Similar claim settlement case determination method and device, equipment and medium

Abstract

The application belongs to the field of artificial intelligence, and relates to a similar claim settlement case determination method, comprising: receiving a claim settlement request of a claim settlement case, the claim settlement request comprising a plurality of claim settlement images of the claim settlement case; performing global feature extraction processing on the multiple claim settlement images to obtain multiple pieces of global feature information of the multiple claim settlement images; based on the multiple pieces of global feature information, determining at least one historical claim settlement case and multiple target historical claim settlement images from a claim settlement retrieval library; and carrying out aggregation processing on the plurality of claim settlement images and the plurality of target historical claim settlement images to obtain a target claim settlement case set. The invention further provides a similar claim settlement case determination device, computer equipment and a storage medium. In addition, the invention also relates to a block chain technology, and a plurality of claim settlement images of the user and a plurality of historical claim settlement images used for establishing the claim settlement retrieval library can be stored in a block chain. According to the invention, the determination efficiency of similar claim settlement cases can be improved, so that the checking efficiency and accuracy of the claim settlement cases can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for determining similar claims cases. Background Art

[0002] In the field of insurance claims, it is common to encounter clients forging insurance claims evidence, using the same case scene images to file claims for multiple insurance policies, or slightly modifying the images, modifying the date or some key information to confuse auditors. Even worse, there are studios or gangs that collect these same or similar case scene inspection photos and apply for claims at different insurance claims agencies, resulting in a significant increase in the insurance agency's claims rate and endangering social order.

[0003] The existing method to solve the above problem is to rely on manual or simple image comparison methods to judge the uploaded on-site images. However, the same insurance institution generates tens of millions of images every year. It is difficult to effectively process such a large amount of data if it relies solely on manual or simple comparison methods. The efficiency and accuracy of image comparison are not high. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a method, device, equipment and medium for determining similar claim cases, so as to solve the problem of low review efficiency and accuracy of claim cases caused by manually or simply comparing images to determine whether the insurance claim certificate uploaded by the customer is forged when the customer applies for insurance claim.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a method for determining similar claims cases, which adopts the following technical solutions:

[0006] receiving a claim request for a claim case, wherein the claim request includes a plurality of claim images of the claim case;

[0007] Performing global feature extraction processing on multiple claim images to obtain multiple global feature information of the multiple claim images;

[0008] Based on the multiple global feature information, determine at least one historical claim case from the claim retrieval library, and determine multiple target historical claim images corresponding to the at least one historical claim case;

[0009] Multiple claim images and multiple target historical claim images are aggregated to obtain a target claim case set, in which the similarity between each historical claim case and the claim case reaches a target similarity threshold.

[0010] Furthermore, before the steps of determining at least one historical claim case from the claim search library based on multiple global feature information and determining multiple target historical claim images corresponding to the at least one historical claim case, the method further includes:

[0011] Acquire historical claims data, the historical claims data including a plurality of historical claims images, a claims case number to which each historical claims image belongs, and image category information of the plurality of historical claims images;

[0012] Performing global feature extraction processing on multiple historical claims images to obtain historical global feature information of the multiple historical claims images;

[0013] Clustering the historical global feature information to obtain multiple feature clusters, each feature cluster is composed of multiple historical global feature information;

[0014] Establish the association relationship between multiple feature clusters, claim case numbers and image category information, and build a claim retrieval library based on the association relationship.

[0015] Further, based on the multiple global feature information, the steps of determining at least one historical claim case from the claim search library, and determining multiple target historical claim images corresponding to the at least one historical claim case specifically include:

[0016] Calculating the first similarity between each global feature information and each feature cluster in the claims retrieval database to obtain multiple first similarities;

[0017] Based on the multiple first similarities, determining a target feature cluster from the multiple feature clusters, wherein the first similarity between the target feature cluster and the multiple global feature information reaches a similarity threshold;

[0018] Calculating the second similarity between each historical global feature information constituting the target feature cluster and each global feature information to obtain multiple second similarities;

[0019] Based on the multiple second similarities, at least one historical claim case is determined from the claim retrieval library, and multiple target historical claim images corresponding to the at least one historical claim case are determined.

[0020] Further, based on the multiple second similarities, the steps of determining at least one historical claim case from the claim search library, and determining multiple target historical claim images corresponding to the at least one historical claim case specifically include:

[0021] Sorting the plurality of second similarities to obtain a similarity sequence;

[0022] Based on the association relationship and the similarity sequence, at least one historical claim case is determined from the claim retrieval library, and a plurality of target historical claim images corresponding to the at least one historical claim case are determined.

[0023] Furthermore, the steps of aggregating the multiple claim images and the multiple target historical claim images to obtain a target claim case set specifically include:

[0024] Based on the plurality of claim images and the plurality of target historical claim images, establishing a claim case graph between the claim case and at least one historical claim case;

[0025] Based on a preset image correction threshold, the claim case map is corrected to obtain a claim case correction map;

[0026] Using a preset search algorithm, image aggregation is performed on the claim case correction graph to obtain claim cases and at least one claim case set;

[0027] Feature aggregation is performed on the claim cases and at least one claim case set to obtain a target claim case set.

[0028] Furthermore, the claim case correction graph includes a plurality of claim case nodes and a plurality of claim case edges. The steps of performing image aggregation on the claim case correction graph using a preset search algorithm to obtain claim cases and at least one claim case set specifically include:

[0029] Using a preset search algorithm, taking any claim case node as the starting point, traverse multiple claim case nodes and multiple claim case edges to generate multiple case paths;

[0030] Based on the multiple case paths, a claim case and at least one claim case set are determined.

[0031] Furthermore, the step of performing feature aggregation on the claim cases and at least one claim case set to obtain a target claim case set specifically includes:

[0032] Based on multiple claim images and global feature information, the mean clustering features of claim cases are calculated;

[0033] Calculate historical mean clustering features of each historical claim case in at least one case set based on a plurality of historical claim images and historical local feature information of historical claim cases in at least one case set;

[0034] The feature similarity between each historical mean clustering feature and the mean clustering feature is calculated respectively to obtain multiple feature similarities;

[0035] The case set of historical claims cases whose feature similarity is greater than a similarity threshold is determined as a similar case set of the claims case;

[0036] Merge the claim cases into similar case sets to obtain the target claim case set.

