Compliance review information generation method and device, equipment and medium
By using a temporal neural network model and a compliance review annotation information generation model, accurate paragraph content vectors and compliance review annotation information are generated, solving the problem of low efficiency in the compliance review of policy documents and achieving efficient and accurate compliance review.
Patent Information
- Application Number
- CN202411402894.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-09
- Publication Date
- 2025-01-24
AI Technical Summary
Current technologies for policy document compliance review are inefficient. Manual sentence-by-sentence comparison is time-consuming, and the paragraph content vectors generated by conventional methods are not accurate enough to fully reflect the semantic information of the policy paragraphs.
An encoding model based on a temporal neural network, downsampling convolutional layers, and fully connected layers are used to generate accurate paragraph content vectors. A pre-trained compliance review annotation information generation model is used to generate a sequence of compliance review annotation information, and a comprehensive compliance review is performed by combining sentence and paragraph granularity.
It enables accurate and efficient compliance review of policy documents, improves review speed and accuracy, and allows for comprehensive assessment of compliance risks at the sentence and paragraph level.
Smart Images
Figure CN119358540A_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to methods, devices, equipment, and media for generating compliance review information. Background Art
[0002] At present, the promulgation of policy documents is of great significance. It can provide protection and benefits to individuals. For example, social security policy documents can provide individuals with basic social security such as medical insurance and pension insurance, and improve health and quality of life. For compliance review of policy documents, the usual method is to verify whether there are compliance risks in policy documents by manually comparing them sentence by sentence.
[0003] However, when using the above method to conduct compliance review of policy documents, the following technical problems often occur:
[0004] The manual comparison of sentences one by one results in a long review time and low efficiency.
[0005] In the process of adopting technical solutions to solve the above technical problems, the following technical problem often arises: how to use the sentence vector sequence to accurately generate the paragraph content vector. For the above technical problems, the conventional solution is generally to directly concatenate the sentence vectors in the sentence vector sequence to generate the paragraph content vector. However, the above conventional solution still has the following problem: the obtained paragraph content vector is not accurate enough and cannot fully reflect the content semantic information of the policy paragraph content.
[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention
[0007] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0008] Some embodiments of the present disclosure propose compliance review information generation methods, devices, equipment and media to solve one or more of the technical problems mentioned in the above background technology section.
[0009] In a first aspect, some embodiments of the present disclosure provide a compliance review information generation method, comprising: obtaining a target policy document; performing paragraph splitting on the target policy document to generate a policy paragraph content set; for each policy paragraph content in the policy paragraph content set, executing the following first generation step: performing sentence processing on the policy paragraph content to generate a policy statement sequence; generating a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence; generating a paragraph content vector corresponding to the policy paragraph content based on the sentence vector sequence; inputting the sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the policy statement sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; inputting the paragraph content vector into a pre-trained second compliance review annotation information generation model to generate a second compliance review annotation information corresponding to the policy paragraph content; generating compliance review comprehensive annotation information for the target policy document based on the obtained first compliance review annotation information sequence set and second compliance review annotation information set; and sending the compliance review comprehensive annotation information to a target review terminal.
[0010] In a second aspect, some embodiments of the present disclosure provide a compliance review information generation device, including: an acquisition unit, configured to acquire a target policy file; a splitting unit, configured to split the target policy file into paragraphs to generate a policy paragraph content set; an execution unit, configured to execute the following first generation step for each policy paragraph content in the policy paragraph content set: sentence processing the policy paragraph content to generate a policy statement sequence; generating a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence; generating a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence; and sending the sentence to The quantity sequence is input into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the above policy statement sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; the above paragraph content vector is input into a pre-trained second compliance review annotation information generation model to generate the second compliance review annotation information corresponding to the above policy paragraph content; a generation unit is configured to generate compliance review comprehensive annotation information for the above target policy document based on the obtained first compliance review annotation information sequence set and the second compliance review annotation information set; and the above compliance review comprehensive annotation information is sent to the target review terminal.
[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.
[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.
