Model training method and claim settlement element extraction method

Abstract

The invention discloses a model training method and a claim settlement element extraction method, and the method comprises the steps: marking an entity in each insurance clause in an insurance clause data set, and obtaining a first data set, the entity comprising a claim settlement element, an element value corresponding to the claim settlement element, and a semantic segment comprising the claim settlement element; aiming at each insurance clause in the first data set, a training instruction is generated based on the insurance clause and the entity corresponding to the insurance clause by utilizing the target template, the training instruction comprises a command part, an input part and an output part, the input part corresponds to the insurance clause, and the output part corresponds to the entity marked in the insurance clause; and a second data set is utilized to train a preset claim settlement element extraction large model to obtain a trained claim settlement element extraction large model, and the second data set comprises a training instruction corresponding to each insurance term.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and specifically relates to a model training method and a claim element extraction method. Background Art

[0002] Traditional information extraction methods usually treat information as a whole, which makes it impossible to conduct in-depth analysis and extract specific details, and there is a problem of too large granularity. Due to the lack of refined processing, the extracted information is not accurate, and redundant or erroneous data is prone to occur, affecting subsequent analysis and application effects. Therefore, traditional information extraction methods are difficult to effectively improve the extraction accuracy of elements in the insurance field. Summary of the invention

[0003] The embodiments of the present application provide a model training method and a claims element extraction method, which can solve the problem of low element extraction accuracy in traditional information extraction methods.

[0004] In a first aspect, an embodiment of the present application provides a model training method, the method comprising: labeling entities in each insurance clause in an insurance clause data set to obtain a first data set, wherein the entities include: claim elements, element values ​​corresponding to the claim elements, and semantic fragments containing the claim elements; for each of the insurance clauses in the first data set, using a target template, and generating training instructions based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part, and an output part, the input part corresponds to the insurance clause, and the output part corresponds to the entities labeled in the insurance clause; using a second data set to train a preset claim element extraction large model to obtain a trained claim element extraction large model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses. In a second aspect, an embodiment of the present application provides a method for extracting claim elements, the method comprising: obtaining insurance clauses to be processed; using a claim element extraction large model to extract claim elements and element values ​​corresponding to the claim elements from the insurance clauses to be processed, wherein the claim element extraction large model is a model trained using the model training method described in the first aspect; outputting the claim elements and element values ​​corresponding to the claim elements.

[0005] In a third aspect, an embodiment of the present application provides a model training device, the model training device comprising: A labeling module is used to label the entities in each insurance clause in the insurance clause data set to obtain a first data set, wherein the entities include: claim elements, element values ​​corresponding to the claim elements, and semantic fragments containing the claim elements; a generation module is used to generate training instructions for each of the insurance clauses in the first data set using a target template and based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part and an output part, wherein the input part corresponds to the insurance clause, and the output part corresponds to the entities labeled in the insurance clause; a training module is used to train a preset claim element extraction model using a second data set to obtain a trained claim element extraction model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses.

[0006] In a fourth aspect, an embodiment of the present application provides a claim element extraction device, which includes: an acquisition module for acquiring insurance clauses to be processed; an extraction module for using a claim element extraction large model to extract claim elements and element values ​​corresponding to the claim elements from the insurance clauses to be processed, wherein the claim element extraction large model is a model trained using the model training method described in the second aspect; an output module for outputting the claim elements and element values ​​corresponding to the claim elements.

[0007] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect, or implements the steps of the method described in the second aspect.

[0008] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method described in the second aspect are implemented.

[0009] In the seventh aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method described in the first aspect, or to implement the steps of the method described in the second aspect.

[0010] In an eighth aspect, an embodiment of the present application provides a computer program product, which includes at least one computer program, and when the computer program is loaded and executed by a processor, it implements the method described in the first aspect, or implements the steps of the method described in the second aspect.

