Claims analysis method, device, equipment and medium based on multi-factor dynamic weights

Abstract

The present invention relates to the field of intelligent decision-making, and discloses a claim analysis method, device, equipment and medium based on multi-factor dynamic weights, the method comprising: constructing a factor image of a claim factor, calculating a convolution resolution map of the factor image, and calculating an attention map of the convolution resolution map, and constructing a multi-branch map of the factor image based on the convolution resolution map and the attention map; performing branch map fusion processing on the multi-branch map to obtain a fused branch map, and calculating the claim grade probability corresponding to the fused branch map; calculating a dynamic allocation index of the claim grade probability, calculating a dynamic allocation weight of the claim grade probability using the dynamic allocation index, and querying the final claim grade of the claim factor based on the dynamic allocation weight and the claim grade probability; according to the final claim grade, identifying the claim item corresponding to the claim factor, and using the claim item as the claim analysis result corresponding to the claim factor. The present invention can use dynamic weights to improve the certainty of claim analysis.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making, and in particular to a claims analysis method, device, equipment and medium based on multi-factor dynamic weights. Background Art

[0002] Claims analysis based on multi-factor dynamic weights refers to the process of assigning dynamic weights to multiple factors used in claims assessment to indicate the importance of the multiple factors, thereby conducting claims assessment.

[0003] At present, claims analysis faces a lot of data analysis and data processing, and neural network models are usually used to process a large amount of data. However, due to the large number of material subdivisions in claims analysis, the mutual influence of various subdivisions leads to the multiple probabilities of the predicted categories output by the neural network model being close, and the two extreme probability values ​​that meet the expected binary classification values ​​do not appear, which ultimately makes it impossible to determine that the category corresponding to the probability value is the final correct category. Therefore, the certainty of claims analysis is insufficient. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a claims analysis method, device, equipment and medium based on multi-factor dynamic weights, which can use dynamic weights to convert claims assessment categories with unclear numerical boundaries into definite claims assessment categories, thereby improving the certainty of claims analysis.

[0005] In a first aspect, the present invention provides a claims analysis method based on multi-factor dynamic weights, comprising:

[0006] Acquire a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, calculate an attention map of the convolution resolution map, and construct a multi-branch map of the factor image based on the convolution resolution map and the attention map;

[0007] Performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph;

[0008] Calculating the dynamic allocation index of the claim level probability, calculating the dynamic allocation weight of the claim level probability using the dynamic allocation index, and querying the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability;

[0009] According to the final claim level, the claim items corresponding to the claim factors are identified, and the claim items are used as the claim analysis results corresponding to the claim factors.

[0010] In a possible implementation manner of the first aspect, calculating the attention map of the convolutional resolution map includes:

[0011] Performing channel processing on the convolution resolution map to obtain a channel processing map;

[0012] Calculating the attention weight of the channel processing graph;

[0013] Performing nonlinear transformation on the convolution resolution map to obtain a nonlinear transformation map;

[0014] determining an activation value of the nonlinear transformation map;

[0015] The attention weight and the activation value are concatenated into a feature map to obtain the attention map.

[0016] In a possible implementation manner of the first aspect, calculating the attention weight of the channel processing graph includes:

[0017] The following formula is used to calculate the proportion of non-zero feature points in the channel processing image:

[0018]

[0019] Among them, D i Indicates the proportion of non-zero feature points, N i represents the number of non-zero feature points in the feature map with sequence number i in the channel processing map, i represents the sequence number of the feature map in the channel processing map, W*H represents the size of the feature map in the channel processing map, W represents the width of the feature map in the channel processing map, and H represents the height of the feature map in the channel processing map;

[0020] According to the proportion of non-zero feature points, the weight of the non-zero feature points in the channel processing graph is calculated using the following formula:

[0021]

[0022] Among them, I weight represents the weight of the non-zero feature point in the channel processing graph, D i represents the proportion of the non-zero feature points, i represents the sequence number of the feature map in the channel processing map, and k represents the total number of feature maps in the channel processing map;

[0023] Based on the non-zero feature point weights, the attention weight of the channel processing graph is calculated using the following formula:

[0024] I weight ′=λD i *δ(I weight ),(λ∈[0,1])

[0025] Among them, I weight ′ represents the attention weight of the channel processing graph, Di represents the proportion of the non-zero feature points, i represents the sequence number of the feature map in the channel processing map, and I weight represents the weight of the non-zero feature point in the channel processing graph, λ represents an adjustment parameter, which is used to indicate the impact of the adjustment of λ on the attention weight, and the range is [0,1].

[0026] In a possible implementation manner of the first aspect, performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph includes:

[0027] Identifying an upper branch graph and a lower branch graph in the multi-branch graph;

[0028] Performing branch graph fusion processing on the upper branch graph in the lower branch graph by a down sampling method to obtain a first fused branch graph;

[0029] Performing branch graph fusion processing on the lower branch graph in the upper branch graph by an upsampling method to obtain a second fused branch graph;

[0030] The first fused branch graph and the second fused branch graph are used as the fused branch graph.