[0037] Furthermore, based on multiple claim images and global feature information, the step of calculating the mean clustering feature of the claim case specifically includes:

[0038] Extracting local features of multiple claim images to obtain local feature information of the multiple claim images;

[0039] The local feature information and the global feature information are spliced ​​to obtain the target global feature information;

[0040] The target global feature information and local feature information are weighted to obtain the mean clustering features of the claim cases.

[0041] Furthermore, based on a plurality of historical claim images and historical local feature information of historical claim cases in at least one case set, the step of calculating the historical mean clustering feature of each historical claim case in at least one case set specifically includes:

[0042] Extracting historical local features of multiple historical claims images of each historical claims case in at least one case set respectively, to obtain historical local feature information of each historical claims case;

[0043] splicing the historical local feature information and multiple historical global feature information of historical claims cases in at least one case set to obtain target historical global feature information;

[0044] A weighted operation is performed on the target historical global feature information and the historical local feature information to obtain the historical mean clustering features of each historical claim case in at least one case set.

[0045] In order to solve the above technical problems, the embodiment of the present application further provides a device for determining similar claims cases, which adopts the following technical solutions:

[0046] A request receiving module, used for receiving a claim request of a claim case, wherein the claim request includes a plurality of claim images of the claim case;

[0047] A first feature extraction module is used to perform global feature extraction processing on multiple claim settlement images to obtain multiple global feature information of the multiple claim settlement images;

[0048] A case determination module, used to determine at least one historical claim case from a claim retrieval library based on multiple global feature information, and to determine multiple target historical claim images corresponding to the at least one historical claim case;

[0049] The case set aggregation module is used to aggregate multiple claim images and multiple target historical claim images to obtain a target claim case set, in which the similarity between each historical claim case and the claim case reaches a target similarity threshold.

[0050] In order to solve the above technical problems, an embodiment of the present application also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for determining similar claims cases as described above are implemented.

[0051] In order to solve the above technical problems, an embodiment of the present application also provides a computer-readable storage medium, which stores computer-readable instructions. The computer-readable instructions can be executed by at least one processor to enable at least one processor to perform the steps of the method for determining similar claims cases as mentioned above.

[0052] Compared with the prior art, the embodiment of the present application has the following beneficial effects: after receiving the claim request of the claim case, the global features of the claim image can be automatically extracted and analyzed to quickly identify images that are highly similar to the historical claim cases, so that at least one historical claim case can be determined from the claim retrieval library. Then, by aggregating multiple claim images and multiple target historical claim images, a set of similar cases whose similarity to the claim case reaches the target similarity threshold is determined to form a target claim case set, which can improve the efficiency of determining similar claim cases, reduce the burden of manual review, effectively shorten the review time, and improve the efficiency and accuracy of claim case review. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0055] Figure 2 It is a flowchart of a method for determining similar claims cases provided by this application;

[0056] Figure 3 It is a schematic diagram of the correction process of a claim case graph of a method for determining similar claim cases provided by the present application;

[0057] Figure 4 It is a structural schematic diagram of a device for determining similar claims cases provided by the present application;

[0058] Figure 5 It is a structural schematic diagram of a computer device provided by this application. DETAILED DESCRIPTION

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0060] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0061] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0063] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0064] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0065] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0066] It should be noted that the method for determining similar claim cases provided in the embodiment of the present application is generally executed by a server, and accordingly, the device for determining similar claim cases is generally set in the server.

[0067] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0068] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for determining similar claim cases according to the present application. The method for determining similar claim cases comprises the following steps:

[0069] Step S201 , receiving a claim request for a claim case, where the claim request includes a plurality of claim images of the claim case.

[0070] In this embodiment, the method for determining similar claims cases is executed on an electronic device (eg, Figure 1 The server shown in the figure) can receive the claim request of the claim case through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0071] Among them, claim images refer to images uploaded by users for assisting insurance claims, such as on-site inspection images, receipt images, bill images, ID card images, bank card images, business license images, etc.

[0072] Specifically, the claim request may include the insured person information, the reporter information, the insurance contract information, the time, location and process of the accident, multiple claim images, etc. When the user submits the claim request through the online reporting system, the system can send the request information to the server, and the server can receive the claim request of the claim case; or submit the claim request through the mobile application of the insurance company, the application will send the data to the server for processing, and the server can receive the claim request of the claim case.

[0073] Step S202, performing global feature extraction processing on multiple claim settlement images to obtain multiple global feature information of the multiple claim settlement images.

[0074] Among them, global feature information is a comprehensive and integrated description of the claim image, capturing the overall structure and content of each claim image. Global feature information can characterize the uniqueness, main content and similarity of each claim image. Uniqueness means that these global features can uniquely represent a corresponding claim image, that is, different images have different representations in the feature space, which makes it easy to distinguish and identify; main content means that global features can capture key information such as the main objects, scene layout, color distribution, etc. in the image, which is crucial for understanding the image content and judging whether it is applicable to claims cases; similarity means that by comparing the global features of different images, the similarity between them can be evaluated, which is very helpful for discovering duplicate or forged claim images.

[0075] In one embodiment, the global feature information can be obtained by processing the claim image through a deep learning model. The global feature information can represent a high-dimensional vector of the overall characteristics of the image (for example, a 512-dimensional global feature vector). Each dimension of these vectors corresponds to a feature or attribute of the image in a certain aspect, and their combination together constitutes a comprehensive and accurate description of the image.

[0076] Step S203, based on the multiple global feature information, determining at least one historical claim case from the claim search library, and determining multiple target historical claim images corresponding to the at least one historical claim case.

[0077] The claim retrieval database is a database dedicated to storing and retrieving the global feature information of insurance claim images. It records the global feature vector, corresponding case number, and image category of each registered image (historical claim image), so that relevant historical claim cases and historical claim images can be quickly located in the subsequent retrieval process.

[0078] The at least one historical claim case refers to a historical claim case in the claim retrieval database that is similar or identical in global features to the claim image input by the current claim case. These cases may contain information that is of reference value to the processing of the current claim case.

[0079] The target historical claims case refers to the historical claims images that are determined to be similar or identical to the claims images input in the current claims case during the retrieval process. These images are registered in the construction of the claims retrieval library and have a unique representation of global feature information.