[0011] In an embodiment of the present application, a first data set is obtained by annotating entities in each insurance clause in an insurance clause data set, wherein the entities include: claim elements, element values ​​corresponding to claim elements, and semantic fragments containing claim elements. Then, for each of the insurance clauses in the first data set, a target template is used, and training instructions are generated based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part, and an output part, the input part corresponds to the insurance clauses, and the output part corresponds to the entities annotated in the insurance clauses. Finally, a second data set is used to train a preset claim element extraction large model to obtain a trained claim element extraction large model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses, so that the trained claim element extraction large model has higher robustness, can improve the accuracy and effectiveness of claim element extraction, thereby meeting subsequent data processing requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flowchart of a model training method provided in an embodiment of the present application; Figure 2 It is a flowchart of a method for extracting claim elements provided in an embodiment of the present application; Figure 3 It is a structural schematic diagram of a model training device provided in an embodiment of the present application; Figure 4 It is a structural schematic diagram of a claim element extraction device provided in an embodiment of the present application; Figure 5 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0014] In the following, in combination with the accompanying drawings, a model training method and a claim element extraction method provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0015] Figure 1A model training method provided by an embodiment of the present application is shown, and the method can be executed by an electronic device, and the electronic device may include: a server and / or a terminal device. In other words, the method can be executed by software or hardware installed in the electronic device, and the method includes the following steps: S110: Labeling entities in each insurance clause in the insurance clause data set to obtain a first data set.

[0016] The entities include: claim elements, element values ​​corresponding to the claim elements, and semantic fragments containing the claim elements.

[0017] S110 may include the following steps: S111: Acquire the insurance clause data set, wherein the insurance clause data set includes a plurality of insurance clauses.

[0018] S112: For each of the insurance clauses, mark the semantic fragments containing the claim elements.

[0019] S113: For each semantic segment containing the claim element, mark the claim element and the element value corresponding to the claim element.

[0020] That is to say, first mark out the complete semantic fragments where the key information of the claim is located, and then mark out the various claim elements from each fragment, where the claim elements include but are not limited to: claim time limit, compensation amount, policyholder, insured, etc. For example, assume that a certain insurance clause includes the following content: after deducting other expenses, the remaining amount will be paid according to the following agreement: if not reimbursed by medical insurance, the deductible is 100 yuan and the payment ratio is 90%; if reimbursed by medical insurance, the deductible for the remaining reimbursable expenses is 0 yuan and the payment ratio is 90%. Among them, in this insurance clause, the semantic fragments containing claim elements are: if not reimbursed by medical insurance, the deductible is 100 yuan and the payment ratio is 90%; if reimbursed by medical insurance, the deductible for the remaining reimbursable expenses is 0 yuan and the payment ratio is 90%. The claim elements in the semantic fragment containing claim elements include: medical insurance deductible, medical payment ratio, non-medical insurance deductible, and non-medical payment ratio. Among them, the element value corresponding to the medical insurance deductible is 0 yuan, the element value corresponding to the medical payment ratio is 90%, the element value corresponding to the non-medical insurance deductible is 100 yuan, and the element value corresponding to the non-medical payment ratio is 90%.

[0021] Optionally, after obtaining the first data set, the method further comprises: cleaning the first data set, that is, deleting insurance clauses that are not marked entities.

[0022] S120: For each of the insurance clauses in the first data set, a target template is used to generate training instructions based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part, and an output part, the input part corresponds to the insurance clause, and the output part corresponds to the entity marked in the insurance clause.

[0023] It is understandable that after acquiring the annotated first data set, for each of the insurance clauses in the first data set, a training instruction is generated based on the insurance clause and the insurance clause, wherein the training instruction may be a prompt instruction. The training instruction may be generated based on a preset target template. Exemplarily, the target template may be: { "instruct": "Please extract the key information of claim elements such as deductible, compensation ratio, compensation limit, etc. from the following insurance terms. Please note that the insurance terms are long and need to be extracted step by step to improve the accuracy of claim element extraction." “input”: “insurance terms” “output”: “In the above insurance clause, the semantic segments containing claim elements are: [span_1, span_2…, span_n], where the claim elements contained in [span_1] include [y_1, y_2], the element value corresponding to [y_1] is [x_1], the element value corresponding to [y_2] is [x_2], the claim elements contained in [span_2] include [y_3, y_4], the element value corresponding to [y_3] is [x_3], the element value corresponding to [y_4] is [x_4], … }.

[0024] Among them, "instruct" is used to indicate the command part, "input" is used to indicate the input part, and "output" is used to indicate the output part. S130: Using the second data set to train the preset large model for extracting claim elements, to obtain a trained large model for extracting claim elements.