[0013] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the compliance review information generation method of some embodiments of the present disclosure, the target policy document can be accurately and efficiently subjected to a comprehensive compliance review. Specifically, the reason for the low efficiency of the relevant compliance review is that the review consumes a long time and is inefficient through the manual sentence-by-sentence comparison. Based on this, the compliance review information generation method of some embodiments of the present disclosure first obtains the target policy document to obtain the object file for subsequent compliance review. Then, the above-mentioned target policy document is segmented into paragraphs to generate a policy paragraph content set, so as to facilitate the subsequent compliance review of the content of each paragraph. Then, for each policy paragraph content in the above-mentioned policy paragraph content set, the following first generation step is performed: the first step is to perform sentence processing on the above-mentioned policy paragraph content to generate a policy statement sequence, so as to facilitate the subsequent compliance review of each statement. The second step is to generate a sentence vector corresponding to each policy statement in the above-mentioned policy statement sequence, and obtain a sentence vector sequence, which is not only generated in a form that is convenient for subsequent input to the model, but also used for the subsequent generation of accurate paragraph content vectors. In the third step, according to the above sentence vector sequence, the paragraph content vector corresponding to the above policy paragraph content can be accurately generated, so as to facilitate the subsequent compliance review of the paragraph content vector using the model. In the fourth step, the above sentence vector sequence is input into the pre-trained first compliance review annotation information generation model to generate the first compliance review annotation information sequence corresponding to the above policy sentence sequence. Among them, the first compliance review annotation information includes: compliance review details information and risk information. Here, through the pre-trained first compliance review annotation information generation model, not only can the compliance review of the target policy document at the sentence granularity be accurately performed, but also the review speed can be improved. In the fifth step, the above paragraph content vector is input into the pre-trained second compliance review annotation information generation model to accurately generate the second compliance review annotation information corresponding to the above policy paragraph content. Here, through the pre-trained second compliance review annotation information generation model, not only can the compliance review of the target policy document at the paragraph granularity be accurately performed, but also the review speed can be improved. Next, according to the obtained first compliance review annotation information sequence set and second compliance review annotation information set, the compliance review comprehensive annotation information for the above target policy document can be accurately generated. Finally, the above-mentioned compliance review comprehensive annotation information is sent to the target review terminal to display the comprehensive compliance review status corresponding to the target policy document. In summary, at the sentence granularity and paragraph granularity, the first compliance review annotation information generation model and the second compliance review annotation information generation model can accurately and efficiently implement the comprehensive compliance review of the target policy document. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.
[0015] Figure 1 is a flow chart of some embodiments of the method for generating compliance review information according to the present disclosure;
[0016] Figure 2 It is a schematic diagram of the structure of some embodiments of the compliance review information generating device according to the present disclosure;
[0017] Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION
[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.
[0019] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.
[0020] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0021] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".
[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.
[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.
[0024] refer to Figure 1, shows a process 100 of some embodiments of the method for generating compliance review information according to the present disclosure. The method for generating compliance review information includes the following steps:
[0025] Step 101, obtaining a target policy file.
[0026] In some embodiments, the executor of the above-mentioned compliance review information generation method (for example, an electronic device) can obtain the target policy document through a wired connection or a wireless connection. Among them, the target policy document may be a policy document to be subject to compliance review. The target policy document may be a document on relevant policy adjustments promulgated by various departments. For example, the target policy document may be a real estate policy adjustment document. The compliance review may be conducted based on the "Fair Competition Review Implementation Rules". Among them, the "Fair Competition Review Implementation Rules" may include four major categories, 18 subcategories and 54 subcategories.
[0027] Specifically, the four major categories, 18 subcategories and 54 detailed categories included in the "Fair Competition Review Implementation Rules" can be found in Table 1. The label sequence numbers corresponding to the specific contents can also be found in
[0028] The "Label" column in Table 1.
[0029]
[0030]
[0031]
[0032]
[0033]
[0034]
[0035] Table 1
[0036] Step 102: split the target policy document into paragraphs to generate a policy paragraph content set.
[0037] In some embodiments, the execution subject may divide the target policy document into paragraphs to generate a policy paragraph content set, wherein each policy paragraph content in the policy paragraph content set is the content of a certain document paragraph.
[0038] As an example, the execution entity may divide the target policy document into paragraphs by various paragraph division methods to generate a policy paragraph content set.
[0039] Step 103: for each policy paragraph content in the policy paragraph content set, perform the following first generation step:
[0040] Step 1031, segment the above policy paragraph content into sentences to generate a policy sentence sequence.
[0041] In some embodiments, the execution entity may perform sentence processing on the policy paragraph content to generate a policy statement sequence, wherein each policy statement in the policy statement sequence is each statement in the policy paragraph content.
[0042] As an example, the execution entity may perform sentence processing on the policy paragraph content according to each period in the policy paragraph content to generate a policy statement sequence.