[0080] In one embodiment, when a claim request for a new claim case is received, all images (claim images) in the current claim case are extracted through a deep learning model to obtain multiple high-dimensional global features (for example, 512 global features). These features can use the cosine distance calculation method to calculate the similarity with all image features in the retrieval library, and finally obtain the top n most similar image retrieval results. So far, the topn retrieval results of all claim images in the new claim case are obtained. Based on the results, at least one historical claim case can be determined from the claim retrieval library, and then multiple target historical claim images corresponding to the at least one historical claim case can be obtained from the claim retrieval library.

[0081] Step S204, aggregating the multiple claim images and the multiple target historical claim images to obtain a target claim case set, wherein the similarity between each historical claim case and the claim case in the target claim case set reaches a target similarity threshold.

[0082] Among them, aggregation processing refers to constructing and modifying a claim case graph to preliminarily aggregate historical claim cases similar to the claim case, and then use the local features of the image (claim image, historical claim image) and the global features to enhance the feature representation, and further refine the aggregation by calculating the similarity between cluster centers, and finally obtain the target claim case set.

[0083] Among them, the similarity threshold is a key parameter used to determine whether the similarity between claim cases is high enough so that they can be merged into a new case set to obtain the target claim case set.

[0084] In one embodiment, a claim case graph between a claim case and at least one historical claim case can be established based on multiple claim images and multiple target historical claim images, and the claim case graph can be a weighted directed graph. Then, a preset image correction threshold is obtained to correct the claim case graph to obtain a corrected claim case graph. Then, a preset search algorithm, such as a depth-first search algorithm (DFS), is used to perform image aggregation on the corrected claim case graph to obtain a claim case and at least one claim case set, and finally, feature aggregation is performed on the claim case and at least one claim case set to obtain a target claim case set.

[0085] In one example, when a claim request for a claim case is received, the server can automatically extract global feature information of multiple claim images contained in the claim request. Global feature extraction can use a convolutional neural network (CNN) model, such as ResNet-50 or VGG-16. The extracted global feature information can usually be a 512-dimensional or 2048-dimensional feature vector. Then, the server can use the extracted global feature information to quickly find multiple historical claim cases with the highest similarity to the current claim case in a claim retrieval library containing multiple historical claim cases, and the claim images (historical claim images) corresponding to these historical claim cases by calculating the Euclidean distance or cosine similarity between the feature vectors. Next, the server can aggregate the claim images of the current claim case and the claim images of the historical claim cases, and use a hierarchical clustering algorithm, such as a bottom-up agglomerative nesting algorithm (AGNES), to classify cases with similarities above a similarity threshold into one category to form a target claim case set. Finally, the server can automatically generate similar claims case reports based on the clustering results for reference by claims specialists, greatly improving the efficiency and accuracy of claims processing.

[0086] After receiving the claim request of the claim case, the embodiment of the present application can automatically extract and analyze the global features of the claim image, and can quickly identify images that are highly similar to the historical claim cases, so as to determine at least one historical claim case from the claim retrieval library. Then, by aggregating multiple claim images and multiple target historical claim images, a set of similar cases whose similarity to the claim case reaches the target similarity threshold is determined to form a target claim case set, which can improve the efficiency of determining similar claim cases, reduce the burden of manual review, effectively shorten the review time, and improve the efficiency and accuracy of claim case review.

[0087] In some optional implementations of this embodiment, before determining at least one historical claim case from the claim search library based on multiple global feature information and determining multiple target historical claim images corresponding to the at least one historical claim case in step 203, the method further includes the following steps:

[0088] Acquire historical claims data, the historical claims data including a plurality of historical claims images, a claims case number to which each historical claims image belongs, and image category information of the plurality of historical claims images;

[0089] Performing global feature extraction processing on multiple historical claims images to obtain historical global feature information of the multiple historical claims images;

[0090] Clustering the historical global feature information to obtain multiple feature clusters, each feature cluster is composed of multiple historical global feature information;

[0091] Establish the association relationship between multiple feature clusters, claim case numbers and image category information, and build a claim retrieval library based on the association relationship.

[0092] In this embodiment, when constructing a claims retrieval library, it is first necessary to obtain a large amount of historical claims data. These data usually include more than tens of thousands of historical claims images, each of which has a corresponding claims case number and image category information, such as "vehicle damage photos", "medical bill scans", etc. After obtaining the historical data, the server can perform global feature extraction on these images. Common feature extraction models include ResNet-50, Inception-V3, etc., and the extracted feature information is usually a 512-dimensional or 2048-dimensional feature vector. Next, the server can perform cluster analysis on the extracted historical global feature information. Common clustering algorithms include K-means, DBSCAN, etc. Through clustering, similar feature vectors can be classified into one category to form multiple feature clusters. The feature vectors in each feature cluster correspond to similar claims images. After clustering is completed, the server can analyze the claims case number and image category of the image in each feature cluster, and establish the association relationship between the feature cluster, case number, and image category. Finally, this association relationship is stored in the database to form an efficient claims retrieval library.

[0093] In one example, historical claims images of historical claims data are registered and stored in the database. Before searching for similar or identical images, a search database needs to be established. The historically retained claims uploaded images plus the case number to which the images belong plus the categories of each image can be used. A 512-dimensional global feature can be extracted through a deep learning model (trained). This feature can uniquely represent a corresponding registered image. When constructing a search database, clustering and quantization methods are required for features in the millions or tens of millions. Here, a more commonly used vector search method that combines inverted index and flat vector storage, namely the inverted index method, is used to construct the search database. The search database here records the case number, features, and categories corresponding to each image. When searching in the search stage, the search results for the same case as the input image can be excluded, and the image category can be used to calculate the edge weight when constructing the directed graph in the aggregation stage.

[0094] The embodiment of the present application obtains historical claims data and establishes a claims retrieval library. When a new claim request arrives, it only needs to extract the global features of its claim image, and then perform similarity matching in the claim retrieval library to quickly find similar historical claims cases and multiple target historical claims images corresponding to historical claims cases, greatly improving the efficiency of claims processing.