[0025] Wherein, the second data set includes the training instructions corresponding to each of the insurance clauses.

[0026] That is to say, the second data set is composed of the training instructions corresponding to each of the insurance clauses, and then the large model for extracting claim elements is trained based on the second data set. Among them, the large model for extracting claim elements can be a large language model (LLM, Large Language Model). After training, the large language model has a wide range of language understanding capabilities, can handle various complex language structures and contexts, and can not only identify explicit elements, but also extract key elements from implicit semantics. For example, assuming that an insurance clause is: the vaccine fee for animal bites and scratches can be included in the calculation at a maximum of 52 yuan per injection, then, through the trained large model for extracting claim elements, the claim element: the single injection limit for animal injuries can be extracted, and its corresponding value is 52 yuan. At the same time, the large language model can also be fine-tuned by adjusting the parameter amount or combining additional data to achieve element extraction of different granularities.

[0027] In this step, through natural language training instructions, the extraction goals and detailed requirements can be more clearly defined, so that the large model for extracting claim elements can focus more clearly on specific information when performing tasks, which can significantly reduce errors caused by task ambiguity, thereby improving extraction accuracy. At the same time, when faced with texts with complex grammatical structures or long dependencies, the training instructions can help the large model for extracting claim elements to better understand the semantic structure of the text, thereby improving extraction accuracy, because in the first data set, the nested entities in the insurance terms are annotated, so that the large model for extracting claim elements can more accurately parse the relationship between different entities and reduce the error rate. For example, the entity "semantic fragment containing the claim element" belongs to the first level, and the entity "claim element" and the entity "element value corresponding to the claim element" belong to the second level.

[0028] In an embodiment of the present application, a first data set is obtained by annotating entities in each insurance clause in an insurance clause data set, wherein the entities include: claim elements, element values ​​corresponding to claim elements, and semantic fragments containing claim elements. Then, for each of the insurance clauses in the first data set, a target template is used, and training instructions are generated based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part, and an output part, the input part corresponds to the insurance clauses, and the output part corresponds to the entities annotated in the insurance clauses. Finally, a second data set is used to train a preset claim element extraction large model to obtain a trained claim element extraction large model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses, so that the trained claim element extraction large model has higher robustness, can improve the accuracy and effectiveness of claim element extraction, thereby meeting subsequent data processing requirements.

[0029] In one implementation, before generating training instructions for each of the insurance clauses in the first data set using a target template and based on the insurance clauses and the entities corresponding to the insurance clauses, the method further includes: enhancing the first data set based on the labeled entities.

[0030] It is understandable that by enhancing the first data set, the diversity of training data can be increased, which can help the large model for extracting claim elements to learn language features and semantic relationships more comprehensively, thereby showing higher extraction accuracy and robustness when facing complex real-world scenarios. In addition, by enhancing the first data set, more training samples can be generated, thereby avoiding the impact of insufficient training data on the learning ability of the large model for extracting claim elements.

[0031] In addition, since the annotated entity includes the semantic segment containing the claim element and the element value corresponding to the claim element, the semantic segment containing the claim element and the element value corresponding to the claim element can be enhanced respectively: (1) In one implementation, the enhancing the first data set based on the annotated entity includes: performing a first enhancement process on the first data set based on the annotated semantic segment containing the claim element to obtain at least one first restructured insurance clause; and adding the at least one first restructured insurance clause to the first data set.

[0032] The first enhancement process includes at least one of the following: Exchanging the semantic fragments containing the claim elements in any two of the insurance clauses; For any of the insurance clauses, deleting the semantic segment containing the claim element from a first position, and inserting the semantic segment containing the claim element into a second position of the insurance clause, wherein the second position is any other position in the insurance clause except the first position; For any of the insurance clauses, insert stop words or invalid words into the insurance clauses; For any of the insurance clauses, a stop word or an invalid word is inserted into at least one semantic segment containing the claim element corresponding to the insurance clause.