[0031] In a possible implementation of the first aspect, the calculating the dynamic allocation index of the claim level probability includes:

[0032] The following formula is used to identify the probability of uncertainty in the claim level probability:

[0033] p un >>p (-u)n

[0034] Among them, p (-u)n represents the probability of uncertainty in the claim level probability, p un represents the probability of the activation output probability with the mark symbol u in the claim level probability, u represents the mark symbol of the activation output probability in the claim level probability, n represents the probability number in the activation output probability of each claim level probability, -u represents the mark symbol of the activation output probability other than u in the claim level probability, and is also p (-u)n The number of the uncertainty probability, p un >>p (-u)n Indicates that when p (-u)n Much smaller than p un When p (-u)n is the probability of uncertainty in the claim level probability;

[0035] The dynamic allocation index of the uncertainty probability is calculated using the following formula:

[0036]

[0037] Wherein, G(p) represents the dynamic allocation index, p n represents the uncertainty probability, N represents the total number of the uncertainty probabilities, and n represents the sequence number of the uncertainty probability.

[0038] In a possible implementation of the first aspect, querying the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability includes:

[0039] Based on the dynamic allocation weight and the claim level probability, the final claim probability of the claim factor is calculated using the following formula:

[0040]

[0041] Among them, O softmax represents the final claim probability of the claim factor, W U represents the weight of the claim level probability, including the dynamic allocation weight for the uncertainty probability and the unallocated weight for the deterministic probability, which can be regarded as 1. U represents the marking symbol of the activated output probability in the claim level probability, including u and -u, that is, the probability sequence number in each activated output probability in the claim level probability, and the marking symbol of the activated output probability in the claim level probability except u, n U represents the sequence number of the probability in the activation output probability in the claim level probability, represents the claim level probability, argmax represents the argmax function;

[0042] Querying a target claim probability with the largest probability value from the final claim probability;

[0043] The claim level corresponding to the target claim probability is obtained, and the target claim level corresponding to the target claim probability is taken as the final claim level from the claim levels.

[0044] In a possible implementation of the first aspect, identifying the claim item corresponding to the claim factor according to the final claim level includes:

[0045] Determine the initial claim items corresponding to the final claim level;

[0046] Identify the factor category of the claim factor;

[0047] The target claim item corresponding to the factor category is searched in the initial claim item, and the target claim item is used as the claim item.

[0048] In a second aspect, the present invention provides a claims analysis device based on multi-factor dynamic weights, the device comprising:

[0049] A multi-branch graph construction module, used to obtain a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, and calculate an attention map of the convolution resolution map, and construct a multi-branch graph of the factor image based on the convolution resolution map and the attention map;

[0050] A claim probability calculation module, used for performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph;

[0051] A claim level query module, used to calculate the dynamic allocation index of the claim level probability, calculate the dynamic allocation weight of the claim level probability using the dynamic allocation index, and query the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability;

[0052] The claim item identification module is used to identify the claim items corresponding to the claim factors according to the final claim level, and use the claim items as the claim analysis results corresponding to the claim factors.

[0053] In a third aspect, the present invention provides an electronic device, comprising:

[0054] at least one processor; and a memory communicatively coupled to the at least one processor;

[0055] Wherein, the memory stores a computer program executable by the at least one processor, so that the at least one processor can execute the claims analysis method based on multi-factor dynamic weights as described in any one of the first aspects above.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a claims analysis method based on multi-factor dynamic weights as described in any one of the first aspects above.

[0057] Compared with the prior art, the technical principle and beneficial effects of this solution are:

[0058] The embodiment of the present invention first obtains claim factors to evaluate whether the claim materials submitted by the user meet the claim conditions. Furthermore, the embodiment of the present invention constructs a factor image of the claim factors to convert the text materials submitted by the user into a text feature map, so as to facilitate the subsequent multi-layer convolution layer and attention layer model to classify the feature map. Furthermore, the embodiment of the present invention calculates the convolution resolution map of the factor image to extract features from the factor image. Furthermore, the embodiment of the present invention calculates the attention map of the convolution resolution map to perform simple linear weighting on all input vectors to further mine the detail features of the convolution resolution map. Furthermore, the embodiment of the present invention constructs a multi-branch map of the factor image based on the convolution resolution map and the attention map to realize the detail feature extraction of the factor image using branches of different resolution sizes. Secondly, the embodiment of the present invention performs branch map fusion processing on the multi-branch map to merge the features extracted from the image into features with better discriminative ability than the input features. Furthermore, the embodiment of the present invention calculates the fused branch. The claim level probability corresponding to the figure is used to output the predicted probability of the claim level of the factor image using the neural network model to realize the classification of the claim level of the factor image, and subsequently dynamically weight the probabilities with similar values, unclear distinction, and inability to judge the final level in the output predicted probabilities, so as to improve the certainty of the output predicted probability; further, the embodiment of the present invention calculates the dynamic allocation index of the claim level probability to mark the uncertain probability in the claim level probability with the dynamic allocation index; further, the embodiment of the present invention calculates the dynamic allocation weight of the claim level probability by using the dynamic allocation index to allocate an adaptive weight to the uncertain probability in the claim level probability, and converts the claim assessment category with unclear numerical boundaries into a certain claim assessment category by using the dynamic weight, so as to improve the certainty of the claim analysis; further, the embodiment of the present invention queries the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability, so as to determine the final claim level after the dynamic weight is allocated to the in-place probability, and is used to grade the claim materials submitted by the user, so as to facilitate the subsequent claim analysis. Therefore, the claims analysis method, device, equipment and medium based on multi-factor dynamic weights proposed in the embodiments of the present invention can use dynamic weights to convert claims assessment categories with unclear numerical boundaries into definite claims assessment categories, thereby improving the certainty of claims analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0061] Figure 1 A flowchart of a claims analysis method based on multi-factor dynamic weights provided by an embodiment of the present invention;