[0043] Step 1032, generating a sentence vector corresponding to each policy statement in the above policy statement sequence to obtain a sentence vector sequence.
[0044] In some embodiments, the execution subject may generate a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence, wherein the sentence vector may represent the sentence semantics of the policy statement.
[0045] As an example, for each policy statement, first, the policy statement is segmented to generate a word set. Then, each word in the word set is embedded to generate a word vector to obtain a word vector set. Finally, the word vectors in the word vector set are concatenated to generate a sentence vector.
[0046] Step 1033: Generate a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence.
[0047] In some embodiments, the execution subject may generate a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence, wherein the paragraph content vector may represent the content semantic information of the policy paragraph content.
[0048] As an example, the execution subject may concatenate the sentence vectors in the sentence vector sequence to generate a concatenated vector, and then input the concatenated vector into a downsampling convolution layer to generate a paragraph content vector.
[0049] In some optional implementations of some embodiments, the execution entity may sequentially input each sentence vector in the sentence vector sequence into a coding model based on a temporal neural network model to generate a coding vector as the paragraph content vector, wherein the coding model based on the temporal neural network model includes: a long short-term memory model (Long-Short Term Memory, LSTM), at least one fully connected layer (Fully Connected Layer) and at least one down-sampling convolution layer. The vector dimension corresponding to the coding vector is smaller than the vector dimension of the combined vector corresponding to each sentence vector. Each down-sampling convolution layer in at least one down-sampling convolution layer is serially connected. The vector dimension of the output vector of each down-sampling convolution layer in at least one down-sampling convolution layer is sequentially reduced.
[0050] Considering the problems of the above conventional solutions, facing the above technical problems: the obtained paragraph content vector is not accurate enough and cannot fully reflect the content semantic information of the policy paragraph content. Combined with the advantages / technical status, we can decide to adopt the following solution.
[0051] In some optional implementations of some embodiments, the long short-term memory model includes: a memory cell unit sequence. There is a one-to-one correspondence between the memory cell units in the memory cell unit sequence and the sentence vectors in the sentence vector sequence. There is a corresponding fully connected layer for at least one target memory cell unit group in the memory cell unit sequence. The memory cell unit may be a cell unit in the long short-term memory model. The target memory cell unit group may be a unit group consisting of two adjacent memory cell units. For details, see Figure 2 , showing the network structure diagram of the encoding model based on the temporal neural network model.
[0052] Optionally, the step of sequentially inputting each sentence vector in the sentence vector sequence into an encoding model based on a temporal neural network model to generate an encoding vector as the paragraph content vector comprises the following steps:
[0053] The first step is to input each sentence vector in the above sentence vector sequence into the above memory cell unit sequence in sequence to generate at least one unit output vector and a model output vector corresponding to the above at least one target memory cell unit group. The unit output vector is a vector obtained by subtracting the output vectors corresponding to two target memory cell units in the target memory cell unit group. There is a one-to-one correspondence between each target memory cell unit group in at least one target memory cell unit group and each unit output vector in at least one unit output vector. The unit output vector can be the output vector difference between the target memory cell unit in the subsequent position and the target memory cell unit in the front position in the target memory cell unit group. The model output vector can be the vector output by the memory cell unit at the last position in the memory cell unit sequence.
[0054] In the second step, the model output vector is input into the at least one downsampling convolutional layer to generate an initial paragraph content vector.
[0055] In a third step, each unit output vector in the at least one unit output vector is input to a corresponding fully connected layer in the at least one fully connected layer to generate a fully connected vector, thereby obtaining at least one fully connected vector.
[0056] The fourth step is to determine the similarity set between the at least one fully connected vector and the corresponding at least one sentence vector. Among them, the similarity in the similarity set has a one-to-one correspondence with the fully connected vector in the at least one fully connected vector. The similarity can be a numerical value between 0 and 1. The larger the numerical value, the more similar the content semantics corresponding to the fully connected vector is to the content semantics corresponding to the sentence vector. The fully connected vector in at least one fully connected vector has a one-to-one correspondence with the target memory cell unit group in at least one target memory cell unit group. The sentence vector in at least one sentence vector is the vector output by the target memory cell unit in the corresponding target memory cell unit group at a subsequent position.
[0057] As an example, for each fully-connected vector in the at least one fully-connected vector, the cosine similarity between the fully-connected vector and the corresponding sentence vector in the at least one sentence vector is determined as the similarity.