[0095] In some optional implementations of this embodiment, step 203, based on multiple global feature information, determines at least one historical claim case from the claim search library, and determines multiple target historical claim images corresponding to the at least one historical claim case, specifically including the following steps:

[0096] Calculating the first similarity between each global feature information and each feature cluster in the claims retrieval database to obtain multiple first similarities;

[0097] Based on the multiple first similarities, determining a target feature cluster from the multiple feature clusters, wherein the first similarity between the target feature cluster and the multiple global feature information reaches a similarity threshold;

[0098] Calculating the second similarity between each historical global feature information constituting the target feature cluster and each global feature information to obtain multiple second similarities;

[0099] Based on the multiple second similarities, at least one historical claim case is determined from the claim retrieval library, and multiple target historical claim images corresponding to the at least one historical claim case are determined.

[0100] In an embodiment of the present application, when determining the target feature cluster, the server calculates the cosine similarity between each global feature information and each feature cluster in the claims retrieval library. After calculation, a similarity list between each global feature information and the feature cluster is obtained. A similarity threshold can be set, for example 8, and only feature clusters with a similarity higher than the threshold will be regarded as target feature clusters. Next, the server calculates the cosine similarity between each global feature information and the historical global features in the target feature cluster. Since the number of target feature clusters is small, the number of historical global features is also greatly reduced, and the amount of calculation is significantly reduced. Select the K historical global features with the highest similarity, for example K = 10, and then find the claim case numbers corresponding to these historical global features as the most similar historical claims cases, and obtain at least one historical claims case. At the same time, the server can obtain multiple target historical claims images of at least one historical claims case from the claims retrieval library.

[0101] The embodiment of the present application determines a target feature cluster from multiple feature clusters by calculating a first similarity between each global feature information and each feature cluster in a claim retrieval library, and then calculates a second similarity between each historical global feature information constituting the target feature cluster and each global feature information, so as to determine at least one historical claim case from the claim retrieval library, and determine multiple target historical claim images corresponding to the at least one historical claim case, so as to quickly and accurately find the historical claim case and multiple target historical claim images that are most similar to the current claim case.

[0102] In some optional implementations of this embodiment, the above step of "determining at least one historical claim case from the claim retrieval library based on multiple second similarities, and determining multiple target historical claim images corresponding to the at least one historical claim case" specifically includes the following steps:

[0103] Sorting the plurality of second similarities to obtain a similarity sequence;

[0104] Based on the association relationship and the similarity sequence, at least one historical claim case is determined from the claim retrieval library, and a plurality of target historical claim images corresponding to the at least one historical claim case are determined.

[0105] In this embodiment, in the process of implementing claim case retrieval, the cosine similarity algorithm is first used to calculate the similarity between each global feature information and each feature cluster in the claim retrieval library. For example, if the global feature information is a vector with a dimension of 100, and the feature cluster in the claim retrieval library is also a vector with a dimension of 100, the first similarity between them is determined by calculating the cosine value of the two vectors. A similarity threshold is set, such as 8. Only when the calculated similarity is greater than or equal to this threshold, the feature cluster is considered to be similar to the input global feature information and is then selected as the target feature cluster. Next, a second similarity calculation is performed between each historical global feature information in the target feature cluster and the input global feature information, also using the cosine similarity algorithm. The calculation results in multiple second similarity values. These second similarity values ​​are sorted to form a similarity sequence. This step is critical because it determines which historical claims cases will be further considered. Finally, based on the sorted similarity sequence and the correlation between historical claims cases, such as case type, occurrence time, etc., at least one historical claims case that is most similar to the current case is selected from the claims retrieval database. At the same time, multiple target historical claims images corresponding to these historical claims cases are determined.

[0106] In the embodiment of the present application, multiple second similarities are sorted to obtain a similarity sequence, and based on the association relationship and the similarity sequence, at least one historical claim case is determined from the claim retrieval library, and multiple target historical claim images corresponding to at least one historical claim case are determined. These processes are completely completed automatically by the algorithm, ensuring the efficiency and accuracy of the processing.

[0107] In some optional implementations of this embodiment, step S204, performing aggregation processing on multiple claim images and multiple target historical claim images to obtain a target claim case set, specifically includes the following steps:

[0108] Based on the plurality of claim images and the plurality of target historical claim images, establishing a claim case graph between the claim case and at least one historical claim case;

[0109] Based on a preset image correction threshold, the claim case map is corrected to obtain a claim case correction map;

[0110] Using a preset search algorithm, image aggregation is performed on the claim case correction graph to obtain claim cases and at least one claim case set;

[0111] Feature aggregation is performed on the claim cases and at least one claim case set to obtain a target claim case set.

[0112] In an embodiment of the present application, in the process of aggregating multiple claim images and multiple target historical claim images, first, based on these images, a claim case graph is established with claim cases and each historical claim case in at least one claim case set as claim case nodes. The weight of the edge in the claim case graph represents the number of images of historical claim images of historical claim cases as the end point that can be retrieved from the claim case as the starting point. Next, in order to eliminate the influence of noise and outliers, the claim case graph is corrected. In this step, an image correction threshold is set, such as 6. If the weight value of the edge between any two claim case nodes is less than this threshold, then this edge will be filtered out from the claim case graph. After such correction, a more accurate claim case correction graph is obtained. Then, a preset search algorithm, such as a DFS algorithm, is used to perform image aggregation on the claim case correction graph to determine the claim case and at least one claim case set. Finally, feature aggregation is performed on the claim case and at least one claim case set to obtain the target claim case set. For example, Figure 3 As shown in the figure, assume that A, B, C, and D represent four claims cases, and they construct a weighted directed graph, namely the claims case graph. The directed edges in the directed graph represent that the starting case can retrieve the end case, and the weight of the edge represents the number of images in the starting case that can retrieve the end case, for example: A->D: all images in case A can retrieve a total of 6 images in case D. After the weighted directed graph is constructed, a threshold δ is set according to the use scenario to correct the graph. The function of this threshold is to filter out edges with edge weights less than δ, thereby reducing the case mismatch caused by multiple matching of highly similar graphs such as bank cards and ID cards. After the graph is corrected, the depth-first search algorithm (DFS) is used to traverse all case nodes that can be reached from any starting point in the direction of the directed arrow, and these cases are aggregated into a suspicious case set. Repeat the above traversal process until all case nodes in the directed graph have been traversed to obtain the results of multiple suspicious case sets. Figure 3In the example, claim case A (claim case) is a single case, and historical claim cases B, C, and D constitute a claim case set. After that, each claim image in the claim case is subjected to PCA-SIFT to obtain 128-dimensional local features, and is spliced ​​with the previously obtained 512-dimensional global features of the deep model to form 640 dimensions. Then, the 640-dimensional global and local features of the above images are averaged in units of cases to obtain the 640-dimensional mean cluster center features of the claim case. For each historical claim case in the claim case set, the above method is also used to calculate the historical mean cluster features of each historical claim case in the claim case set. Finally, the feature similarity between each historical mean cluster feature and the mean cluster feature is calculated respectively. If the similarity between the two claim cases is greater than the threshold σ (selected according to the actual scenario), the case sets to which the two claim cases belong are merged into a new case set to obtain the target claim case set, that is, the final aggregation result.