[0062] Figure 2 In one embodiment of the present invention Figure 1 A flowchart of one of the steps of a claims analysis method based on multi-factor dynamic weights is provided;

[0063] Figure 3 In one embodiment of the present invention Figure 1 A flowchart of another step of a claims analysis method based on multi-factor dynamic weights is provided;

[0064] Figure 4 A schematic diagram of a module of a claims analysis device based on multi-factor dynamic weights provided by an embodiment of the present invention;

[0065] Figure 5 A schematic diagram of the internal structure of an electronic device for implementing a claims analysis method based on multi-factor dynamic weights provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0066] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0067] The embodiment of the present invention provides a claim analysis method based on multi-factor dynamic weights, and the execution subject of the claim analysis method based on multi-factor dynamic weights includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present invention. In other words, the claim analysis method based on multi-factor dynamic weights can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0068] See also Figure 1 FIG. 1 is a flow chart of a method for claim analysis based on multi-factor dynamic weights provided by an embodiment of the present invention. Figure 1 The claims analysis method based on multi-factor dynamic weights described in the includes:

[0069] S1. Obtain a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, and calculate an attention map of the convolution resolution map, and construct a multi-branch map of the factor image based on the convolution resolution map and the attention map.

[0070] The embodiment of the present invention obtains claim factors to evaluate whether the claim materials submitted by the user meet the claim conditions. The claim factors refer to the claim materials submitted by the user for evaluating whether to pay the claim. For example, in the auto insurance claim scenario, the claim materials submitted by the user include a traffic accident identification certificate, a certificate of personal injury and disability, and a certificate of personal and property damage.

[0071] Furthermore, the embodiment of the present invention constructs a factor image of the claim factor to convert the text material submitted by the user into a text feature map, thereby facilitating the subsequent multi-layer convolutional layer and attention layer model to classify the feature map.

[0072] In one embodiment of the present invention, referring to Figure 2 As shown, the factor image of constructing the claim factor includes:

[0073] S201, dividing the state word blocks of the claim factors;

[0074] S202, identifying the state set and state category of the state word block;

[0075] S203: Based on the state set and the state category, construct a factor image of the claim factor by using one-hot encoding.

[0076] Among them, the one-hot encoding refers to (One-Hot Encoding), also known as one-bit effective encoding, which uses an N-bit state register to encode N states. Each state has its own independent register bit, and at any time, only one of them is valid, that is, only one bit is 1, and the rest are zero values. The one-hot encoding uses 0 and 1 to represent some parameters and uses an N-bit state register to encode N states; the state word block refers to the word block after removing some useless words. The length of the word block depends on the situation. For example, "disability degree" and "disability" are two blocks of different lengths, and they cannot be It is a useless word block such as "residual process"; the state set is composed of the state word blocks, which is usually a set of word blocks in a continuous text; the state category refers to a word block set composed of all state sets; for example, "A, B, C, D, E" is a state set, and the state set of all texts is "a, A, a, B, C, D, E, a", that is, the state category, then through one-hot encoding, it can be obtained that a vector in the factor image of the claim factor is "01011110", and multiple vectors are merged into a matrix to obtain the factor image.

[0077] Furthermore, the embodiment of the present invention calculates a convolution resolution map of the factor image to extract features from the factor image.

[0078] In one embodiment of the present invention, the calculating of the convolution resolution map of the factor image includes: performing regional sliding on the factor image using a resolution convolution kernel; when performing regional sliding on the factor image using the resolution convolution kernel, calculating the convolution value of the factor image using the following formula:

[0079]

[0080] in, represents the convolution value of the factor image, represents the mth neuron in layer l-1. It should be noted that each neuron is connected to a feature map. Therefore, the mth neuron in layer l-1 is connected to the mth feature map in layer l-1. Represents the weight between the mth feature map of the previous layer input and the nth feature map of the current layer, Represents the bias term between the lth layer and the nth feature map, m∈Map n Represents the feature map between the previous layer l-1 layer and the nth feature map of the target layer l layer;

[0081] Based on the convolution value, after performing regional sliding on the factor image using the resolution convolution kernel, a convolution resolution map of the factor image is constructed.

[0082] The resolution convolution kernel refers to convolution kernels of different resolution sizes. Exemplarily, the process of calculating the convolution resolution map of the factor image is to slide the convolution kernel from left to right and from top to bottom with a weighted convolution kernel, perform a product operation on the entire image area, and then add the values, and fill the calculated values ​​in the corresponding positions in the feature map in turn to extract image features.