[0058] In step 5, in response to determining that each similarity in the similarity set is greater than a target value, the initial paragraph content vector is determined as the paragraph content vector, wherein the target value may be a preset value, for example, the target value may be 0.5.
[0059] The content in "some optional implementations of some embodiments" as an inventive point of the present disclosure solves the technical problem mentioned in the background technology that "the obtained paragraph content vector is not accurate enough and cannot fully reflect the content semantic information of the policy paragraph content." Based on this, the present disclosure can ensure the accuracy of the output paragraph content vector on the premise of ensuring the similarity between the output of the memory cell unit and the corresponding at least one sentence vector by utilizing an encoding model based on a temporal neural network model, at least one downsampling convolution layer and at least one fully connected layer.
[0060] In some optional implementations of some embodiments, after step 1033, the steps further include:
[0061] In the first step, the paragraph content vector is input into a pre-trained first paragraph content semantic information generation model to generate first paragraph content semantic information. The first paragraph content semantic information generation model may be a neural network model for generating first paragraph content semantic information. The first paragraph content semantic information may be the predicted semantic subject content of the paragraph corresponding to the paragraph content vector. For example, the first paragraph content semantic information generation model may be a large language generation model (LLMs, Large Language Models).
[0062] In the second step, each sentence vector in the above sentence vector sequence is input into a pre-trained second paragraph content semantic information generation model to generate second paragraph content semantic information. The second paragraph content semantic information may be the total main semantic content predicted by the sentence vector sequence corresponding to each sentence. The second paragraph content semantic information may be the paragraph main content of the paragraph corresponding to the predicted sentence vector sequence. For example, the second paragraph content semantic information generation model may also be a large language generation model.
[0063] The third step is to determine the semantic similarity between the semantic information of the content of the first paragraph and the semantic information of the content of the second paragraph. The semantic similarity can represent the degree of content similarity between the semantic information of the content of the first paragraph and the semantic information of the content of the second paragraph. The semantic similarity can be a value between 0 and 1. The higher the value, the more similar the content between the semantic information of the content of the first paragraph and the semantic information of the content of the second paragraph is.
[0064] As an example, first, the execution subject may generate a first paragraph content semantic vector corresponding to the first paragraph content semantic information and a second paragraph content semantic vector corresponding to the second paragraph content semantic information. Then, the cosine similarity between the first paragraph content semantic vector and the second paragraph content semantic vector is determined as the semantic similarity.
[0065] Step 4: In response to determining that the semantic similarity is greater than or equal to a target value, information indicating that the paragraph content vector is verified to be correct is generated. The target value may be a preset value. For example, the target value is 0.7.
[0066] Step 1034, input the above-mentioned sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the above-mentioned policy statement sequence.
[0067] In some embodiments, the execution entity may input the sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the policy statement sequence. The first compliance review annotation information includes: compliance review details information and risk information. The first compliance review annotation information generation model may be a neural network model that generates compliance review annotation information at the sentence granularity. In practice, the first compliance review annotation information generation model may be a time series neural network model (e.g., a recurrent neural network model). The compliance review details information may be the contents of each detail in the "Detailed Rules for the Implementation of Compliance Review". For example, the "Detailed Rules for the Implementation of Compliance Review" includes 54 detailed categories. The risk information may represent whether the policy statement corresponding to the sentence vector has a compliance review anomaly. For example, the risk information may be "0" or "1". The value "1" may represent that the policy statement corresponding to the sentence vector has a compliance review anomaly. The value "0" may represent that the policy statement corresponding to the sentence vector does not have a compliance review anomaly. There is a one-to-one correspondence between the policy statements in the policy statement sequence and the first compliance review annotation information in the first compliance review annotation information sequence.
[0068] In some optional implementations of some embodiments, the first compliance review annotation information generation model is generated by the following steps:
[0069] The first step is to obtain the first policy document set.
[0070] In the second step, for each first policy file in the first policy file set, the following second generation step is performed:
[0071] Sub-step 1: Generate a first policy statement sequence set corresponding to the first policy file, wherein each first policy statement sequence in the first policy statement sequence set corresponds to a certain content in the first policy file.
[0072] Sub-step 2: Generate a first sentence vector sequence set corresponding to the first policy sentence sequence set. There is a one-to-one correspondence between the first policy sentence in the first policy sentence sequence set and the first sentence vector in the first sentence vector sequence set. The first sentence vector can represent the sentence content semantics corresponding to the first policy sentence.