[0033] It can be understood that the method of performing data enhancement processing on the semantic fragment and generating the first reorganized insurance clause may include: randomly exchanging a semantic fragment with other insurance clause texts to generate the first reorganized insurance clause; deleting the semantic fragment from the original position of the original text and randomly inserting it to other positions to generate the first reorganized insurance clause; randomly inserting stop words or invalid words into the insurance clause to generate the first reorganized insurance clause; randomly inserting stop words or invalid words into the semantic fragment containing the claim element corresponding to the insurance clause to generate the first reorganized insurance clause, wherein the stop words or invalid words are used to increase noise to improve the robustness of the large model for claim element extraction.

[0034] (2) In one implementation, the enhancing the first data set based on the marked entity includes: performing a second enhancement process on the first data set based on the element value corresponding to the marked claim element to obtain at least one second restructured insurance clause; and adding the at least one second restructured insurance clause to the first data set.

[0035] The second enhancement process includes at least one of the following: For any of the insurance clauses, modifying the element value corresponding to at least one of the claim elements in the insurance clause to a first value, wherein the first value is within a preset range, and the first values ​​corresponding to different claim elements are different; For any of the insurance clauses, the element value corresponding to at least one of the claim elements in the insurance clause is modified to a second value, and the second value is marked as a result, wherein the second value is outside a preset range, and the result marking is used to indicate that the second value is incorrect, and the second values ​​corresponding to different claim elements are different.

[0036] It is understandable that the enhancement of claim elements may include: within a preset range, randomly modifying the element value corresponding to at least one of the claim elements, for example, replacing "the compensation ratio is 85%" with "the compensation ratio is 25%"; outside the preset range, randomly modifying the element value corresponding to at least one of the claim elements, for example, replacing "the compensation ratio is 85%" with "the compensation ratio is 135%", where "135%" is not within the preset range and can therefore be marked as incorrect to help the model understand and thus enhance the model's learning ability.

[0037] In one implementation, the use of the second data set to train the preset claim element extraction large model to obtain a trained claim element extraction large model includes: dividing the training instructions in the second data set into N batches of training data to obtain a third data set consisting of N batches of training data, wherein N is an integer greater than 0, and each batch includes a preset number of the training instructions; using the third data set, performing multiple rounds of iterative training on the preset claim element extraction large model, each round of training using a batch of training data in the third data set.

[0038] It is understandable that the entire training process will be iterated multiple times, that is, multiple batches will be traversed until all batches of data are trained or the preset number of iterations is reached. Each training batch is called an epoch, that is, a round of training. Through multiple epochs, the weight of the model can be continuously optimized to gradually improve its extraction accuracy on the entire data set.

[0039] In this implementation, since there is new sample data in each iteration, the large model for claim factor extraction can adapt to changes more quickly. Therefore, small batch training can update the model weights more frequently, thereby accelerating the convergence process.

[0040] Furthermore, in another implementation method, the third data set is used to perform multiple rounds of iterative training on the preset claim element extraction large model, including: in each round of training, for each of the training instructions, calculating the target cross entropy loss between the standard probability corresponding to the marked output part and the predicted probability corresponding to the output part predicted by the claim element extraction large model; summing the target cross entropy losses corresponding to each of the training instructions to obtain a total cross entropy loss; based on the total cross entropy loss, using a stochastic gradient descent algorithm to update the weights of the claim element extraction large model.

[0041] It can be understood that in each iteration, for each training instruction in the current batch, the large model for claim factor extraction calculates the target cross entropy loss based on the predicted output and the true annotation, and the target cross entropy loss is used to quantify the gap between the predicted entity and the pre-annotated entity. Then the target cross entropy losses corresponding to all training instructions in the current batch are added together to obtain the total loss value of the batch, i.e., the total cross entropy loss. The total cross entropy loss can reflect the overall performance of the large model for claim factor extraction on the current batch. In this way, a moderate batch size can enable the large model for claim factor extraction to gradually adjust its weights, and balance gradient fluctuations and stability, further improving the final performance of the model.

[0042] Among them, for each of the training instructions, calculating the target cross entropy loss between the standard probability corresponding to the labeled entity and the predicted probability of the predicted entity output by the claim element extraction large model means serializing the training instruction, for example: <istct>{instruct} <eistct> <input> {input} <einput> <response>{output} <eresponse>,in, <istct> 、 <eistct> 、 <input> 、 <einput> 、 <response> 、 <eresponse>Used to indicate the beginning or end of the corresponding content, and then maximize the predicted probability of the sequence, for example: L(X)= ; Among them, L(X) represents the predicted probability of the sequence, Represents each word in the sequence. It should be noted that in the training instruction, since the instruction and input parts are known information, in order to reduce the amount of calculation, increase the calculation speed and ensure the fine-tuning effect, in the embodiment of the present application, only the cross entropy summation is performed on the text of the output part. Exemplarily, the target cross entropy loss can be expressed by the following formula: Target-Cross-Entropy(p, ) = sum(plog( )) ,i>index of <einput>.

