence level of the matching results is determined by a path sorting algorithm;

[0019] The matching results are weighted based on the confidence level combined with an attention mechanism.

[0020] In some exemplary embodiments, after concatenating the vector representation and the text representation to obtain intermediate data, the method further includes:

[0021] A fact consistency check is performed between the intermediate data and the corresponding data in the knowledge graph.

[0022] In some exemplary embodiments, after the intermediate data conforms to a set rule, the method further includes:

[0023] Determine whether the intermediate data contains all required information;

[0024] In response to the absence of at least one required field, the acquisition channel for the missing at least one required field is determined, and the missing at least one required field is completed through the acquisition channel.

[0025] In some exemplary embodiments, before acquiring the input data, the method further includes:

[0026] In response to receiving a user's question request, the question information is determined based on the question request;

[0027] Based on the question information, a similarity query is performed in the knowledge graph using an embedding model;

[0028] Generate and output the answer information based on the query results.

[0029] Based on the same concept, this disclosure also provides a data analysis apparatus, including:

[0030] The first module is used to acquire input data;

[0031] The second module is used to perform data matching in the knowledge graph based on the input data using a graph sampling and aggregation model, perform vector representation conversion and text representation conversion based on the matching results, and concatenate the vector representation and the text representation to obtain intermediate data.

[0032] The third module is used to determine whether the intermediate data conforms to the set rules;

[0033] The fourth module is used to respond to the intermediate data conforming to the set rules, determine the calculation rules for the analysis results in the knowledge graph based on the intermediate data, determine the analysis results based on the calculation rules and the intermediate data, and output the analysis results.

[0034] Based on the same concept, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as described in any of the preceding claims.

[0035] Based on the same concept, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method as described in any of the preceding claims.

[0036] Based on the same concept, this disclosure also provides a computer program product, including computer program instructions that, when run on a computer, cause the computer to perform the method as described in any of the preceding claims.

[0037] As can be seen from the above, this disclosure provides a data analysis method, apparatus, electronic device, storage medium, and program product. The method includes: acquiring input data; performing data matching in a knowledge graph using a graph sampling and aggregation model based on the input data; performing vector representation transformation and text representation transformation based on the matching results; concatenating the vector representation and the text representation to obtain intermediate data; determining whether the intermediate data conforms to a set rule; in response to the intermediate data conforming to the set rule, determining the calculation rule for the analysis result in the knowledge graph based on the intermediate data; determining the analysis result based on the calculation rule and the intermediate data; and outputting the analysis result. This disclosure uses a graph sampling and aggregation model to accurately retrieve and match information between input data and a knowledge graph, thereby improving the accuracy of data retrieval. The retrieval results are then transformed into both vector and text representations. When the results meet subsequent requirements, vector representation allows for rapid data matching and calculation; when they do not meet subsequent requirements, corresponding text representation can be output to assist manual review and rapid problem identification. Subsequently, when the intermediate data meets the corresponding rules, the calculation rule requirements that it meets can be determined based on the intermediate data. This determines the various specific conditions that the intermediate data needs to meet when calculating the analysis results. Finally, the analysis results are calculated using the intermediate data. This allows for the fully automated generation of analysis results corresponding to the input data, making the entire process more reliant on automated programs and improving overall processing efficiency and accuracy. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments or related technologies of this disclosure, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A flowchart illustrating an exemplary method provided in an embodiment of this disclosure.

[0040] Figure 2 This is a flowchart illustrating an exemplary method for intelligent question answering provided in an embodiment of this disclosure.

[0041] Figure 3 This is a schematic flowchart illustrating the exemplary method for creating a knowledge graph provided in this embodiment of the disclosure.

[0042] Figure 4 A schematic diagram of the structure of an exemplary device provided in an embodiment of this disclosure.