[0073] The third step is to obtain the first compliance review annotation label corresponding to each first policy statement in the obtained first policy statement set, and obtain the first compliance review annotation label set. The first policy statement set is each first policy statement corresponding to the first policy document set. The first compliance review annotation label can be a real annotation label after the compliance review of the first policy statement.
[0074] The fourth step is to generate a first training data set based on the first compliance review annotation label set and the first policy statement set.
[0075] As an example, the above-mentioned execution entity can combine each first compliance review annotation label in the first compliance review annotation label set with the corresponding first policy statement in the above-mentioned first policy statement set to generate first training data and obtain a first training data set.
[0076] The fifth step is to perform model training on the initial first compliance review annotation information generation model based on the above-mentioned first training data set to generate the above-mentioned first compliance review annotation information generation model.
[0077] As an example, the above-mentioned execution entity can use the gradient descent method to perform model training on the initial first compliance review annotation information generation model based on the above-mentioned first training data set to generate the above-mentioned first compliance review annotation information generation model.
[0078] Step 1035, input the above paragraph content vector into a pre-trained second compliance review annotation information generation model to generate the second compliance review annotation information corresponding to the above policy paragraph content.
[0079] In some embodiments, the above-mentioned execution entity may input the above-mentioned paragraph content vector into a pre-trained second compliance review annotation information generation model to generate the second compliance review annotation information corresponding to the above-mentioned policy paragraph content. Among them, the second compliance review annotation information includes: compliance review details information and risk information. The second compliance review annotation information generation model may be a neural network model that generates compliance review annotation information at the paragraph granularity. In practice, the second compliance review annotation information generation model may be a temporal neural network model (for example, a recurrent neural network model). The second compliance review annotation information may characterize whether there is a compliance review anomaly in the paragraph corresponding to the policy paragraph content.
[0080] It should be noted that the training of the first compliance review annotation information generation model and the second compliance review annotation information generation model can be based on the gradient descent method and trained through conventional training methods.
[0081] In some optional implementations of some embodiments, the first compliance review annotation information generation model and the second compliance review annotation information generation model are generated by the following steps:
[0082] The first step is to obtain the second policy file set.
[0083] In the second step, for each second policy file in the second policy file set, the following third generation step is performed:
[0084] Sub-step 1: Generate a second policy statement sequence set corresponding to the second policy file, wherein each second policy statement sequence in the second policy statement sequence set corresponds to a certain content in the second policy file.
[0085] Sub-step 2: Generate a second sentence vector sequence set corresponding to the second policy sentence sequence set. There is a one-to-one correspondence between the second policy sentence in the second policy sentence sequence set and the second sentence vector in the second sentence vector sequence set. The second sentence vector can represent the sentence content semantics corresponding to the second policy sentence.
[0086] Sub-step 3: for each second sentence vector sequence in the second sentence vector sequence set, generate a paragraph content vector corresponding to the second sentence vector sequence as the target paragraph content vector, and obtain a target paragraph content vector sequence. The specific implementation method will not be repeated here.
[0087] The third step is to obtain the second compliance review annotation label corresponding to each second policy statement in the obtained second policy statement set to obtain the second compliance review annotation label set. The second policy statement set is each second policy statement corresponding to the second policy document set. The second compliance review annotation label can be a real annotation label after the compliance review of the second policy statement.
[0088] The fourth step is to obtain the third compliance review annotation label corresponding to each target paragraph content vector in the obtained target paragraph content vector sequence set to obtain a third compliance review annotation label set. The third compliance review annotation label can be a real annotation label after compliance review of the corresponding paragraph of the target paragraph content vector.
[0089] The fifth step is to generate a second training data set based on the second compliance review annotation label set and the second policy statement set.
[0090] As an example, the above-mentioned execution entity can combine each second compliance review annotation label in the second compliance review annotation label set with the corresponding second policy statement in the above-mentioned second policy statement set to generate second training data and obtain a second training data set.
[0091] Step 6: Generate a third training data set based on the target paragraph content vector sequence set and the paragraph content set, wherein the paragraph content set is the content of each paragraph corresponding to the second policy document set.
[0092] As an example, the above-mentioned execution entity may combine each third compliance review annotation tag in the third compliance review annotation tag set with the corresponding paragraph content in the above-mentioned paragraph content set to generate third training data and obtain a third training data set.