[0113] The embodiment of the present application compares the claim image with the historical claim image to quickly establish a correlation diagram between the claim case and the historical case (claim case diagram). This process can be completed automatically, greatly reducing the time for manual review and improving the efficiency of case processing. By correcting the claim case diagram through a preset image correction threshold, some noise or abnormal data can be filtered out, making subsequent classification more accurate. The steps of image aggregation and feature aggregation further refine the common features between cases, which helps to classify similar or related cases into the same category (claim case set), thereby improving the accuracy of classification.

[0114] In some optional implementations of this embodiment, the claim case correction graph includes multiple claim case nodes and multiple claim case edges. The above step of "using a preset search algorithm to perform image aggregation on the claim case correction graph to obtain claim cases and at least one claim case set" specifically includes the following steps:

[0115] Using a preset search algorithm, taking any claim case node as the starting point, traverse multiple claim case nodes and multiple claim case edges to generate multiple case paths;

[0116] Based on the multiple case paths, a claim case and at least one claim case set are determined.

[0117] In an embodiment of the present application, when processing a claim case correction graph, first identify each claim case node in the graph and the claim case edges connecting them. Using a preset search algorithm, such as depth-first search (DFS), take any claim case node as the starting point, traverse all nodes and edges, and generate a case path. During the traversal process, the sequence of all passed nodes and edges can be recorded to form multiple case paths. Next, based on the generated case path, a density-based spatial clustering of applications with noise (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) is used to determine the claim cases and at least one claim case set.

[0118] The embodiment of the present application can quickly and accurately locate claim cases with similar characteristics through a preset search algorithm, and obtain claim cases and at least one claim case set.

[0119] In some optional implementations of this embodiment, the above step of “performing feature aggregation on the claim cases and at least one claim case set to obtain a target claim case set” specifically includes the following steps:

[0120] Based on multiple claim images and global feature information, the mean clustering features of claim cases are calculated;

[0121] Calculate historical mean clustering features of each historical claim case in at least one case set based on a plurality of historical claim images and historical local feature information of historical claim cases in at least one case set;

[0122] The feature similarity between each historical mean clustering feature and the mean clustering feature is calculated respectively to obtain multiple feature similarities;

[0123] The case set of historical claims cases whose feature similarity is greater than a similarity threshold is determined as a similar case set of the claims case;

[0124] Merge the claim cases into similar case sets to obtain the target claim case set.

[0125] In an embodiment of the present application, when performing feature aggregation on claim cases and case sets, first, based on multiple claim images and global feature information of the claim cases, a scale-invariant feature transform algorithm based on principal component analysis (Principal Component Analysis), i.e., a PCA-SIFT algorithm, can be used to extract local feature information of multiple claim images, and then the local feature information and the global feature information are spliced ​​to obtain the target global feature information. The mean of the target global feature information and the local feature information is calculated to obtain the mean clustering feature vector of the claim case. Similarly, for each case in the historical claim case set, the same method is used to calculate its historical mean clustering feature. Next, the feature similarity between each historical mean clustering feature and the mean clustering feature of the current claim case is calculated using cosine similarity. The value range of cosine similarity is 0 to 1, and the larger the value, the more similar the two feature vectors are. A similarity threshold is set, for example 8, and the case set to which the historical claim cases with feature similarity greater than the threshold belong is determined as the similar case set of the current claim case. Finally, the current claim case is merged into a set of similar cases to obtain the target claim case set.

[0126] The embodiment of the present application calculates the mean clustering features of the claim cases and the historical mean clustering features of the historical claim cases, and determines the similar case set based on the similarity between these features, so as to more accurately identify the historical case set similar to the current claim case. This method is more accurate and reliable than the traditional classification method based on a single feature or manual judgment.

[0127] In some optional implementations of this embodiment, the above step of "calculating the mean clustering features of the claim cases based on multiple claim images and global feature information" specifically includes the following steps:

[0128] Extracting local features of multiple claim images to obtain local feature information of the multiple claim images;

[0129] The local feature information and the global feature information are spliced ​​to obtain the target global feature information;

[0130] The target global feature information and local feature information are weighted to obtain the mean clustering features of the claim cases.

[0131] In an embodiment of the present application, when processing a claim case, local features are first extracted from each claim image. For example, the PCA-SIFT algorithm is used to extract local feature vectors of 100 key points for each image, and the dimension of each local vector can be 128. Then, these local feature vectors are spliced ​​with the global feature information of the claim case to form a more comprehensive feature representation. Next, these spliced ​​feature vectors are weighted to calculate the mean clustering features of the claim case. Specifically, a weighted average method can be used. Through this weighted method, it can be ensured that both local image features and global case information are properly reflected in the final clustering features. Finally, this weighted mean clustering feature vector is used to represent the entire claim case. In subsequent case processing and analysis, this will provide an accurate and comprehensive case feature description.

[0132] In the embodiments of the present application, local features can capture detailed information in the claims image, such as edges, corners, etc., which are key to distinguishing different claims cases. Global features can reflect the overall attributes and contextual information of the image, which helps to understand the image content from a more macro perspective. By splicing local features with global features, the advantages of both can be combined to form a more comprehensive and rich feature representation, thereby improving the ability to characterize claims cases. The mean clustering feature is obtained by performing weighted operations on local features and global features. This process can balance the impact of different features on classification and recognition results.