[0043] Figure 5This is a schematic diagram of the electronic device structure provided in an embodiment of this disclosure. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this specification clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0045] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element, object, or method step preceding the term covers the element, object, or method step listed after the term and its equivalents, but does not exclude other elements, objects, or method steps. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0046] As described in the background section, with the rapid development of the insurance industry, traditional claims processes mainly rely on manual review, leading to problems such as large data volumes, low processing efficiency, high labor costs, and susceptibility to errors. Driven by rapid digitalization, insurance claims processes are gradually shifting from traditional manual review to intelligent and automated processes. Early claims processing relied primarily on paper documents and manual review, which had limited processing capacity and was time-consuming and labor-intensive. With the rise of big data and artificial intelligence technologies, insurance companies have begun to utilize machine learning algorithms to optimize claims processes, such as using OCR technology to identify key information in claims materials or using natural language processing technology to analyze medical reports. Currently, existing technologies in the insurance claims field mainly include rule engine-based automation technology and machine learning-based intelligent review technology. Rule engine technology processes claims applications through predefined logical rules, such as determining whether a disease is covered or calculating the compensation amount. The advantages of this technology are logical transparency and ease of maintenance, but its disadvantages include poor flexibility and difficulty in handling complex or marginal cases.

[0047] Another common technology is intelligent review based on machine learning, which trains models to identify key information in claim materials and make decisions. This type of technology typically combines OCR technology to extract text features or uses natural language processing to parse medical reports. Its advantage lies in its ability to handle unstructured data, but it relies on large amounts of labeled data, and the interpretability of the models is relatively poor.

[0048] In addition, some insurance companies have attempted to introduce knowledge graph technology to construct the relationship between diseases and insurance liabilities in order to support more complex reasoning. However, existing knowledge graphs are usually limited to a single data source (such as insurance terms) and lack dynamic connections with medical data and customer information, resulting in limited reasoning capabilities.

[0049] In specific application scenarios, some embodiments of intelligent claims technology solutions mainly include the following: (1) Automated claims technology based on rule engines: mainly processes claims applications through predefined rules (such as the mapping between disease codes and insurance liabilities). For example, when a customer submits a diagnosis certificate, the system will query the rule base to determine whether the disease is within the coverage scope and calculate the compensation amount according to the rules. Its core components include the rule base, query engine and user interface. (2) Intelligent review technology based on machine learning: the system uses supervised learning models to classify and make decisions on claims materials. For example, it extracts the cost information in medical invoices through OCR technology and inputs it into the model to predict the compensation amount. Its core process includes data preprocessing, feature extraction, model training and prediction. (3) Intelligent question answering technology based on knowledge graphs: intelligent question answering based on knowledge graphs graphs the information such as liability, disease, premium, and insurance period in insurance terms, and quickly realizes knowledge query through question mapping and knowledge reasoning, effectively reducing the cost for consumers to obtain product information and reducing the operating costs of customer service personnel.

[0050] Based on this, the applicant analyzed the above embodiments and found that the above embodiments have some shortcomings and deficiencies, mainly including: (1) Limitations of the rule engine: The rule engine relies on manually written logic, which is difficult to cover all possible claims scenarios, especially complex or marginal cases. For example, for rare diseases or multiple complications, the rule engine may not be able to accurately match insurance liability. (2) Deficiencies of machine learning models: Machine learning models require a large amount of labeled data, and the decision-making process lacks transparency, making it difficult for insurance companies and customers to trust the results. In addition, the model's sensitivity to data distribution may lead to insufficient generalization ability. (3) Static nature of knowledge graphs: Existing knowledge graphs are usually built based on static data, which is difficult to update in real time to reflect changes in insurance terms or medical standards. For example, the emergence of new drugs or treatment technologies may cause the relationships in the graph to become invalid. (4) Cross-institutional data barriers: Claims involve data from multiple parties such as hospitals, medical insurance, and insurance companies. Existing technologies are difficult to achieve efficient data sharing and integration, resulting in information silos. (5) Insufficient user experience: Existing systems usually lack interactivity, and customers cannot understand the progress of claims or obtain personalized suggestions in real time, which reduces satisfaction.

[0051] In light of the aforementioned practical situation and to address some of the problems, this disclosure provides a data analysis method. This disclosure utilizes graph sampling and aggregation models to accurately retrieve and match information between input data and knowledge graphs, thereby improving the accuracy of data retrieval. The retrieval results are then converted into both vector and text representations. If the results meet subsequent requirements, vector representation allows for rapid data matching and calculation. If they do not meet requirements, corresponding text representations are output to aid in manual review and rapid problem identification. Furthermore, when intermediate data conforms to relevant rules, the calculation rules it meets can be determined, thus identifying the specific conditions that the intermediate data must satisfy when calculating the analysis results. Finally, the analysis results are calculated using the intermediate data. This fully automated generation of analysis results corresponding to the input data allows for a significant degree of reliance on automated programs, improving overall processing efficiency and accuracy.