[0043] in, <einput>Indicates the end of the input part, i>index of <einput>Represents the characters in the output part, P is the standard probability, is the predicted probability.

[0044] like Figure 2 As shown, the embodiment of the present application also provides a method for extracting claim elements, using the above Figure 1 The model obtained by training the model training method in the method embodiment is used for recognition, and the method comprises the following steps: S210: Obtain the insurance terms to be processed.

[0045] S220: Utilizing a large model for extracting claim elements, extracting claim elements and element values ​​corresponding to the claim elements from the insurance clauses to be processed.

[0046] Among them, the large model for extracting claim elements is based on Figure 1 The model is obtained by training using the model training method in the illustrated embodiment.

[0047] S230: Output the claim elements and the element values ​​corresponding to the claim elements.

[0048] In the embodiment of the present application, the insurance clauses to be processed are first obtained, and then the claim element extraction model is used to extract the claim elements and the element values ​​corresponding to the claim elements from the insurance clauses to be processed, and finally the claim elements and the element values ​​corresponding to the claim elements are output. Figure 1 The model obtained by training the model training method provided in the illustrated embodiment has strong robustness and high extraction accuracy, so that the final output claim elements are more accurate.

[0049] Figure 3 A schematic diagram of the structure of a model training device provided in an embodiment of this specification is shown. Figure 3 As shown, the model training device 300 may include: a labeling module 310 , a generating module 320 and a training module 330 .

[0050] In this embodiment, the labeling module 310 is used to label the entities in each insurance clause in the insurance clause data set to obtain a first data set, wherein the entities include: claim elements, element values ​​corresponding to the claim elements, and semantic fragments containing the claim elements; the generation module 320 is used to generate training instructions for each of the insurance clauses in the first data set using a target template and based on the insurance clauses and the entities corresponding to the insurance clauses, wherein the training instructions include a command part, an input part and an output part, wherein the input part corresponds to the insurance clause, and the output part corresponds to the entities labeled in the insurance clause; the training module 330 is used to train the preset claim element extraction model using the second data set to obtain a trained claim element extraction model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses.

[0051] In one implementation, the model training device 300 may further include an enhancement module for performing enhancement processing on the first data set based on the labeled entities.

[0052] In one implementation, the enhancement module is also specifically used to perform a first enhancement processing on the first data set based on the marked semantic fragments containing the claim elements to obtain at least one first reorganized insurance clause; add the at least one first reorganized insurance clause to the first data set; wherein the first enhancement processing includes at least one of the following: exchanging the semantic fragments containing the claim elements in any two of the insurance clauses; for any of the insurance clauses, deleting the semantic fragment containing the claim elements from the first position, and inserting the semantic fragment containing the claim elements into the second position of the insurance clause, wherein the second position is any other position in the insurance clause except the first position; for any of the insurance clauses, inserting stop words or invalid words into the insurance clause; for any of the insurance clauses, inserting stop words or invalid words into at least one of the semantic fragments containing the claim elements corresponding to the insurance clause.

[0053] In one implementation, the enhancement module is further specifically used to perform a second enhancement processing on the first data set based on the element value corresponding to the marked claim element to obtain at least one second reorganized insurance clause; and add the at least one second reorganized insurance clause to the first data set; wherein the second enhancement processing includes at least one of the following: for any of the insurance clauses, modifying the element value corresponding to at least one of the claim elements in the insurance clause to a first value, wherein the first value is within a preset range, and the first values ​​corresponding to different claim elements are different; for any of the insurance clauses, modifying the element value corresponding to at least one of the claim elements in the insurance clause to a second value, and performing result annotation on the second value, wherein the second value is outside the preset range, and the result annotation is used to indicate that the second value is incorrect, and the second values ​​corresponding to different claim elements are different.