[0093] In the seventh step, based on the above-mentioned second training data set, the initial first compliance review annotation information generation model is trained to generate a first compliance review annotation information generation model, and based on the above-mentioned third training data set, the initial second compliance review annotation information generation model is trained to generate a second compliance review annotation information generation model.
[0094] As an example, the execution entity may perform model training on the initial second compliance review annotation information generation model using the gradient descent method based on the second training data set to generate the second compliance review annotation information generation model. The execution entity may perform model training on the initial third compliance review annotation information generation model using the gradient descent method based on the third training data set to generate the third compliance review annotation information generation model.
[0095] Step 104: Generate comprehensive compliance review annotation information for the target policy document based on the obtained first compliance review annotation information sequence set and the second compliance review annotation information set.
[0096] In some embodiments, the execution entity may generate compliance review information for the target policy file based on the first compliance review annotation information sequence set and the second compliance review annotation information set. The comprehensive compliance review annotation information may represent the contents of each file in the target policy file that is annotated with compliance review risks.
[0097] As an example, the above-mentioned execution entity can directly filter out at least one policy paragraph content and at least one policy statement with corresponding risk information of 1 from the first compliance review annotation information sequence set and the second compliance review annotation information set as comprehensive compliance review annotation information.
[0098] In some optional implementations of some embodiments, generating the compliance review comprehensive annotation information for the target policy document according to the obtained first compliance review annotation information sequence set and the second compliance review annotation information set may include the following steps:
[0099] The first step is to set the content identifier corresponding to each policy paragraph content in the policy paragraph content set, and determine the sentence identifier corresponding to each policy sentence in the policy sentence sequence set, to obtain a content identifier set and a sentence identifier sequence set.
[0100] The second step is to generate a compliance review anomaly table based on the content identifier set, the sentence identifier sequence set, the first compliance review annotation information sequence set, and the second compliance review annotation information set as the above-mentioned compliance review comprehensive annotation information. The content identifier or sentence identifier in the above-mentioned compliance review anomaly table that represents the corresponding compliance review anomaly is annotated with a target color. For example, the target color may be red.
[0101] As an example, the execution entity may fill the content identification set, the sentence identification sequence set, the first compliance review annotation information sequence set, and the second compliance review annotation information set into an empty table to generate an initial compliance review exception table. Then, the content identification or sentence identification in the initial compliance review exception table that represents the corresponding compliance review exception is annotated with a target color to generate a compliance review exception table as the compliance review information.
[0102] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the compliance review information generation method of some embodiments of the present disclosure, the target policy document can be accurately and efficiently subject to compliance review. Specifically, the reason for the low efficiency of the relevant compliance review is that the review takes a long time and is inefficient through the manual sentence-by-sentence comparison. Based on this, the compliance review information generation method of some embodiments of the present disclosure first obtains the target policy document to obtain the object file for subsequent compliance review. Then, the above-mentioned target policy document is segmented into paragraphs to generate a policy paragraph content set, so as to facilitate the subsequent compliance review of the content of each paragraph. Then, for each policy paragraph content in the above-mentioned policy paragraph content set, the following first generation step is performed: the first step is to perform sentence processing on the above-mentioned policy paragraph content to generate a policy statement sequence, so as to facilitate the subsequent compliance review of each statement. The second step is to generate a sentence vector corresponding to each policy statement in the above-mentioned policy statement sequence, and obtain a sentence vector sequence, which is not only generated in a form that is convenient for subsequent input to the model, but also used for the subsequent generation of accurate paragraph content vectors. In the third step, according to the above sentence vector sequence, the paragraph content vector corresponding to the above policy paragraph content can be accurately generated, so as to facilitate the subsequent compliance review of the paragraph content vector using the model. In the fourth step, the above sentence vector sequence is input into the pre-trained first compliance review annotation information generation model to generate the first compliance review annotation information sequence corresponding to the above policy sentence sequence. Among them, the first compliance review annotation information includes: compliance review details information and risk information. Here, through the pre-trained first compliance review annotation information generation model, not only can the compliance review of the target policy document at the sentence granularity be accurately performed, but also the review speed can be improved. In the fifth step, the above paragraph content vector is input into the pre-trained second compliance review annotation information generation model to accurately generate the second compliance review annotation information corresponding to the above policy paragraph content. Here, through the pre-trained second compliance review annotation information generation model, not only can the compliance review of the target policy document at the paragraph granularity be accurately performed, but also the review speed can be improved. Next, according to the obtained first compliance review annotation information sequence set and second compliance review annotation information set, the compliance review comprehensive annotation information for the above target policy document can be accurately generated. Finally, the above-mentioned compliance review comprehensive annotation information is sent to the target review terminal to display the comprehensive compliance review status corresponding to the target policy document. In summary, at the sentence granularity and paragraph granularity, the first compliance review annotation information generation model and the second compliance review annotation information generation model can accurately and efficiently implement the comprehensive compliance review of the target policy document.