[0083] Furthermore, the embodiment of the present invention calculates an attention map of the convolutional resolution map to perform simple linear weighting on all input vectors, thereby further mining the detail features of the convolutional resolution map.

[0084] In one embodiment of the present invention, the calculation of the attention map of the convolution resolution map includes: performing channel processing on the convolution resolution map to obtain a channel processing map; calculating the attention weight of the channel processing map; performing nonlinear transformation on the convolution resolution map to obtain a nonlinear transformation map; determining the activation value of the nonlinear transformation map; and performing feature map splicing of the attention weight and the activation value to obtain the attention map.

[0085] Optionally, the channel processing of the convolution resolution map to obtain the channel processing map can be: first, the convolution resolution map is normalized by a batch normalization layer of a neural network model, and then multi-channel output is performed using channel processing to obtain a final channel processing map. The batch normalization layer refers to BatchNormalization, BN, which is used to keep the input of each layer of the neural network in the same distribution during the deep neural network training process. The batch of Batch normalization is batch data, which divides the data into small batches for stochastic gradient descent, and when each batch of data is forward passed, each layer is normalized.

[0086] Furthermore, in another embodiment of the present invention, the calculating the attention weight of the channel processing graph includes: calculating the proportion of non-zero feature points in the channel processing graph using the following formula:

[0087]

[0088] Among them, D i Indicates the proportion of non-zero feature points, N i represents the number of non-zero feature points in the feature map with sequence number i in the channel processing map, i represents the sequence number of the feature map in the channel processing map, W*H represents the size of the feature map in the channel processing map, W represents the width of the feature map in the channel processing map, and H represents the height of the feature map in the channel processing map;

[0089] According to the proportion of non-zero feature points, the weight of the non-zero feature points in the channel processing graph is calculated using the following formula:

[0090]

[0091] Among them, I weight represents the weight of the non-zero feature point in the channel processing graph, D i represents the proportion of the non-zero feature points, i represents the sequence number of the feature map in the channel processing map, and k represents the total number of feature maps in the channel processing map;

[0092] Based on the non-zero feature point weights, the attention weight of the channel processing graph is calculated using the following formula:

[0093] I weight ′=λD i *δ(I weight ),(λ∈[0,1])

[0094] Among them, I weight ′ represents the attention weight of the channel processing graph, D i represents the proportion of non-zero feature points, i represents the sequence number of the feature map in the channel processing map, and I weight represents the weight of the non-zero feature point in the channel processing graph, λ represents an adjustment parameter, which is used to indicate the impact of the adjustment of λ on the attention weight, and the range is [0,1].

[0095] Optionally, the nonlinear transformation of the convolution resolution map to obtain a nonlinear transformation map can be implemented by a convolution layer, a batch normalization layer and a RELU activation function, and is mainly used for feature extraction; the determination of the activation value of the nonlinear transformation map can be implemented by an adaptive average pooling layer (Adaptive Average Pooling, AAP), a picture resize layer (Resize), a fully connected layer, a batch normalization layer and an activation function; the attention weight and the activation value are spliced ​​into a feature map, and the attention map can be obtained by the splicing formula r 3 =concat[r 1 ,r 2 ]accomplish.

[0096] Furthermore, the embodiment of the present invention constructs a multi-branch graph of the factor image based on the convolution resolution map and the attention map, so as to realize detail feature extraction of the factor image by utilizing branches of different resolution sizes.

[0097] In one embodiment of the present invention, referring to Figure 3 As shown, the multi-branch graph of the factor image is constructed based on the convolution resolution map and the attention map, including:

[0098] S301, performing downsampling processing on the convolution resolution map to obtain a downsampling branch map of the factor image;

[0099] S302, performing multi-layer convolution processing and multi-layer attention extraction processing on the attention map, and performing multi-layer convolution processing and multi-layer attention extraction processing on the down-sampled branch map to obtain a multi-branch map of the factor image.

[0100] Exemplarily, after the original input image is converted into the convolution resolution map through three convolution layers, the attention map of the convolution resolution map is determined, and the convolution resolution map is downsampled to a smaller resolution to obtain a first downsampled branch map, and then the attention map is subjected to multi-layer convolution and attention extraction to obtain the output map of the final convolution layer. At the same time, the first downsampled branch map is also subjected to multi-layer convolution and attention extraction to obtain the output map of the final convolution layer, and the output map of the final convolution layer corresponding to the first downsampled branch map is downsampled to obtain a second downsampled branch map, and the first downsampled branch map, the second downsampled branch map and the output map of the final convolution layer corresponding to the attention map (three branches) are used as the final multi-branch map.

[0101] S2. Perform branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculate the claim level probability corresponding to the fused branch graph.

[0102] The embodiment of the present invention performs branch graph fusion processing on the multiple branch graphs to combine the features extracted from the image into features with higher discriminative power than the input features.