[0133] In some optional implementations of this embodiment, the above step of "calculating the historical mean clustering features of each historical claim case in at least one case set based on multiple historical claim images and historical local feature information of historical claim cases in at least one case set" specifically includes the following steps:

[0134] Extracting historical local features of multiple historical claims images of each historical claims case in at least one case set respectively, to obtain historical local feature information of each historical claims case;

[0135] splicing the historical local feature information and multiple historical global feature information of historical claims cases in at least one case set to obtain target historical global feature information;

[0136] A weighted operation is performed on the target historical global feature information and the historical local feature information to obtain the historical mean clustering features of each historical claim case in at least one case set.

[0137] In an embodiment of the present application, when processing historical claims cases in at least one case set, local features are first extracted from multiple historical claims images of each historical claims case. Then, these historical local feature information are spliced ​​with the historical global feature information of the case. Thereafter, these spliced ​​feature vectors are weighted to calculate the historical mean clustering features of each historical claims case. Specifically, a weighted average method can be used, which ensures that both historical image features and historical case information are properly reflected in the final clustering features, thereby providing an accurate and comprehensive historical case feature description for subsequent case analysis.

[0138] The embodiment of the present application can fully capture the detailed information and overall attributes of the case image by extracting the local features and global features of the historical claims image respectively. By splicing the local features and the global features, a richer and more comprehensive feature representation can be formed, thereby more accurately describing the characteristics of each historical claims case. The historical mean clustering feature is obtained by weighting the target historical global feature information and the historical local feature information. This process can balance the influence of local features and global features in cluster analysis.

[0139] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned multiple claim images and the multiple historical claim images used to establish a claim retrieval library, the above-mentioned multiple claim images and the multiple historical claim images used to establish a claim retrieval library can also be stored in a node of a blockchain.

[0140] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0141] The embodiment of the present application can acquire and process multiple claim images and historical claim images based on the relevant algorithms of computer vision technology under artificial intelligence. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0142] Computer vision technology is an important branch of artificial intelligence, which is dedicated to enabling computers to have human-like visual understanding and analysis capabilities. By simulating certain functions of the human visual system, computer vision technology can process, analyze and interpret image and video data to extract useful information or perform specific tasks. Computer vision technology can be used to extract image features of claims images. This is achieved through a series of algorithms, such as edge detection, corner detection, texture analysis, etc., to identify key elements and attributes in the image, and obtain global feature information and local feature information of multiple claims images.

[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0144] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0145] Further references Figure 4 , as a response to the above Figure 2 In order to realize the method shown in the figure, the present application provides an embodiment of a device for determining similar claims cases, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0146] like Figure 4 As shown, the similar claim case determination device 400 of this embodiment includes: a request receiving module 401, a first feature extraction module 402, a case determination module 403, and a case set aggregation module 404. Among them:

[0147] A request receiving module 401 is used to receive a claim request of a claim case, wherein the claim request includes a plurality of claim images of the claim case;

[0148] A first feature extraction module 402 is used to perform global feature extraction processing on multiple claim settlement images to obtain multiple global feature information of the multiple claim settlement images;

[0149] A case determination module 403 is used to determine at least one historical claim case from the claim search library based on multiple global feature information, and to determine multiple target historical claim images corresponding to the at least one historical claim case;

[0150] The case set aggregation module 404 is used to aggregate multiple claim images and multiple target historical claim images to obtain a target claim case set, in which the similarity between each historical claim case and the claim case reaches a target similarity threshold.

[0151] After receiving the claim request of the claim case, the embodiment of the present application can automatically extract and analyze the global features of the claim image, and can quickly identify images that are highly similar to the historical claim cases, so as to determine at least one historical claim case from the claim retrieval library. Then, by aggregating multiple claim images and multiple target historical claim images, a set of similar cases whose similarity to the claim case reaches the target similarity threshold is determined to form a target claim case set, which can improve the efficiency of determining similar claim cases, reduce the burden of manual review, effectively shorten the review time, and improve the efficiency and accuracy of claim case review.

[0152] In this embodiment, the device for determining similar claim cases may further include:

[0153] A data acquisition module, used to acquire historical claims data, the historical claims data including a plurality of historical claims images, a claims case number to which each historical claims image belongs, and image category information of the plurality of historical claims images;

[0154] A second feature extraction module is used to perform global feature extraction processing on multiple historical claims images to obtain historical global feature information of multiple historical claims images;

[0155] An information clustering module is used to cluster historical global feature information to obtain multiple feature clusters, each of which is composed of multiple historical global feature information;

[0156] The relationship building module is used to establish the association relationship between multiple feature clusters, claim case numbers and image category information, and build a claim retrieval library based on the association relationship.

[0157] The embodiment of the present application obtains historical claims data and establishes a claims retrieval library. When a new claim request arrives, it only needs to extract the global features of its claim image, and then perform similarity matching in the claim retrieval library to quickly find similar historical claims cases and multiple target historical claims images corresponding to historical claims cases, greatly improving the efficiency of claims processing.

[0158] In this embodiment, the case determination module 403 includes:

[0159] A first calculation submodule is used to calculate a first similarity between each global feature information and each feature cluster in the claim retrieval library to obtain a plurality of first similarities;

[0160] A feature cluster determination submodule, used to determine a target feature cluster from multiple feature clusters based on multiple first similarities, wherein the first similarity between the target feature cluster and multiple global feature information reaches a similarity threshold;

[0161] A second calculation submodule is used to calculate a second similarity between each historical global feature information constituting the target feature cluster and each global feature information to obtain a plurality of second similarities;

[0162] The information determination submodule is used to determine at least one historical claim case from the claim retrieval library based on multiple second similarities, and to determine multiple target historical claim images corresponding to the at least one historical claim case.

[0163] The embodiment of the present application determines a target feature cluster from multiple feature clusters by calculating a first similarity between each global feature information and each feature cluster in a claim retrieval library, and then calculates a second similarity between each historical global feature information constituting the target feature cluster and each global feature information, so as to determine at least one historical claim case from the claim retrieval library, and determine multiple target historical claim images corresponding to the at least one historical claim case, so as to quickly and accurately find the historical claim case and multiple target historical claim images that are most similar to the current claim case.