[0052] Figure 1 A flowchart illustrating an exemplary method provided by an embodiment of this disclosure is shown.

[0053] like Figure 1 As shown in the embodiments of this disclosure, the data analysis method specifically includes the following steps.

[0054] Step 102: Obtain input data.

[0055] In this step, the input data can be basic data used for data analysis. In specific medical claims applications, this data can include the user's basic data, medical records, treatment-related information, etc. In specific applications, the user can input the relevant data into the system through a corresponding interface, or the system can obtain the data from the internet or third parties through various channels with authorization.

[0056] In some embodiments, due to varying levels of understanding of specific data analysis processes—for example, different users' understanding of medical claims—they may be unable to determine the types of data required or predict the potential claims outcome. Consequently, users may have questions about different details of the overall process. Therefore, this embodiment can provide users with a pre-emptive intelligent Q&A session to help them understand the overall processing or answer their specific questions. Figure 2As shown, when a user uses the system, they typically first provide a specific question, which serves as the query request. Operations are then performed based on this request to obtain the final answer. After receiving the query request, the question information can be determined. In specific applications, the Fine-tune medical version of the BERT model can be used to identify various broad intents (such as "insurance coverage query," "claim progress," etc.). Then, ElasticSearch can be used to build a medical terminology thesaurus, standardizing the user's expression "myocardial infarction" into a "myocardial infarction" node in the knowledge graph. Next, the knowledge graph can be used to perform information retrieval queries using various methods. For example, based on the obtained question information, keyword matching can be used to directly match specific insurance clause nodes; similarity-based similarity searches can be performed using embedding models to find related disease treatment plans; and predefined meta-paths such as "disease → treatment → drug → side effects" can be used to expand the query. The embedding model here can be a model that establishes a text-vector mapping relationship, such as an embedding model. Finally, the answer information is generated based on the query results and output. In specific scenarios, output can be provided in a multimodal form, such as automatically generating answers with charts (such as insurance liability comparison tables) for complex questions, and directly returning text for simple questions. Afterwards, a certain traceability mechanism can be provided, such as marking the source information at the end of the answer as "according to Article 3.2 of the XX insurance terms", which meets the compliance requirements of the financial industry. For example, when a user asks "Is breast cancer targeted therapy covered by insurance?": (1) The question-and-answer interface identifies the intent as "insurance liability query", and the entities as "breast cancer" and "targeted therapy". (2) GraphRAG retrieves from the knowledge graph (graph database query and subgraph retrieval): the "list of special drugs for malignant tumors" subgraph in the insurance terms knowledge graph, and the "breast cancer → HER2 positive → applicable trastuzumab" path in the medical knowledge graph. (3) The workflow engine verifies whether the drug is within the policy's effective period and calculates it in conjunction with the deductible. (4) Finally, the answer is generated (answer generation and format output): "Your policy covers trastuzumab (Herceptin), with an expected reimbursement rate of 70%. A pathology report and gene testing results are required."

[0057] That is, in some embodiments, before acquiring input data, the method further includes: in response to acquiring a user's question request, determining question information based on the question request; performing a similarity query in the knowledge graph using an embedding model based on the question information; generating answer information based on the query result, and outputting the answer information. Here, performing a similarity query in the knowledge graph using an embedding model based on the question information corresponds to "finding related disease treatment plans through similarity using an embedding model" in the aforementioned embodiments.

[0058] Step 104: Based on the input data, use a graph sampling and aggregation model to perform data matching in the knowledge graph, perform vector representation transformation and text representation transformation based on the matching results, and concatenate the vector representation and the text representation to obtain intermediate data.