[0103] In one embodiment of the present invention, the branch graph fusion processing is performed on the multiple branch graphs to obtain a fused branch graph, including: identifying an upper branch graph and a lower branch graph in the multiple branch graphs; performing branch graph fusion processing on the upper branch graph in the lower branch graph by a downsampling method to obtain a first fused branch graph; performing branch graph fusion processing on the lower branch graph in the upper branch graph by an upsampling method to obtain a second fused branch graph; and using the first fused branch graph and the second fused branch graph as the fused branch graph.

[0104] Furthermore, the embodiment of the present invention calculates the claim level probability corresponding to the fusion branch diagram, so as to utilize the neural network model to output the predicted probability of the claim level of the factor image, thereby realizing the classification of the claim level of the factor image, and subsequently dynamically assigning weights to the probabilities in the output predicted probabilities that are close in value, unclear in distinction, and unable to determine the final level, thereby improving the certainty of the output predicted probability.

[0105] Among them, the claim level probability refers to the probability of the claim level to which the claim materials submitted by the user belong. For example, the claim company stipulates in advance that the claim levels include heavy, medium, medium-high and light. Each level corresponds to the items that the claim company should claim for the user, including the death and disability compensation limit and the no-fault death and disability compensation limit. The compensation for funeral expenses, death compensation, transportation expenses incurred by the victim's relatives for funeral matters, disability compensation, disability assistive device expenses, nursing expenses, rehabilitation expenses, transportation expenses, dependents' living expenses, accommodation expenses, and loss of work wages. The insured shall, in accordance with the court For the compensation for mental damage borne by judgment or mediation, etc., for the claim level specified by the insurance company, the neural network outputs the predicted probability of each level based on the input claim materials, including the probability corresponding to the severe level, the probability corresponding to the intermediate level, the probability corresponding to the intermediate and advanced level, and the probability corresponding to the mild level. If the probability corresponding to the severe level is 0.01, the probability corresponding to the intermediate level is 0.015, the probability corresponding to the intermediate and advanced level is 0.02, and the probability corresponding to the mild level is 0.011, it can be seen that the values ​​of these four probabilities are relatively close, and it cannot be determined with certainty that the probability of 0.02 corresponding to the intermediate and advanced level is the final result.

[0106] In one embodiment of the present invention, the calculation of the claim level probability corresponding to the fused branch graph is implemented by an activation function.

[0107] S3. Calculate the dynamic allocation index of the claim level probability, use the dynamic allocation index to calculate the dynamic allocation weight of the claim level probability, and query the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability.

[0108] The embodiment of the present invention calculates the dynamic allocation index of the claim level probability to mark the uncertain probability in the claim level probability with the dynamic allocation index.

[0109] In one embodiment of the present invention, the calculation of the dynamic allocation index of the claim level probability includes: identifying the uncertainty probability in the claim level probability using the following formula:

[0110] p un >>p (-u)n

[0111] Among them, p (-u)n represents the probability of uncertainty in the claim level probability, p un represents the probability of the activation output probability with the mark symbol u in the claim level probability, u represents the mark symbol of the activation output probability in the claim level probability, n represents the probability number in the activation output probability of each claim level probability, -u represents the mark symbol of the activation output probability other than u in the claim level probability, and is also p (-u)n The number of the uncertainty probability, pun >>p (-u)n Indicates that when p (-u)n Much smaller than p un When p (-u)n is the probability of uncertainty in the probability of the claim level;

[0112] The dynamic allocation index of the uncertainty probability is calculated using the following formula:

[0113]

[0114] Wherein, G(p) represents the dynamic allocation index, p n represents the uncertainty probability, N represents the total number of the uncertainty probabilities, and n represents the sequence number of the uncertainty probability.

[0115] Furthermore, the embodiment of the present invention calculates the dynamic allocation weight of the claim level probability by utilizing the dynamic allocation index, so as to allocate adaptive weights to the uncertain probabilities in the claim level probability, and utilizes the dynamic weights to convert the claim assessment categories with unclear numerical boundaries into definite claim assessment categories, thereby improving the certainty of the claim analysis.

[0116] In one embodiment of the present invention, the calculation of the dynamic allocation weight of the claim level probability using the dynamic allocation index includes: calculating the dynamic allocation weight of the claim level probability using the following formula according to the dynamic allocation index:

[0117]

[0118] Among them, w represents the dynamic allocation weight of the claim level probability, G and G(p) represent the dynamic allocation index, - represents the negative sign, and p represents the probability.

[0119] Furthermore, the embodiment of the present invention queries the final claim level of the claim factors based on the dynamically allocated weights and the claim level probability, so as to determine the final claim level after allocating dynamic weights by probability, and to grade the claim materials submitted by the user, so as to facilitate subsequent claim analysis.