[0054] In one implementation, the training module 330 is further used to divide the training instructions in the second data set into N batches of training data to obtain a third data set consisting of N batches of training data, wherein N is an integer greater than 0, and each batch includes a preset number of the training instructions; using the third data set, the preset large model for extracting claims elements is subjected to multiple rounds of iterative training, and each round of training uses a batch of training data in the third data set.

[0055] In one implementation, the training module 330 is also used to calculate, for each of the training instructions in each round of training, the target cross entropy loss between the standard probability corresponding to the marked output part and the predicted probability corresponding to the output part predicted by the claim element extraction large model; summing the target cross entropy losses corresponding to each of the training instructions to obtain a total cross entropy loss; and based on the total cross entropy loss, using a stochastic gradient descent algorithm to update the weights of the claim element extraction large model.

[0056] A model training device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented in the illustrated method embodiment will not be described again here.

[0057] Figure 4 A claim element extraction device provided in an embodiment of the present application is shown. The claim element extraction device 400 may include: an acquisition module 410, an extraction module 420, and an output module 430.

[0058] In this embodiment, the acquisition module 410 is used to obtain the insurance clauses to be processed; the extraction module 420 is used to extract the claim elements and the element values ​​corresponding to the claim elements from the insurance clauses to be processed by using the claim element extraction model, wherein the claim element extraction model is based on the above Figure 1 The model is obtained by training using the model training method shown; an output module 430 is used to output the claim elements and the element values ​​corresponding to the claim elements.

[0059] The claim element extraction device provided in the embodiment of the present application can achieve Figure 2 To avoid repetition, the various processes implemented in the illustrated method embodiment will not be described again here.

[0060] A model training device and a claim element extraction device in the embodiment of the present application can be a device, or a component, integrated circuit, or chip in an electronic device. The embodiment of the present application is not specifically limited.

[0061] A model training device and a claim element extraction device in the embodiment of the present application may be a device with an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0062] Optional, such as Figure 5 As shown, an embodiment of the present application also provides an electronic device 500, including a processor 510, a memory 520, and a program or instruction stored in the memory 520 and executable on the processor 510. When the program or instruction is executed by the processor 510, the various processes of the above-mentioned model training method embodiment are implemented, or the various processes of the above-mentioned claim element extraction method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0063] An embodiment of the present application also provides a computer-readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned model training method embodiment are implemented, or the various processes of the above-mentioned claim element extraction method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0064] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0065] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned model training method embodiment, or to implement the various processes of the above-mentioned claim element extraction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0066] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0067] An embodiment of the present application also provides a computer program product, which includes at least one computer program. When the computer program is loaded and executed by a processor, it implements the various processes of the above-mentioned model training method embodiment, or implements the various processes of the above-mentioned claim element extraction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0068] It should be noted that, in this article, the terms "comprise", "include" 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, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0069] 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 magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0070] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.< / einput> < / einput> < / einput> < / eresponse> < / response> < / einput> < / eistct> < / istct> < / eresponse> < / response> < / einput> < / eistct> < / istct>

Claims

1. A model training method, characterized in that: include: Annotating entities in each insurance clause in the insurance clause data set to obtain a first data set, wherein the entities include: claim elements, element values ​​corresponding to the claim elements, and semantic segments containing the claim elements; For each of the insurance clauses in the first data set, using a target template, and generating a training instruction based on the insurance clause and the entity corresponding to the insurance clause, wherein the training instruction includes a command part, an input part, and an output part, the input part corresponds to the insurance clause, and the output part corresponds to the entity marked in the insurance clause; The preset claim element extraction large model is trained using the second data set to obtain a trained claim element extraction large model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses.

2. The method according to claim 1, characterized in that Before generating a training instruction for each of the insurance clauses in the first data set using a target template and based on the insurance clause and the entity corresponding to the insurance clause, the method further includes: Based on the marked entities, the first data set is enhanced.