[0059] In this step, the graph sampling and aggregation model can be a GNN model such as GraphSAGE. GraphSAGE is a graph neural network model for learning graph node embeddings, particularly suitable for inductive learning, capable of generating embedding vectors on unseen graph nodes. The knowledge graph typically consists of a network composed of multiple subgraphs. In this scenario, a subgraph can be understood as information corresponding to a specific keyword. In application, the input information is generally matched with relevant subgraphs in the knowledge graph. In specific application scenarios, GNN models such as GraphSAGE can be used to perform graph embedding on the input data, locate relevant subgraphs in the knowledge graph, and support 2- to 3-hop reasoning paths (e.g., "diabetes → complications → retinopathy").

[0060] After matching the corresponding results, the matching results can be transformed into vector representation and text representation respectively. Specifically, the results can be transformed into vector representation using the Graph2Seq model and into text representation using pre-trained generation models such as T5 / BART. Then, the two representations can be concatenated to generate intermediate data. For example, Cross-Attention can be used to fusion and generate intermediate data.

[0061] In some embodiments, the knowledge graph construction process, such as Figure 3As shown, the first step is to determine the basic data required for the knowledge graph. When establishing a knowledge graph, the staff will inevitably have a plan regarding its purpose, which will then determine the basic data that needs to be included. Since the storage formats of basic data may differ across different fields (e.g., hospitals, medical insurance, and insurance companies have inconsistent data standards), it is necessary to first uniformly acquire, organize, and clean the data. Here, the required data format can be pre-defined as a condition. If some basic data meets the pre-defined conditions (i.e., it is structured data), it can be directly obtained from the corresponding database using data interfaces. Conversely, if some basic data does not meet the pre-defined conditions (i.e., it is unstructured data), web crawlers can be used to obtain the data content, which can then be output in a standard format. For example, it interfaces with hospital systems, medical insurance platforms, and core insurance systems, using medical data standards such as HL7 / FHIR and obtaining structured data such as patient records and insurance terms through OAuth 2.0 authentication. For unstructured data such as medical forums and drug instructions, it uses the Scrapy framework to build a distributed crawler, combined with BeautifulSoup to parse HTML, employing dynamic IP proxies and request frequency control to circumvent anti-crawling strategies. This method ensures the acquisition of all basic data.

[0062] After completing the basic data acquisition, collection, and cleaning, entity recognition, association, and knowledge fusion can be performed. This involves contextual encoding of the basic data using a pre-trained model, capturing sequence features through a bidirectional long short-term memory model, and then extracting features. In specific application scenarios, a BERT pre-trained model can be used for contextual encoding, a BiLSTM model can capture sequence features, and a CRF layer (Conditional Random Field) can be used to constrain label transition rules to identify entities such as diseases, drugs, and surgeries. This yields the entity recognition results. Next, an attention-based relation classification model can be used for relation classification, extracting <entity 1, relation, entity 2> triple structures from corpora such as "patient-taking-drug," supporting medical-specific relation types such as causal relationships and taboo relationships. Finally, a knowledge graph is generated. The extracted triple structures can be stored in graph databases such as Neo4j to form a knowledge graph. Furthermore, a native graph storage engine can be used to implement multi-hop queries, such as "querying common treatment plans for a certain disease," which requires a 3-hop traversal. Finally, in specific applications, if certain entities are identified as high-frequency entities for querying, attribute indexes can be established for these high-frequency entities, which can support complex graph analysis using Cypher / Gremlin query languages. That is, in some embodiments, before data matching in the knowledge graph, the method further includes: determining the basic data required for the knowledge graph; obtaining structured data that meets set conditions from the basic data through a preset interface, and obtaining unstructured data that does not meet the set conditions from the basic data using a web crawler framework, thereby obtaining all the basic data; in response to completing the acquisition of the basic data, encoding the basic data through a pre-trained model, and using a bidirectional long short-term memory model to perform feature recognition and extraction on the encoding results to obtain recognition results; classifying the recognition results into relationships through an attention mechanism to generate triple structures for assembling the knowledge graph; and using a graph database to collect and store the triple structures to generate the knowledge graph.