[0120] In one embodiment of the present invention, querying the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability includes: calculating the final claim probability of the claim factor based on the dynamic allocation weight and the claim level probability using the following formula:

[0121]

[0122] Among them, O softmax represents the final claim probability of the claim factor, W Urepresents the weight of the claim level probability, including the dynamic allocation weight for the uncertainty probability and the unallocated weight for the deterministic probability, which can be regarded as 1. U represents the marking symbol of the activated output probability in the claim level probability, including u and -u, that is, the probability sequence number in each activated output probability in the claim level probability, and the marking symbol of the activated output probability in the claim level probability except u, n U represents the sequence number of the probability in the activation output probability in the claim level probability, represents the claim level probability, argmax represents the argmax function;

[0123] The target claim probability with the largest probability value is queried from the final claim probabilities; the claim level corresponding to the target claim probability is obtained, and the target claim level corresponding to the target claim probability is used as the final claim level from the claim levels.

[0124] S4. According to the final claim level, identify the claim items corresponding to the claim factors, and use the claim items as the claim analysis results corresponding to the claim factors.

[0125] In one embodiment of the present invention, identifying the claim item corresponding to the claim factor based on the final claim level includes: determining the initial claim item corresponding to the final claim level; identifying the factor category of the claim factor; querying the target claim item corresponding to the factor category in the initial claim item, and using the target claim item as the claim item.

[0126] Among them, the initial claim item refers to the claim item corresponding to each claim level, and the target claim item refers to the claim item that meets the user's demands and corresponds to the factor category. For example, if there is no requirement for accommodation fees in the user's demands, that is, in the claim materials, it will be deleted from the initial claim item to obtain the target claim item.

[0127] It can be seen that the embodiment of the present invention first obtains claim factors to evaluate whether the claim materials submitted by the user meet the claim conditions. Furthermore, the embodiment of the present invention constructs a factor image of the claim factors to convert the text materials submitted by the user into a text feature map, so as to facilitate the subsequent multi-layer convolution layer and attention layer model to classify the feature map. Furthermore, the embodiment of the present invention calculates the convolution resolution map of the factor image to extract features from the factor image. Furthermore, the embodiment of the present invention calculates the attention map of the convolution resolution map to perform simple linear weighting on all input vectors to further mine the detail features of the convolution resolution map. Furthermore, the embodiment of the present invention constructs a multi-branch map of the factor image based on the convolution resolution map and the attention map to realize the detail feature extraction of the factor image using branches of different resolution sizes. Secondly, the embodiment of the present invention performs branch map fusion processing on the multi-branch map to merge the features extracted from the image into features with better discriminative ability than the input features. Furthermore, the embodiment of the present invention calculates the fusion. The claim level probability corresponding to the branch diagram is used to output the predicted probability of the claim level of the factor image using the neural network model to realize the classification of the claim level of the factor image, and subsequently dynamically weight the probabilities with similar values, unclear distinction, and inability to judge the final level in the output predicted probabilities, so as to improve the certainty of the output predicted probability; further, the embodiment of the present invention calculates the dynamic allocation index of the claim level probability to mark the uncertain probability in the claim level probability with the dynamic allocation index; further, the embodiment of the present invention calculates the dynamic allocation weight of the claim level probability by using the dynamic allocation index to allocate an adaptive weight to the uncertain probability in the claim level probability, and converts the claim assessment category with unclear numerical boundaries into a certain claim assessment category by using the dynamic weight, so as to improve the certainty of the claim analysis; further, the embodiment of the present invention queries the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability, so as to determine the final claim level after the in-place probability is allocated with the dynamic weight, and is used to grade the claim materials submitted by the user, so as to facilitate the subsequent claim analysis. Therefore, a claims analysis method based on multi-factor dynamic weights proposed in an embodiment of the present invention can utilize dynamic weights to convert claims assessment categories with unclear numerical boundaries into definite claims assessment categories, thereby improving the certainty of claims analysis.

[0128] like Figure 4 The figure shows a functional module diagram of the claims analysis device based on multi-factor dynamic weights of the present invention.

[0129] The claim analysis device 400 based on multi-factor dynamic weights of the present invention can be installed in an electronic device. According to the functions implemented, the claim analysis device based on multi-factor dynamic weights can include a multi-branch diagram construction module 401, a claim probability calculation module 402, a claim level query module 403 and a claim item identification module 404. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0130] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0131] The multi-branch graph construction module 401 is used to obtain claim factors, construct a factor image of the claim factors, calculate a convolution resolution map of the factor image, and calculate an attention map of the convolution resolution map, and construct a multi-branch graph of the factor image based on the convolution resolution map and the attention map;

[0132] The claim probability calculation module 402 is used to perform branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculate the claim level probability corresponding to the fused branch graph;

[0133] The claim level query module 403 is used to calculate the dynamic allocation index of the claim level probability, calculate the dynamic allocation weight of the claim level probability using the dynamic allocation index, and query the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability;

[0134] The claim item identification module 404 is used to identify the claim items corresponding to the claim factors according to the final claim level, and use the claim items as the claim analysis results corresponding to the claim factors.

[0135] In detail, each module in the claim analysis device 400 based on multi-factor dynamic weights in the embodiment of the present invention is used in the same manner as described above. Figures 1 to 3 The same technical means are used as the claims analysis method based on multi-factor dynamic weights described in , and can produce the same technical effects, so I will not go into details here.

[0136] like Figure 5 The figure is a schematic diagram of the structure of an electronic device for implementing a claims analysis method based on multi-factor dynamic weights according to the present invention.