[0103] Further references Figure 2As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a compliance review information generating device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the compliance review information generating device can be specifically applied to various electronic devices.
[0104] like Figure 2 As shown, a compliance review information generating device 200 includes: an acquiring unit 201, a splitting unit 202, an executing unit 203, a generating unit 204 and a sending unit 205. Among them, the acquisition unit 201 is configured to acquire the target policy file; the splitting unit 202 is configured to perform paragraph splitting on the above-mentioned target policy file to generate a policy paragraph content set; the execution unit 203 is configured to perform the following first generation step for each policy paragraph content in the above-mentioned policy paragraph content set: sentence processing on the above-mentioned policy paragraph content to generate a policy statement sequence; generate a sentence vector corresponding to each policy statement in the above-mentioned policy statement sequence to obtain a sentence vector sequence; generate a paragraph content vector corresponding to the above-mentioned policy paragraph content according to the above-mentioned sentence vector sequence; input the above-mentioned sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the above-mentioned policy statement sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; input the above-mentioned paragraph content vector into a pre-trained second compliance review annotation information generation model to generate a second compliance review annotation information corresponding to the above-mentioned policy paragraph content; the generation unit 204 is configured to generate compliance review comprehensive annotation information for the above-mentioned target policy file according to the obtained first compliance review annotation information sequence set and the second compliance review annotation information set. The sending unit 205 is configured to send the above-mentioned compliance review comprehensive annotation information to the target review terminal.
[0105] It is understandable that the units recorded in the compliance review information generating device 200 and the reference Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features and beneficial effects described above for the method are also applicable to the compliance review information generating device 200 and the units contained therein, and will not be repeated here.
[0106] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.
[0107] like Figure 3As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.
[0108] Typically, the following devices may be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.
[0109] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.
[0110] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
[0111] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
[0112] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a target policy file; splits the target policy file into paragraphs to generate a policy paragraph content set; for each policy paragraph content in the policy paragraph content set, executes the following first generation step: performs sentence processing on the policy paragraph content to generate a policy statement sequence; generates a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence; generates a paragraph content vector corresponding to the policy paragraph content based on the sentence vector sequence; inputs the sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the policy statement sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; inputs the paragraph content vector into a pre-trained second compliance review annotation information generation model to generate a second compliance review annotation information corresponding to the policy paragraph content; generates compliance review comprehensive annotation information for the target policy file based on the obtained first compliance review annotation information sequence set and second compliance review annotation information set.
[0113] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0114] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0115] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The described units may also be set in a processor, for example, it may be described as: a processor includes an acquisition unit, a splitting unit, an execution unit, a generation unit, and a sending unit. The names of these units do not constitute a limitation on the units themselves in some cases, for example, the acquisition unit may also be described as a "unit for acquiring a target policy file".
[0116] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.
[0117] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.
Claims
1. A method for generating compliance review information, comprising: Obtain target policy documents; Splitting the target policy document into paragraphs to generate a policy paragraph content set; For each policy paragraph content in the policy paragraph content set, the following first generation step is performed: Sentence processing is performed on the content of the policy paragraph to generate a policy statement sequence; Generate a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence; Generating a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence; Inputting the sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the policy sentence sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; Input the paragraph content vector into a pre-trained second compliance review annotation information generation model to generate second compliance review annotation information corresponding to the policy paragraph content; generate comprehensive compliance review annotation information for the target policy document based on the obtained first compliance review annotation information sequence set and the second compliance review annotation information set; The compliance review comprehensive annotation information is sent to the target review terminal.
2. The method according to claim 1, wherein: Generating a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence includes: Input each sentence vector in the sentence vector sequence into an encoding model based on a temporal neural network model in sequence to generate an encoding vector as the paragraph content vector, wherein the encoding model based on the temporal neural network model includes: a long short-term memory model, at least one fully connected layer and at least one downsampling convolutional layer.