[0063] In some embodiments, when performing data matching in a knowledge graph, the resulting matching may contain multiple pieces of information, which may have varying degrees of relevance to the input data or the information the customer wants to know. Therefore, to increase the weight of information that the user is more interested in, a path ranking algorithm can be used to determine the confidence level of different matching results, and then an attention mechanism can be used to dynamically weight the importance of each subgraph. The path ranking algorithm here can be a PRA (PathRanking Algorithm), which is an algorithm used to rank nodes in a directed graph. That is, in some embodiments, after performing data matching in the knowledge graph, the method further includes: determining the confidence level of the matching results using a path ranking algorithm; and weighting the matching results based on the confidence level combined with an attention mechanism.

[0064] In some embodiments, to ensure the consistency between the output intermediate data and the data within the knowledge graph, after obtaining the intermediate data, a fact consistency detection model such as FactCC can be further used to perform fact consistency detection between the intermediate data and the data in the knowledge graph, ensuring that the generated content is consistent with the graph data. That is, in some embodiments, after concatenating the vector representation and the text representation to obtain the intermediate data, the method further includes: performing fact consistency detection between the data corresponding to the intermediate data in the knowledge graph and the intermediate data itself.

[0065] In more specific application scenarios, during the operation of this solution model, if staff determine that a certain question and answer is of high user interest but not found in the knowledge graph, a reward function can be constructed for that question and answer that includes factors such as answer accuracy, fluency, and user satisfaction (e.g., feedback from tracking points). The PPO algorithm can then be used to continuously optimize the joint performance of the retrieval and generation modules. Furthermore, to enhance the model's online learning capabilities, an A / B testing framework can be designed, where the new model is first deployed with 5% of the traffic to validate its effectiveness before being gradually rolled out to the full user base.

[0066] Step 106: Determine whether the intermediate data conforms to the set rules.

[0067] In this step, it is necessary to determine whether the intermediate data conforms to the set rules. These set rules can be requirements that the intermediate data needs to meet based on specific application scenarios. For example, in a medical claims scenario, the set rules could be the types of data that the intermediate data needs to include, or the standards that a certain data item needs to meet. If the intermediate data conforms to the set rules, proceed to step 108; if the intermediate data does not conform to the set rules, it can be returned to the user side for supplementation, or returned to the manual review side for determination of specific issues through manual review.

[0068] Step 108: In response to the intermediate data conforming to the set rules, determine the calculation rules for the analysis results in the knowledge graph based on the intermediate data, determine the analysis results based on the calculation rules and the intermediate data, and output the analysis results.

[0069] In this step, once the intermediate data conforms to the set rules, the calculation rules for the analysis results can be determined in the knowledge graph based on the intermediate data. In specific application scenarios, information extraction can first be performed based on the document understanding model of the LayoutLMv3 model, supporting the recognition of complex formats such as medical invoices and diagnostic certificates (field recognition accuracy >95%), and automatically classifying them into "outpatient claims" or "inpatient claims" types. Here, the LayoutLMv3 model can be used for OCR recognition processing (Optical Character Recognition). Then, the disease codes extracted by OCR are matched with insurance terms and other information in the knowledge graph, and the coverage is determined through predefined mapping rules (such as the correspondence table between ICD-10 and insurance diseases). Next, it is connected to the medical insurance drug catalog knowledge graph to calculate the reimbursement ratio and out-of-pocket portion within medical insurance, and generate pre-calculated results by combining the deductible and reimbursement ratio of the policy. Finally, the reimbursement details are dynamically populated based on a preset template (such as the Apache Velocity template engine), supporting the generation of multi-language versions. Finally, the results can be sent synchronously via SMS, APP push, and email, and the case status can be updated.

[0070] In more specific application scenarios, such as the medical claims scenario described in the aforementioned embodiments, the Drools rule base can be used to implement over 2000 claims rules (e.g., "a pathology report is required for a confirmed diagnosis of thyroid cancer"), supporting real-time rule hit detection. Simultaneously, for disputed cases (e.g., suspected insurance purchase with pre-existing conditions), the XGBoost model can be used to comprehensively analyze features such as medical history and insurance purchase time to output a fraud probability score.