[0137] The electronic device may include a processor 50, a memory 51, a communication bus 52 and a communication interface 53, and may also include a computer program stored in the memory 51 and executable on the processor 50, such as a claims analysis program based on multi-factor dynamic weights.

[0138] In some embodiments, the processor 50 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 50 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 51 (for example, executing a claims analysis program based on multi-factor dynamic weights, etc.), and calls data stored in the memory 51 to execute various functions of the electronic device and process data.

[0139] The memory 51 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. The memory 51 may be an internal storage unit of an electronic device in some embodiments, such as a mobile hard disk of the electronic device. The memory 51 may also be an external storage device of an electronic device in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 51 may also include both an internal storage unit of the electronic device and an external storage device. The memory 51 may not only be used to store application software and various types of data installed in the electronic device, such as the code of the database configuration connection program, but also be used to temporarily store data that has been output or is to be output.

[0140] The communication bus 52 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 51 and at least one processor 50, etc.

[0141] The communication interface 53 is used for communication between the above-mentioned electronic device 5 and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visual user interface.

[0142] Figure 5 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 5 The structure shown does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0143] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 50 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0144] It should be understood that the embodiment is for illustrative purposes only and the scope of the patent invention is not limited to this structure.

[0145] The database configuration connection program stored in the memory 51 in the electronic device is a combination of multiple computer programs. When running in the processor 50, it can achieve:

[0146] Acquire a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, calculate an attention map of the convolution resolution map, and construct a multi-branch map of the factor image based on the convolution resolution map and the attention map;

[0147] Performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph;

[0148] Calculating the dynamic allocation index of the claim level probability, calculating the dynamic allocation weight of the claim level probability using the dynamic allocation index, and querying the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability;

[0149] According to the final claim level, the claim items corresponding to the claim factors are identified, and the claim items are used as the claim analysis results corresponding to the claim factors.

[0150] Specifically, the specific implementation method of the processor 50 for the above computer program can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0151] Furthermore, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0152] The present invention further provides a storage medium, wherein the readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0153] Acquire a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, calculate an attention map of the convolution resolution map, and construct a multi-branch map of the factor image based on the convolution resolution map and the attention map;

[0154] Performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph;

[0155] Calculating the dynamic allocation index of the claim level probability, calculating the dynamic allocation weight of the claim level probability using the dynamic allocation index, and querying the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability;

[0156] According to the final claim level, the claim items corresponding to the claim factors are identified, and the claim items are used as the claim analysis results corresponding to the claim factors.

[0157] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0158] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0159] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0160] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0161] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0162] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0163] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A claims analysis method based on multi-factor dynamic weights, characterized in that: The method comprises: Obtaining a claim factor, constructing a factor image of the claim factor, calculating a convolution resolution map of the factor image, and calculating an attention map of the convolution resolution map, and constructing a multi-branch map of the factor image based on the convolution resolution map and the attention map, wherein constructing the factor image of the claim factor includes: Dividing the state word blocks of the claim factors; Identify the state set and state category of the state word block; Based on the state set and the state category, construct a factor image of the claim factor by using one-hot encoding; The step of constructing a multi-branch graph of the factor image based on the convolution resolution graph and the attention graph includes: Downsampling the convolution resolution map to obtain a downsampling branch map of the factor image; Performing multi-layer convolution processing and multi-layer attention extraction processing on the attention map, and performing multi-layer convolution processing and multi-layer attention extraction processing on the down-sampled branch map to obtain a multi-branch map of the factor image; Performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph; Calculate the dynamic allocation index of the claim level probability, calculate the dynamic allocation weight of the claim level probability using the dynamic allocation index, and query the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability, wherein calculating the dynamic allocation index of the claim level probability includes: The following formula is used to identify the probability of uncertainty in the claim level probability: in, represents the probability of uncertainty in the claim level probability, represents the probability of the sequence number n in the activation output probability marked with the symbol u in the claim level probability, u represents the mark symbol of the activation output probability in the claim level probability, and n represents the probability sequence number in the activation output probability in each claim level probability, The symbol representing the probability of activation outputs other than u in the claim level probability, and also The sequence number of the uncertainty probability, Indicates when Much smaller than hour, is the probability of uncertainty in the probability of the claim level; The dynamic allocation index of the uncertainty probability is calculated using the following formula: in, represents the dynamic allocation index, represents the uncertainty probability, represents the total number of uncertain probabilities, a sequence number representing the uncertainty probability; According to the final claim level, the claim items corresponding to the claim factors are identified, and the claim items are used as the claim analysis results corresponding to the claim factors.

2. The method according to claim 1, characterized in that The calculating the attention map of the convolution resolution map includes: Performing channel processing on the convolution resolution map to obtain a channel processing map; Calculating the attention weight of the channel processing graph; Performing nonlinear transformation on the convolution resolution map to obtain a nonlinear transformation map; determining an activation value of the nonlinear transformation map; The attention weight and the activation value are concatenated into a feature map to obtain the attention map.