[0164] In this embodiment, the information determination submodule is also used to sort multiple second similarities to obtain a similarity sequence; based on the association relationship and the similarity sequence, determine at least one historical claim case from the claim retrieval library, and determine multiple target historical claim images corresponding to at least one historical claim case.

[0165] In an embodiment of the present application, multiple second similarities are sorted to obtain a similarity sequence, and based on the association relationship and the similarity sequence, at least one historical claim case is determined from a claim retrieval library, and multiple target historical claim images corresponding to at least one historical claim case are determined. These processes are completely completed automatically by the algorithm, ensuring the efficiency and accuracy of the processing.

[0166] In this embodiment, the case collection aggregation module 404 includes:

[0167] Establishing a submodule for establishing a claim case graph between a claim case and at least one historical claim case based on a plurality of claim images and a plurality of target historical claim images;

[0168] An image correction submodule, used to correct the claim case map based on a preset image correction threshold value to obtain a claim case correction map;

[0169] An image aggregation submodule, for performing image aggregation on the claim case correction graph using a preset search algorithm to obtain claim cases and at least one claim case set;

[0170] The feature aggregation submodule is used to perform feature aggregation on the claim case and at least one claim case set to obtain a target claim case set.

[0171] The embodiment of the present application compares the claim image with the historical claim image to quickly establish a correlation diagram between the claim case and the historical case (claim case diagram). This process can be completed automatically, greatly reducing the time for manual review and improving the efficiency of case processing. By correcting the claim case diagram through a preset image correction threshold, some noise or abnormal data can be filtered out, making subsequent classification more accurate. The steps of image aggregation and feature aggregation further refine the common features between cases, which helps to classify similar or related cases into the same category (claim case set), thereby improving the accuracy of classification.

[0172] In this embodiment, the image aggregation submodule is also used to adopt a preset search algorithm, traverse multiple claim case nodes and multiple claim case edges with any claim case node as the starting point, and generate multiple case paths; based on the multiple case paths, determine the claim case and at least one claim case set.

[0173] The embodiment of the present application can quickly and accurately locate claim cases with similar characteristics through a preset search algorithm, and obtain claim cases and at least one claim case set.

[0174] In this embodiment, the feature aggregation submodule is also used to calculate the mean clustering features of claim cases based on multiple claim images and global feature information; calculate the historical mean clustering features of each historical claim case in at least one case set based on multiple historical claim images and historical local feature information of historical claim cases in at least one case set; calculate the feature similarity between each historical mean clustering feature and the mean clustering feature respectively to obtain multiple feature similarities; determine the case set to which the historical claim cases whose feature similarity is greater than the similarity threshold belong as the similar case set of the claim cases; merge the claim cases into the similar case set to obtain the target claim case set.

[0175] The embodiment of the present application calculates the mean clustering features of the claim cases and the historical mean clustering features of the historical claim cases, and determines the similar case set based on the similarity between these features, so as to more accurately identify the historical case set similar to the current claim case. This method is more accurate and reliable than the traditional classification method based on a single feature or manual judgment.

[0176] In this embodiment, the feature aggregation submodule is also used to extract local features of multiple claims images to obtain local feature information of multiple claims images; to splice the local feature information and the global feature information to obtain the target global feature information; to perform weighted operations on the target global feature information and the local feature information to obtain the mean clustering features of the claims case.

[0177] In the embodiments of the present application, local features can capture detailed information in the claims image, such as edges, corners, etc., which are key to distinguishing different claims cases. Global features can reflect the overall attributes and contextual information of the image, which helps to understand the image content from a more macro perspective. By splicing local features with global features, the advantages of both can be combined to form a more comprehensive and rich feature representation, thereby improving the ability to characterize claims cases. The mean clustering feature is obtained by performing weighted operations on local features and global features. This process can balance the impact of different features on classification and recognition results.

[0178] In this embodiment, the feature aggregation submodule is also used to extract the historical local features of multiple historical claims images of each historical claims case in at least one case set, respectively, to obtain the historical local feature information of each historical claims case; to splice the historical local feature information and multiple historical global feature information of the historical claims cases in at least one case set, to obtain the target historical global feature information; to perform weighted operations on the target historical global feature information and the historical local feature information, to obtain the historical mean clustering features of each historical claims case in at least one case set.

[0179] The embodiment of the present application can fully capture the detailed information and overall attributes of the case image by extracting the local features and global features of the historical claims image respectively. By splicing the local features and the global features, a richer and more comprehensive feature representation can be formed, thereby more accurately describing the characteristics of each historical claims case. The historical mean clustering feature is obtained by weighting the target historical global feature information and the historical local feature information. This process can balance the influence of local features and global features in cluster analysis.

[0180] To solve the above technical problems, the present application also provides a computer device. Figure 5 , Figure 5 This is a basic structural block diagram of the computer device in this embodiment.

[0181] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 6 having a memory 61, a processor 62, and a network interface 63, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASIC), programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.

[0182] Computer devices can be computing devices such as desktop computers, notebooks, PDAs, and cloud servers. Computer devices can interact with users through keyboards, mice, remote controls, touch pads, or voice control devices.

[0183] The memory 61 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as a hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk equipped on the computer device 6, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 61 can also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for the method of determining similar claims cases. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or are to be output.

[0184] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to run computer-readable instructions stored in the memory 61 or process data, such as computer-readable instructions for running a method for determining similar claim cases.

[0185] The network interface 63 may include a wireless network interface or a wired network interface, and the network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0186] After receiving the claim request of the claim case, the embodiment of the present application can automatically extract and analyze the global features of the claim image, and can quickly identify images that are highly similar to the historical claim cases, so as to determine at least one historical claim case from the claim retrieval library. Then, by aggregating multiple claim images and multiple target historical claim images, a set of similar cases whose similarity to the claim case reaches the target similarity threshold is determined to form a target claim case set, which can improve the efficiency of determining similar claim cases, reduce the burden of manual review, effectively shorten the review time, and improve the efficiency and accuracy of claim case review.

[0187] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions. The computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the method for determining similar claims cases as described above.