3. The method according to claim 2, characterized in that The step of enhancing the first data set based on the labeled entity includes: Based on the annotated semantic segments containing the claim elements, performing a first enhancement process on the first data set to obtain at least one first reorganized insurance clause; adding the at least one first restructuring insurance clause to the first data set; The first enhancement process includes at least one of the following: Exchanging the semantic fragments containing the claim elements in any two of the insurance clauses; For any of the insurance clauses, deleting the semantic segment containing the claim element from a first position, and inserting the semantic segment containing the claim element into a second position of the insurance clause, wherein the second position is any other position in the insurance clause except the first position; For any of the insurance clauses, insert stop words or invalid words into the insurance clauses; For any of the insurance clauses, a stop word or an invalid word is inserted into at least one semantic segment containing the claim element corresponding to the insurance clause.

4. The method according to claim 2, characterized in that: The step of enhancing the first data set based on the labeled entity includes: Based on the element value corresponding to the marked claim element, performing a second enhancement process on the first data set to obtain at least one second restructured insurance clause; adding the at least one second restructuring insurance clause to the first data set; The second enhancement process includes at least one of the following: For any of the insurance clauses, modifying the element value corresponding to at least one of the claim elements in the insurance clause to a first value, wherein the first value is within a preset range, and the first values ​​corresponding to different claim elements are different; For any of the insurance clauses, the element value corresponding to at least one of the claim elements in the insurance clause is modified to a second value, and the second value is marked as a result, wherein the second value is outside a preset range, and the result marking is used to indicate that the second value is incorrect, and the second values ​​corresponding to different claim elements are different.

5. The method according to claim 1, characterized in that The method of using the second data set to train the preset claim factor extraction model to obtain a trained claim factor extraction model includes: Dividing the training instructions in the second data set into N batches of training data to obtain a third data set consisting of N batches of training data, wherein N is an integer greater than 0, and each batch includes a preset number of the training instructions; The third data set is used to perform multiple rounds of iterative training on the preset large model for extracting claim elements, and each round of training uses a batch of training data from the third data set.

6. The method according to claim 5, characterized in that The method of using the third data set to perform multiple rounds of iterative training on the preset large model for extracting claim elements includes: In each round of training, for each of the training instructions, the target cross entropy loss between the standard probability corresponding to the labeled output part and the predicted probability corresponding to the output part predicted by the large model for claim element extraction is calculated; The target cross entropy loss corresponding to each of the training instructions is summed to obtain a total cross entropy loss; Based on the total cross entropy loss, the weights of the claim factor extraction model are updated using a stochastic gradient descent algorithm.

7. A method for extracting claim elements, characterized in that: The method comprises: Obtain pending insurance terms; Utilizing a large model for extracting claim elements, extracting claim elements and element values ​​corresponding to the claim elements from the insurance clauses to be processed, wherein the large model for extracting claim elements is a model trained by the model training method described in any one of claims 1 to 6; Output the claim elements and the element values ​​corresponding to the claim elements.

8. A model training device, characterized in that: include: A labeling module is used to label entities in each insurance clause in the insurance clause data set to obtain a first data set, wherein the entities include: claim elements, element values ​​corresponding to the claim elements, and semantic segments containing the claim elements; a generating module, configured to generate, for each of the insurance clauses in the first data set, a training instruction using a target template and based on the insurance clause and the entity corresponding to the insurance clause, wherein the training instruction includes a command part, an input part and an output part, the input part corresponds to the insurance clause, and the output part corresponds to the entity marked in the insurance clause; A training module is used to train a preset claim factor extraction model using a second data set to obtain a trained claim factor extraction model, wherein the second data set includes the training instructions corresponding to each of the insurance clauses.

9. A claim element extraction device, characterized in that: The device comprises: An acquisition module, used to acquire insurance clauses to be processed; An extraction module, used to extract claim elements and element values ​​corresponding to the claim elements from the insurance clauses to be processed by using a claim element extraction model, wherein the claim element extraction model is a model trained by the model training method described in any one of claims 1 to 6; An output module is used to output the claim elements and the element values ​​corresponding to the claim elements.

10. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the model training method as described in any one of claims 1 to 6 are implemented, or the steps of the claim element extraction method as described in claim 7 are implemented.

11. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the steps of the model training method as described in any one of claims 1 to 6 are implemented, or the steps of the claim element extraction method as described in claim 7 are implemented.

12. A computer program product, characterized in that The computer program product includes program instructions, and when the program instructions are executed by a computer, the computer implements the steps of the model training method as described in any one of claims 1 to 6, or implements the steps of the claim element extraction method as described in claim 7.

Images