3. The method according to claim 1, wherein: After generating the paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence, the method further includes: Inputting the paragraph content vector into a pre-trained first paragraph content semantic information generation model to generate first paragraph content semantic information; Inputting each sentence vector in the sentence vector sequence into a pre-trained second paragraph content semantic information generation model to generate second paragraph content semantic information; Determining the semantic similarity between the first paragraph content semantic information and the second paragraph content semantic information; In response to determining that the semantic similarity is greater than or equal to the target value, information indicating that the paragraph content vector is verified to be correct is generated.
4. The method according to claim 1, wherein: The step of generating the compliance review comprehensive annotation information for the target policy document according to the obtained first compliance review annotation information sequence set and the second compliance review annotation information set includes: Setting a content identifier corresponding to each policy paragraph content in the policy paragraph content set, and determining a sentence identifier corresponding to each policy sentence in the policy sentence sequence set, to obtain a content identifier set and a sentence identifier sequence set; A compliance review anomaly table is generated based on the content identification set, the statement identification sequence set, the first compliance review annotation information sequence set and the second compliance review annotation information set as the compliance review comprehensive annotation information, wherein the content identification or statement identification in the compliance review anomaly table that represents the corresponding compliance review risk is marked with a target color.
5. The method according to claim 1, wherein: The first compliance review annotation information generation model is generated by the following steps: Get the first policy document set; For each first policy file in the first policy file set, the following second generation step is performed: generating a first policy statement sequence set corresponding to the first policy file; Generate a first statement vector sequence set corresponding to the first policy statement sequence set; obtain a first compliance review annotation label corresponding to each first policy statement in the obtained first policy statement set to obtain a first compliance review annotation label set, wherein the first policy statement set is each first policy statement corresponding to the first policy document set; Generate a first training data set based on the first compliance review annotation label set and the first policy statement set; Based on the first training data set, model training is performed on the initial first compliance review annotation information generation model to generate the first compliance review annotation information generation model.
6. The method according to claim 1, wherein: The first compliance review annotation information generation model and the second compliance review annotation information generation model are generated by the following steps: Get the second policy document set; For each second policy file in the second policy file set, the following third generation step is performed: generating a second policy statement sequence set corresponding to the second policy file; generating a second statement vector sequence set corresponding to the second policy statement sequence set; For each second sentence vector sequence in the second sentence vector sequence set, generating a paragraph content vector corresponding to the second sentence vector sequence as a target paragraph content vector, and obtaining a target paragraph content vector sequence; Obtaining a second compliance review annotation label corresponding to each second policy statement in the obtained second policy statement set, to obtain a second compliance review annotation label set, wherein the second policy statement set is each second policy statement corresponding to the second policy document set; Obtain a third compliance review annotation label corresponding to each target paragraph content vector in the obtained target paragraph content vector sequence set to obtain a third compliance review annotation label set; generating a second training data set according to the second compliance review annotation label set and the second policy statement set; Generate a third training data set according to the target paragraph content vector sequence set and the paragraph content set, wherein the paragraph content set is each paragraph content corresponding to the second policy document set; According to the second training data set, the initial first compliance review annotation information generation model is trained to generate a first compliance review annotation information generation model, and according to the third training data set, the initial second compliance review annotation information generation model is trained to generate a second compliance review annotation information generation model.
7. A compliance review information generating device, comprising: an acquisition unit, configured to acquire a target policy file; a splitting unit, configured to split the target policy file into paragraphs to generate a policy paragraph content set; The execution unit is configured to execute the following first generation step for each policy paragraph content in the policy paragraph content set: sentence processing is performed on the policy paragraph content to generate a policy statement sequence; Generate a sentence vector corresponding to each policy statement in the policy statement sequence to obtain a sentence vector sequence; generate a paragraph content vector corresponding to the policy paragraph content according to the sentence vector sequence; input the sentence vector sequence into a pre-trained first compliance review annotation information generation model to generate a first compliance review annotation information sequence corresponding to the policy statement sequence, wherein the first compliance review annotation information includes: compliance review details information and risk information; input the paragraph content vector into a pre-trained second compliance review annotation information generation model to generate second compliance review annotation information corresponding to the policy paragraph content; A generating unit is configured to generate the compliance review comprehensive annotation information for the target policy document according to the obtained first compliance review annotation information sequence set and the second compliance review annotation information set; The sending unit is configured to send the compliance review comprehensive annotation information to a target review terminal.
8. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.
9. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.