[0071] In some embodiments, after the intermediate data conforms to the set rules, the completeness of the data contained in the intermediate data can be determined. During the generation of analysis results, some information is necessary but not necessarily required by the user. For example, the analysis results may require the user's invoice information. Since the user provides their name and other identity information, this invoice information can be automatically determined. Therefore, invoice information is not mandatory for the user, but it is required by the analysis results. Thus, after the intermediate data conforms to the set rules, the mandatory information required for the analysis results, or the mandatory information required for subsequent rule retrieval, can be determined. Then, it is determined whether the intermediate data contains this mandatory information. If a mandatory information item is missing, the information can be retrieved through certain information in the intermediate data using the retrieval channel, thereby completing the missing mandatory information. For example, the aforementioned invoice information can be retrieved through the corresponding tax channels using the user's relevant information, thus completing the missing information. In more specific scenarios, the rule engine can check mandatory fields (such as invoice codes and diagnostic codes), and automatically trigger a completion process when missing fields (such as querying a hospital system). That is, in some embodiments, after the intermediate data conforms to the set rules, the method further includes: determining whether the intermediate data contains all required information; in response to the absence of at least one required information, determining the acquisition channel of the missing at least one required information, and completing the missing at least one required information through the acquisition channel.

[0072] Finally, the analysis results can be output, for example, displayed on a corresponding device to provide feedback to the operator. Of course, in other embodiments, the output method of the analysis results is not limited to display; it can also be used to store, display, use, or further process the analysis results. The specific output method of the analysis results can be flexibly selected according to different application scenarios and implementation needs.

[0073] Specifically, for example, in the application scenario where the method of this embodiment is executed on a single device, the analysis results can be directly output on the display component (monitor, projector, etc.) of the current device, so that the operator of the current device can directly see the content of the analysis results on the display component.

[0074] For example, in application scenarios where the method of this embodiment is executed on a system composed of multiple devices, the analysis results can be sent to other preset devices within the system, i.e., synchronization terminals, as recipients, via any data communication method (wired connection, NFC, Bluetooth, Wi-Fi, cellular network, etc.), so that the synchronization terminals can perform subsequent processing. Optionally, the synchronization terminal can be a preset server, which is generally located in the cloud and serves as a data processing and storage center, capable of storing and distributing the analysis results; wherein, the recipients of the distribution are terminal devices, and the owners or operators of these terminal devices can be administrators, supervisors, maintenance personnel, sales personnel, etc., of the medical claims system.

[0075] For example, in the application scenario where the method of this embodiment is executed on a system composed of multiple devices, the analysis results can be directly sent to a preset terminal device through any data communication method. The terminal device can be one or more of the devices listed in the preceding paragraphs.

[0076] As can be seen from the above embodiments, this disclosure provides a data analysis method, which includes: acquiring input data; performing data matching in a knowledge graph using a graph sampling and aggregation model based on the input data; performing vector representation transformation and text representation transformation based on the matching results; concatenating the vector representation and the text representation to obtain intermediate data; determining whether the intermediate data conforms to a set rule; in response to the intermediate data conforming to the set rule, determining the calculation rule for the analysis result in the knowledge graph based on the intermediate data; determining the analysis result based on the calculation rule and the intermediate data; and outputting the analysis result. This disclosure uses a graph sampling and aggregation model to accurately retrieve and match information between input data and a knowledge graph, thereby improving the accuracy of data retrieval. The retrieval results are then transformed into both vector and text representations. When the results meet subsequent requirements, vector representation allows for rapid subsequent data matching and calculation; when they do not meet subsequent requirements, corresponding text representation can be output to assist in manual review and rapid problem identification. Subsequently, when the intermediate data meets the corresponding rules, the calculation rule requirements that it meets can be determined based on the intermediate data. This determines the various specific conditions that the intermediate data needs to meet when calculating the analysis results. Finally, the analysis results are calculated using the intermediate data. This allows for the fully automated generation of analysis results corresponding to the input data, making the entire process more reliant on automated programs and improving overall processing efficiency and accuracy.