3. The method according to claim 2, characterized in that The calculating the attention weight of the channel processing graph includes: The following formula is used to calculate the proportion of non-zero feature points in the channel processing image: in, represents the proportion of non-zero feature points, represents the number of non-zero feature points in the feature map with sequence number i in the channel processing map, i represents the sequence number of the feature map in the channel processing map, represents the size of the feature map in the channel processing map, represents the width of the feature map in the channel processing map, Represents the height of the feature map in the channel processing map; According to the proportion of non-zero feature points, the weight of the non-zero feature points in the channel processing graph is calculated using the following formula: in, represents the weight of the non-zero feature point in the channel processing graph, represents the proportion of the non-zero feature points, i represents the sequence number of the feature map in the channel processing map, and k represents the total number of feature maps in the channel processing map; Based on the non-zero feature point weights, the attention weight of the channel processing graph is calculated using the following formula: in, represents the attention weight of the channel processing graph, represents the proportion of the non-zero feature points, i represents the sequence number of the feature map in the channel processing map, represents the weight of the non-zero feature point in the channel processing graph, Represents adjustment parameters, used to represent The adjustment affects the attention weights in the range .

4. The method according to claim 1, characterized in that The performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph includes: Identifying an upper branch graph and a lower branch graph in the multi-branch graph; Performing branch graph fusion processing on the upper branch graph in the lower branch graph by a down sampling method to obtain a first fused branch graph; Performing branch graph fusion processing on the lower branch graph in the upper branch graph by an upsampling method to obtain a second fused branch graph; The first fused branch graph and the second fused branch graph are used as the fused branch graph.

5. The method according to claim 1, characterized in that The querying of the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability includes: Based on the dynamic allocation weight and the claim level probability, the final claim probability of the claim factor is calculated using the following formula: in, represents the final claim probability of the claim factor, The weight representing the probability of the claim level includes the dynamically assigned weight for the uncertainty probability and the unassigned weight for the deterministic probability. The symbol representing the probability of activating the output in the claim level probability, including u and , that is, the probability sequence number in the activation output probability in each of the claim level probabilities, and the marking symbol of the activation output probability other than u in the claim level probability, represents the sequence number of the probability in the activation output probability in the claim level probability, represents the claim level probability, argmax represents the argmax function; Querying a target claim probability with the largest probability value from the final claim probability; The claim level corresponding to the target claim probability is obtained, and the target claim level corresponding to the target claim probability is taken as the final claim level from the claim levels.

6. The method according to claim 1, characterized in that The step of identifying the claim items corresponding to the claim factors according to the final claim level includes: Determine the initial claim items corresponding to the final claim level; Identify the factor category of the claim factor; The target claim item corresponding to the factor category is searched in the initial claim item, and the target claim item is used as the claim item.

7. A claims analysis device based on multi-factor dynamic weights, characterized in that: The device comprises: A multi-branch graph construction module is used to obtain a claim factor, construct a factor image of the claim factor, calculate a convolution resolution map of the factor image, and calculate an attention map of the convolution resolution map. Based on the convolution resolution map and the attention map, a multi-branch graph of the factor image is constructed, wherein the construction of the factor image of the claim factor includes: Dividing the state word blocks of the claim factors; Identify the state set and state category of the state word block; Based on the state set and the state category, construct a factor image of the claim factor by using one-hot encoding; The step of constructing a multi-branch graph of the factor image based on the convolution resolution graph and the attention graph includes: Downsampling the convolution resolution map to obtain a downsampling branch map of the factor image; Performing multi-layer convolution processing and multi-layer attention extraction processing on the attention map, and performing multi-layer convolution processing and multi-layer attention extraction processing on the down-sampled branch map to obtain a multi-branch map of the factor image; A claim probability calculation module, used for performing branch graph fusion processing on the multiple branch graphs to obtain a fused branch graph, and calculating the claim level probability corresponding to the fused branch graph; A claim level query module is used to calculate the dynamic allocation index of the claim level probability, calculate the dynamic allocation weight of the claim level probability using the dynamic allocation index, and query the final claim level of the claim factor based on the dynamic allocation weight and the claim level probability, wherein the calculation of the dynamic allocation index of the claim level probability includes: The following formula is used to identify the probability of uncertainty in the claim level probability: in, represents the probability of uncertainty in the claim level probability, represents the probability of the sequence number n in the activation output probability marked with the symbol u in the claim level probability, u represents the mark symbol of the activation output probability in the claim level probability, and n represents the probability sequence number in the activation output probability in each claim level probability, The symbol representing the probability of activation outputs other than u in the claim level probability, and also The sequence number of the uncertainty probability, Indicates when Much smaller than hour, is the probability of uncertainty in the probability of the claim level; The dynamic allocation index of the uncertainty probability is calculated using the following formula: in, represents the dynamic allocation index, represents the uncertainty probability, represents the total number of uncertain probabilities, a sequence number representing the uncertainty probability; The claim item identification module is used to identify the claim items corresponding to the claim factors according to the final claim level, and use the claim items as the claim analysis results corresponding to the claim factors.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the claims analysis method based on multi-factor dynamic weights as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the claims analysis method based on multi-factor dynamic weights as described in any one of claims 1 to 6.

Images