[0188] After receiving the claim request of the claim case, the embodiment of the present application can automatically extract and analyze the global features of the claim image, and can quickly identify images that are highly similar to the historical claim cases, so as to determine at least one historical claim case from the claim retrieval library. Then, by aggregating multiple claim images and multiple target historical claim images, a set of similar cases whose similarity to the claim case reaches the target similarity threshold is determined to form a target claim case set, which can improve the efficiency of determining similar claim cases, reduce the burden of manual review, effectively shorten the review time, and improve the efficiency and accuracy of claim case review.

[0189] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0190] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for determining similar claim cases, characterized in that: The steps include: receiving a claim request for a claim case, wherein the claim request includes a plurality of claim images for the claim case; Performing global feature extraction processing on the multiple claim settlement images to obtain multiple global feature information of the multiple claim settlement images; Based on the multiple global feature information, determining at least one historical claim case from a claim retrieval library, and determining multiple target historical claim images corresponding to the at least one historical claim case; Aggregation processing is performed on the multiple claim images and the multiple target historical claim images to obtain a target claim case set, wherein the similarity between each historical claim case in the target claim case set and the claim case reaches a target similarity threshold.

2. The method according to claim 1, characterized in that Before the step of determining at least one historical claim case from the claim retrieval library based on the multiple global feature information and determining multiple target historical claim images corresponding to the at least one historical claim case, the method further includes: Acquire historical claims data, the historical claims data including a plurality of historical claims images, a claims case number to which each historical claims image belongs, and image category information of the plurality of historical claims images; Performing global feature extraction processing on the multiple historical claims images to obtain historical global feature information of the multiple historical claims images; Clustering the historical global feature information to obtain a plurality of feature clusters, each feature cluster consisting of a plurality of historical global feature information; Establish associations among the multiple feature clusters, the claim case number, and the image category information, and construct a claim retrieval library based on the associations.

3. The method according to claim 2, characterized in that The step of determining at least one historical claim case from a claim search library based on the multiple global feature information, and determining multiple target historical claim images corresponding to the at least one historical claim case, specifically includes: Calculating a first similarity between each global feature information and each feature cluster in the claims retrieval library to obtain a plurality of first similarities; Based on the multiple first similarities, determining a target feature cluster from the multiple feature clusters, wherein the first similarity between the target feature cluster and the multiple global feature information reaches a similarity threshold; Calculating a second similarity between each piece of historical global feature information constituting the target feature cluster and each piece of global feature information to obtain a plurality of second similarities; Based on the multiple second similarities, at least one historical claim case is determined from the claim retrieval library, and multiple target historical claim images corresponding to the at least one historical claim case are determined.

4. The method according to claim 3, characterized in that: The step of determining at least one historical claim case from the claim search library based on the multiple second similarities, and determining multiple target historical claim images corresponding to the at least one historical claim case, specifically includes: Sorting the multiple second similarities to obtain a similarity sequence; Based on the association relationship and the similarity sequence, at least one historical claim case is determined from the claim retrieval library, and a plurality of target historical claim images corresponding to the at least one historical claim case are determined.

5. The method according to claim 1, characterized in that The step of aggregating the multiple claim images and the multiple target historical claim images to obtain a target claim case set specifically includes: establishing a claim case graph between the claim case and the at least one historical claim case based on the multiple claim images and the multiple target historical claim images; Based on a preset image correction threshold, the claim case map is corrected to obtain a claim case correction map; Using a preset search algorithm, perform image aggregation on the claim case correction graph to obtain the claim case and at least one claim case set; Feature aggregation is performed on the claim case and the at least one claim case set to obtain a target claim case set.

6. The method according to claim 5, characterized in that The claim case correction graph includes a plurality of claim case nodes and a plurality of claim case edges. The step of using a preset search algorithm to perform image aggregation on the claim case correction graph to obtain the claim case and at least one claim case set specifically includes: Using a preset search algorithm, taking any claim case node as a starting point, traversing the multiple claim case nodes and the multiple claim case edges, and generating multiple case paths; Based on the multiple case paths, the claim case and at least one claim case set are determined.

7. The method according to claim 5, characterized in that The step of performing feature aggregation on the claim case and the at least one claim case set to obtain a target claim case set specifically includes: Calculating a mean cluster feature of the claim case based on the multiple claim images and the global feature information; Calculating historical mean clustering features of each historical claim case in the at least one case set based on a plurality of historical claim images and historical local feature information of the historical claim cases in the at least one case set; Calculating the feature similarity between each historical mean cluster feature and the mean cluster feature respectively to obtain multiple feature similarities; Determine the case set to which the historical claim cases with feature similarity greater than a similarity threshold belong as a similar case set of the claim case; The claim cases are merged into the similar case set to obtain a target claim case set.

8. The method according to claim 7, characterized in that The step of calculating the mean clustering feature of the claim case based on the multiple claim images and the global feature information specifically includes: Extracting local features of the multiple claims images to obtain local feature information of the multiple claims images; splicing the local feature information and the global feature information to obtain target global feature information; A weighted operation is performed on the target global feature information and the local feature information to obtain a mean clustering feature of the claim case.

9. The method according to claim 7, characterized in that: The step of calculating the historical mean clustering features of each historical claim case in the at least one case set based on the multiple historical claim images and historical local feature information of the historical claim cases in the at least one case set specifically includes: Respectively extracting historical local features of a plurality of historical claims images of each historical claims case in the at least one case set to obtain historical local feature information of each historical claims case; splicing the historical local feature information and a plurality of historical global feature information of historical claims cases in the at least one case set to obtain target historical global feature information; A weighted operation is performed on the target historical global feature information and the historical local feature information to obtain historical mean clustering features of each historical claim case in at least one case set.

10. A device for determining similar claim cases, characterized in that: include: A request receiving module, configured to receive a claim request of a claim case, wherein the claim request includes a plurality of claim images of the claim case; A first feature extraction module is used to perform global feature extraction processing on the multiple claim settlement images to obtain multiple global feature information of the multiple claim settlement images; A case determination module, configured to determine at least one historical claim case from a claim retrieval library based on the plurality of global feature information, and to determine a plurality of target historical claim images corresponding to the at least one historical claim case; The case set aggregation module is used to aggregate the multiple claim images and the multiple target historical claim images to obtain a target claim case set, and the similarity between each historical claim case in the target claim case set and the claim case reaches a target similarity threshold.

11. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the method according to any one of claims 1 to 9 when executing the computer-readable instructions.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

Images