[0077] In specific application scenarios, this disclosure presents a knowledge graph-based intelligent insurance claims model. This model constructs multi-dimensional knowledge graphs encompassing insurance products, medical diagnoses, and drug catalogs, combining natural language processing, graph neural networks, and GraphRAG (Graph Retrieval Augmentation) technology to automate, intelligentize, and streamline the claims process. Application scenarios primarily include claims processing for health insurance and medical insurance products, covering core aspects such as policyholder information association, claims document identification, disease liability matching, and payout calculation. This model structures insurance terms, medical data, and customer information using knowledge graph technology, leveraging graph algorithms and machine learning models to achieve intelligent reasoning and decision-making, significantly improving claims efficiency and accuracy. It supports intelligent question answering, risk control, and competitor analysis, providing comprehensive intelligent services for insurance businesses and customers. The core technologies of this model include knowledge graph construction, GraphRAG technology, OCR recognition, and intelligent decision-making. Through the synergistic effect of these technologies, the entire process from claims document submission to payout decision-making is automated. The development direction of this technology is to further integrate multimodal data and reinforcement learning to improve the system's adaptability and intelligence level, and meet the growing personalized needs of the insurance industry.

[0078] Ultimately, through automated processing, the traditional claims processing cycle of several days has been shortened to hours, significantly improving customer satisfaction. Real-world testing data shows an over 80% increase in processing efficiency. The intelligent system reduces reliance on professional claims personnel, decreasing manual review workload by 60% and reducing repetitive work caused by human error. The intelligent question-and-answer system provides 24 / 7 professional consultation services, achieving a 92% accuracy rate in answering questions and significantly reducing customer wait times. All claims decisions are traceable to relevant nodes and rules in a knowledge graph, enhancing the explainability and credibility of the decisions.

[0079] In specific application scenarios, the embodiments of this disclosure can be applied to: auto insurance claims: constructing knowledge graphs of vehicle repair, spare parts prices, etc., to automate accident damage assessment; property insurance: integrating building structure, disaster data, etc., to assist in underwriting and claims decisions; social welfare distribution: automating the review of application eligibility; tax audits: discovering tax evasion through enterprise relationship graphs; and so on.

[0080] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this disclosure embodiment can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0081] It should be noted that the above description describes specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] Based on the same concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a data analysis apparatus.

[0083] refer to Figure 4 The data analysis device includes:

[0084] The first module 410 is used to acquire input data.

[0085] The second module 420 is used to perform data matching in the knowledge graph based on the input data using a graph sampling and aggregation model, perform vector representation transformation and text representation transformation based on the matching results, and concatenate the vector representation and the text representation to obtain intermediate data.

[0086] The third module 430 is used to determine whether the intermediate data conforms to the set rules.

[0087] The fourth module 440 is used to respond to the intermediate data conforming to the set rules, determine the calculation rules for the analysis results in the knowledge graph based on the intermediate data, determine the analysis results based on the calculation rules and the intermediate data, and output the analysis results.

[0088] In some exemplary embodiments, the second module 420 is further configured to:

[0089] Determine the basic data required for the knowledge graph;

[0090] The system obtains structured data that meets the set conditions from the basic data through a preset interface, and obtains unstructured data that does not meet the set conditions from the basic data using a web crawler framework, thereby obtaining all the basic data.

[0091] In response to the completion of acquiring the basic data, the basic data is encoded using a pre-trained model, and the encoding result is extracted using a bidirectional long short-term memory model to obtain the recognition result;

[0092] The recognition results are classified into relationships using an attention mechanism to generate a triplet structure for assembling the knowledge graph.

[0093] The knowledge graph is generated by collecting and storing the triplet structures using a graph database.

[0094] In some exemplary embodiments, the second module 420 is further configured to:

[0095] The confidence level of the matching results is determined by a path sorting algorithm;

[0096] The matching results are weighted based on the confidence level combined with an attention mechanism.

[0097] In some exemplary embodiments, the second module 420 is further configured to:

[0098] A fact consistency check is performed between the intermediate data and the corresponding data in the knowledge graph.

[0099] In some exemplary embodiments, the fourth module 440 is further configured to:

[0100] Determine whether the intermediate data contains all required information;

[0101] In response to the absence of at least one required field, the acquisition channel for the missing at least one required field is determined, and the missing at least one required field is completed through the acquisition channel.

[0102] In some exemplary embodiments, the first module 410 is further configured to:

[0103] In response to receiving a user's question request, the question information is determined based on the question request;

[0104] Based on the question information, a similarity query is performed in the knowledge graph using an embedding model;

[0105] Generate and output the answer information based on the query results.

[0106] For ease of description, the above apparatus is described in terms of its functions, divided into various modules. Of course, in implementing the embodiments of this disclosure, the functions of each module can be implemented in one or more software and / or hardware.

[0107] The apparatus described above is used to implement the corresponding data analysis methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0108] Based on the same concept, corresponding to the methods of any of the above embodiments, this disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the data analysis method as described in any of the above embodiments.

[0109] Figure 5 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0110] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0111] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0112] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0113] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0114] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0115] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0116] The electronic devices described above are used to implement the corresponding data analysis methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0117] Based on the same concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the data analysis method as described in any of the above embodiments.

[0118] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0119] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the data analysis method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0120] Based on the same concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer to cause the computer and / or the processors to perform the data analysis method. Corresponding to the execution entity for each step in each embodiment of the data analysis method, the processor executing the corresponding step may belong to the corresponding execution entity.

[0121] The computer program products of the above embodiments are used to cause the computer and / or the processor to execute the data analysis method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0122] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this disclosure as described above, which are not provided in detail for the sake of brevity.

[0123] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this disclosure, the provided drawings may or may not show well-known power / ground connections to integrated circuit (IC) chips and other components. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this disclosure, and this also takes into account the fact that the details of implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this disclosure will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuitry) have been set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that the embodiments of this disclosure may be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0124] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0125] This disclosure is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A data analysis method, characterized in that, include: Get the input data; Based on the input data, a graph sampling and aggregation model is used to perform data matching in the knowledge graph. Based on the matching results, vector representation transformation and text representation transformation are performed, and the vector representation and text representation are concatenated to obtain intermediate data. Determine whether the intermediate data conforms to the set rules; In response to the intermediate data conforming to the set rules, the calculation rules for the analysis results are determined in the knowledge graph based on the intermediate data, the analysis results are determined based on the calculation rules and the intermediate data, and the analysis results are output.

2. The method according to claim 1, characterized in that, Before performing data matching in the knowledge graph, the method further includes: Determine the basic data required for the knowledge graph; The system obtains structured data that meets the set conditions from the basic data through a preset interface, and obtains unstructured data that does not meet the set conditions from the basic data using a web crawler framework, thereby obtaining all the basic data. In response to the completion of acquiring the basic data, the basic data is encoded using a pre-trained model, and the encoding result is extracted using a bidirectional long short-term memory model to obtain the recognition result; The recognition results are classified into relationships using an attention mechanism to generate a triplet structure for assembling the knowledge graph. The knowledge graph is generated by collecting and storing the triplet structures using a graph database.

3. The method according to claim 1, characterized in that, After performing data matching in the knowledge graph, the method further includes: The confidence level of the matching results is determined by a path sorting algorithm; The matching results are weighted based on the confidence level combined with an attention mechanism.

4. The method according to claim 1, characterized in that, After concatenating the vector representation and the text representation to obtain intermediate data, the method further includes: A fact consistency check is performed between the intermediate data and the corresponding data in the knowledge graph.

5. The method according to claim 1, characterized in that, After the intermediate data conforms to the set rules, the method further includes: Determine whether the intermediate data contains all required information; In response to the absence of at least one required field, the acquisition channel for the missing at least one required field is determined, and the missing at least one required field is completed through the acquisition channel.

6. The method according to claim 1, characterized in that, Before acquiring the input data, the method further includes: In response to receiving a user's question request, the question information is determined based on the question request; Based on the question information, a similarity query is performed in the knowledge graph using an embedding model; Generate and output the answer information based on the query results.

7. A data analysis device, characterized in that, include: The first module is used to acquire input data; The second module is used to perform data matching in the knowledge graph based on the input data using a graph sampling and aggregation model, perform vector representation transformation and text representation transformation based on the matching results, and concatenate the vector representation and the text representation to obtain intermediate data. The third module is used to determine whether the intermediate data conforms to the set rules; The fourth module is used to respond to the intermediate data conforming to the set rules, determine the calculation rules for the analysis results in the knowledge graph based on the intermediate data, determine the analysis results based on the calculation rules and the intermediate data, and output the analysis results.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing a computer to perform the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer program instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 6.

Images